Loading...
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
51struct 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
59struct 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
73static struct io_thread_req * (*irq_req_buffer)[];
74static struct io_thread_req *irq_remainder;
75static int irq_remainder_size;
76
77static struct io_thread_req * (*io_req_buffer)[];
78static struct io_thread_req *io_remainder;
79static int io_remainder_size;
80
81
82
83static 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
94static 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
108static DEFINE_MUTEX(ubd_lock);
109static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
110
111static int ubd_open(struct block_device *bdev, fmode_t mode);
112static void ubd_release(struct gendisk *disk, fmode_t mode);
113static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
114 unsigned int cmd, unsigned long arg);
115static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
116
117#define MAX_DEV (16)
118
119static const struct block_device_operations ubd_blops = {
120 .owner = THIS_MODULE,
121 .open = ubd_open,
122 .release = ubd_release,
123 .ioctl = ubd_ioctl,
124 .compat_ioctl = blkdev_compat_ptr_ioctl,
125 .getgeo = ubd_getgeo,
126};
127
128/* Protected by ubd_lock */
129static struct gendisk *ubd_gendisk[MAX_DEV];
130
131#ifdef CONFIG_BLK_DEV_UBD_SYNC
132#define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
133 .cl = 1 })
134#else
135#define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
136 .cl = 1 })
137#endif
138static struct openflags global_openflags = OPEN_FLAGS;
139
140struct cow {
141 /* backing file name */
142 char *file;
143 /* backing file fd */
144 int fd;
145 unsigned long *bitmap;
146 unsigned long bitmap_len;
147 int bitmap_offset;
148 int data_offset;
149};
150
151#define MAX_SG 64
152
153struct ubd {
154 /* name (and fd, below) of the file opened for writing, either the
155 * backing or the cow file. */
156 char *file;
157 char *serial;
158 int count;
159 int fd;
160 __u64 size;
161 struct openflags boot_openflags;
162 struct openflags openflags;
163 unsigned shared:1;
164 unsigned no_cow:1;
165 unsigned no_trim:1;
166 struct cow cow;
167 struct platform_device pdev;
168 struct request_queue *queue;
169 struct blk_mq_tag_set tag_set;
170 spinlock_t lock;
171};
172
173#define DEFAULT_COW { \
174 .file = NULL, \
175 .fd = -1, \
176 .bitmap = NULL, \
177 .bitmap_offset = 0, \
178 .data_offset = 0, \
179}
180
181#define DEFAULT_UBD { \
182 .file = NULL, \
183 .serial = NULL, \
184 .count = 0, \
185 .fd = -1, \
186 .size = -1, \
187 .boot_openflags = OPEN_FLAGS, \
188 .openflags = OPEN_FLAGS, \
189 .no_cow = 0, \
190 .no_trim = 0, \
191 .shared = 0, \
192 .cow = DEFAULT_COW, \
193 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
194}
195
196/* Protected by ubd_lock */
197static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
198
199static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
200 const struct blk_mq_queue_data *bd);
201
202static int fake_ide_setup(char *str)
203{
204 pr_warn("The fake_ide option has been removed\n");
205 return 1;
206}
207__setup("fake_ide", fake_ide_setup);
208
209__uml_help(fake_ide_setup,
210"fake_ide\n"
211" Obsolete stub.\n\n"
212);
213
214static int parse_unit(char **ptr)
215{
216 char *str = *ptr, *end;
217 int n = -1;
218
219 if(isdigit(*str)) {
220 n = simple_strtoul(str, &end, 0);
221 if(end == str)
222 return -1;
223 *ptr = end;
224 }
225 else if (('a' <= *str) && (*str <= 'z')) {
226 n = *str - 'a';
227 str++;
228 *ptr = str;
229 }
230 return n;
231}
232
233/* If *index_out == -1 at exit, the passed option was a general one;
234 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
235 * should not be freed on exit.
236 */
237static int ubd_setup_common(char *str, int *index_out, char **error_out)
238{
239 struct ubd *ubd_dev;
240 struct openflags flags = global_openflags;
241 char *file, *backing_file, *serial;
242 int n, err = 0, i;
243
244 if(index_out) *index_out = -1;
245 n = *str;
246 if(n == '='){
247 str++;
248 if(!strcmp(str, "sync")){
249 global_openflags = of_sync(global_openflags);
250 return err;
251 }
252
253 pr_warn("fake major not supported any more\n");
254 return 0;
255 }
256
257 n = parse_unit(&str);
258 if(n < 0){
259 *error_out = "Couldn't parse device number";
260 return -EINVAL;
261 }
262 if(n >= MAX_DEV){
263 *error_out = "Device number out of range";
264 return 1;
265 }
266
267 err = -EBUSY;
268 mutex_lock(&ubd_lock);
269
270 ubd_dev = &ubd_devs[n];
271 if(ubd_dev->file != NULL){
272 *error_out = "Device is already configured";
273 goto out;
274 }
275
276 if (index_out)
277 *index_out = n;
278
279 err = -EINVAL;
280 for (i = 0; i < sizeof("rscdt="); i++) {
281 switch (*str) {
282 case 'r':
283 flags.w = 0;
284 break;
285 case 's':
286 flags.s = 1;
287 break;
288 case 'd':
289 ubd_dev->no_cow = 1;
290 break;
291 case 'c':
292 ubd_dev->shared = 1;
293 break;
294 case 't':
295 ubd_dev->no_trim = 1;
296 break;
297 case '=':
298 str++;
299 goto break_loop;
300 default:
301 *error_out = "Expected '=' or flag letter "
302 "(r, s, c, t or d)";
303 goto out;
304 }
305 str++;
306 }
307
308 if (*str == '=')
309 *error_out = "Too many flags specified";
310 else
311 *error_out = "Missing '='";
312 goto out;
313
314break_loop:
315 file = strsep(&str, ",:");
316 if (*file == '\0')
317 file = NULL;
318
319 backing_file = strsep(&str, ",:");
320 if (backing_file && *backing_file == '\0')
321 backing_file = NULL;
322
323 serial = strsep(&str, ",:");
324 if (serial && *serial == '\0')
325 serial = NULL;
326
327 if (backing_file && ubd_dev->no_cow) {
328 *error_out = "Can't specify both 'd' and a cow file";
329 goto out;
330 }
331
332 err = 0;
333 ubd_dev->file = file;
334 ubd_dev->cow.file = backing_file;
335 ubd_dev->serial = serial;
336 ubd_dev->boot_openflags = flags;
337out:
338 mutex_unlock(&ubd_lock);
339 return err;
340}
341
342static int ubd_setup(char *str)
343{
344 char *error;
345 int err;
346
347 err = ubd_setup_common(str, NULL, &error);
348 if(err)
349 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
350 "%s\n", str, error);
351 return 1;
352}
353
354__setup("ubd", ubd_setup);
355__uml_help(ubd_setup,
356"ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n"
357" This is used to associate a device with a file in the underlying\n"
358" filesystem. When specifying two filenames, the first one is the\n"
359" COW name and the second is the backing file name. As separator you can\n"
360" use either a ':' or a ',': the first one allows writing things like;\n"
361" ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
362" while with a ',' the shell would not expand the 2nd '~'.\n"
363" When using only one filename, UML will detect whether to treat it like\n"
364" a COW file or a backing file. To override this detection, add the 'd'\n"
365" flag:\n"
366" ubd0d=BackingFile\n"
367" Usually, there is a filesystem in the file, but \n"
368" that's not required. Swap devices containing swap files can be\n"
369" specified like this. Also, a file which doesn't contain a\n"
370" filesystem can have its contents read in the virtual \n"
371" machine by running 'dd' on the device. <n> must be in the range\n"
372" 0 to 7. Appending an 'r' to the number will cause that device\n"
373" to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
374" an 's' will cause data to be written to disk on the host immediately.\n"
375" 'c' will cause the device to be treated as being shared between multiple\n"
376" UMLs and file locking will be turned off - this is appropriate for a\n"
377" cluster filesystem and inappropriate at almost all other times.\n\n"
378" 't' will disable trim/discard support on the device (enabled by default).\n\n"
379" An optional device serial number can be exposed using the serial parameter\n"
380" on the cmdline which is exposed as a sysfs entry. This is particularly\n"
381" useful when a unique number should be given to the device. Note when\n"
382" specifying a label, the filename2 must be also presented. It can be\n"
383" an empty string, in which case the backing file is not used:\n"
384" ubd0=File,,Serial\n"
385);
386
387static int udb_setup(char *str)
388{
389 printk("udb%s specified on command line is almost certainly a ubd -> "
390 "udb TYPO\n", str);
391 return 1;
392}
393
394__setup("udb", udb_setup);
395__uml_help(udb_setup,
396"udb\n"
397" This option is here solely to catch ubd -> udb typos, which can be\n"
398" to impossible to catch visually unless you specifically look for\n"
399" them. The only result of any option starting with 'udb' is an error\n"
400" in the boot output.\n\n"
401);
402
403/* Only changed by ubd_init, which is an initcall. */
404static int thread_fd = -1;
405
406/* Function to read several request pointers at a time
407* handling fractional reads if (and as) needed
408*/
409
410static int bulk_req_safe_read(
411 int fd,
412 struct io_thread_req * (*request_buffer)[],
413 struct io_thread_req **remainder,
414 int *remainder_size,
415 int max_recs
416 )
417{
418 int n = 0;
419 int res = 0;
420
421 if (*remainder_size > 0) {
422 memmove(
423 (char *) request_buffer,
424 (char *) remainder, *remainder_size
425 );
426 n = *remainder_size;
427 }
428
429 res = os_read_file(
430 fd,
431 ((char *) request_buffer) + *remainder_size,
432 sizeof(struct io_thread_req *)*max_recs
433 - *remainder_size
434 );
435 if (res > 0) {
436 n += res;
437 if ((n % sizeof(struct io_thread_req *)) > 0) {
438 /*
439 * Read somehow returned not a multiple of dword
440 * theoretically possible, but never observed in the
441 * wild, so read routine must be able to handle it
442 */
443 *remainder_size = n % sizeof(struct io_thread_req *);
444 WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
445 memmove(
446 remainder,
447 ((char *) request_buffer) +
448 (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
449 *remainder_size
450 );
451 n = n - *remainder_size;
452 }
453 } else {
454 n = res;
455 }
456 return n;
457}
458
459/* Called without dev->lock held, and only in interrupt context. */
460static void ubd_handler(void)
461{
462 int n;
463 int count;
464
465 while(1){
466 n = bulk_req_safe_read(
467 thread_fd,
468 irq_req_buffer,
469 &irq_remainder,
470 &irq_remainder_size,
471 UBD_REQ_BUFFER_SIZE
472 );
473 if (n < 0) {
474 if(n == -EAGAIN)
475 break;
476 printk(KERN_ERR "spurious interrupt in ubd_handler, "
477 "err = %d\n", -n);
478 return;
479 }
480 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
481 struct io_thread_req *io_req = (*irq_req_buffer)[count];
482
483 if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
484 blk_queue_max_discard_sectors(io_req->req->q, 0);
485 blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
486 }
487 blk_mq_end_request(io_req->req, io_req->error);
488 kfree(io_req);
489 }
490 }
491}
492
493static irqreturn_t ubd_intr(int irq, void *dev)
494{
495 ubd_handler();
496 return IRQ_HANDLED;
497}
498
499/* Only changed by ubd_init, which is an initcall. */
500static int io_pid = -1;
501
502static void kill_io_thread(void)
503{
504 if(io_pid != -1)
505 os_kill_process(io_pid, 1);
506}
507
508__uml_exitcall(kill_io_thread);
509
510static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
511{
512 char *file;
513 int fd;
514 int err;
515
516 __u32 version;
517 __u32 align;
518 char *backing_file;
519 time64_t mtime;
520 unsigned long long size;
521 int sector_size;
522 int bitmap_offset;
523
524 if (ubd_dev->file && ubd_dev->cow.file) {
525 file = ubd_dev->cow.file;
526
527 goto out;
528 }
529
530 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
531 if (fd < 0)
532 return fd;
533
534 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
535 &mtime, &size, §or_size, &align, &bitmap_offset);
536 os_close_file(fd);
537
538 if(err == -EINVAL)
539 file = ubd_dev->file;
540 else
541 file = backing_file;
542
543out:
544 return os_file_size(file, size_out);
545}
546
547static int read_cow_bitmap(int fd, void *buf, int offset, int len)
548{
549 int err;
550
551 err = os_pread_file(fd, buf, len, offset);
552 if (err < 0)
553 return err;
554
555 return 0;
556}
557
558static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
559{
560 time64_t modtime;
561 unsigned long long actual;
562 int err;
563
564 err = os_file_modtime(file, &modtime);
565 if (err < 0) {
566 printk(KERN_ERR "Failed to get modification time of backing "
567 "file \"%s\", err = %d\n", file, -err);
568 return err;
569 }
570
571 err = os_file_size(file, &actual);
572 if (err < 0) {
573 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
574 "err = %d\n", file, -err);
575 return err;
576 }
577
578 if (actual != size) {
579 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
580 * the typecast.*/
581 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
582 "vs backing file\n", (unsigned long long) size, actual);
583 return -EINVAL;
584 }
585 if (modtime != mtime) {
586 printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
587 "backing file\n", mtime, modtime);
588 return -EINVAL;
589 }
590 return 0;
591}
592
593static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
594{
595 struct uml_stat buf1, buf2;
596 int err;
597
598 if (from_cmdline == NULL)
599 return 0;
600 if (!strcmp(from_cmdline, from_cow))
601 return 0;
602
603 err = os_stat_file(from_cmdline, &buf1);
604 if (err < 0) {
605 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
606 -err);
607 return 0;
608 }
609 err = os_stat_file(from_cow, &buf2);
610 if (err < 0) {
611 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
612 -err);
613 return 1;
614 }
615 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
616 return 0;
617
618 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
619 "\"%s\" specified in COW header of \"%s\"\n",
620 from_cmdline, from_cow, cow);
621 return 1;
622}
623
624static int open_ubd_file(char *file, struct openflags *openflags, int shared,
625 char **backing_file_out, int *bitmap_offset_out,
626 unsigned long *bitmap_len_out, int *data_offset_out,
627 int *create_cow_out)
628{
629 time64_t mtime;
630 unsigned long long size;
631 __u32 version, align;
632 char *backing_file;
633 int fd, err, sectorsize, asked_switch, mode = 0644;
634
635 fd = os_open_file(file, *openflags, mode);
636 if (fd < 0) {
637 if ((fd == -ENOENT) && (create_cow_out != NULL))
638 *create_cow_out = 1;
639 if (!openflags->w ||
640 ((fd != -EROFS) && (fd != -EACCES)))
641 return fd;
642 openflags->w = 0;
643 fd = os_open_file(file, *openflags, mode);
644 if (fd < 0)
645 return fd;
646 }
647
648 if (shared)
649 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
650 else {
651 err = os_lock_file(fd, openflags->w);
652 if (err < 0) {
653 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
654 file, -err);
655 goto out_close;
656 }
657 }
658
659 /* Successful return case! */
660 if (backing_file_out == NULL)
661 return fd;
662
663 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
664 &size, §orsize, &align, bitmap_offset_out);
665 if (err && (*backing_file_out != NULL)) {
666 printk(KERN_ERR "Failed to read COW header from COW file "
667 "\"%s\", errno = %d\n", file, -err);
668 goto out_close;
669 }
670 if (err)
671 return fd;
672
673 asked_switch = path_requires_switch(*backing_file_out, backing_file,
674 file);
675
676 /* Allow switching only if no mismatch. */
677 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
678 mtime)) {
679 printk(KERN_ERR "Switching backing file to '%s'\n",
680 *backing_file_out);
681 err = write_cow_header(file, fd, *backing_file_out,
682 sectorsize, align, &size);
683 if (err) {
684 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
685 goto out_close;
686 }
687 } else {
688 *backing_file_out = backing_file;
689 err = backing_file_mismatch(*backing_file_out, size, mtime);
690 if (err)
691 goto out_close;
692 }
693
694 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
695 bitmap_len_out, data_offset_out);
696
697 return fd;
698 out_close:
699 os_close_file(fd);
700 return err;
701}
702
703static int create_cow_file(char *cow_file, char *backing_file,
704 struct openflags flags,
705 int sectorsize, int alignment, int *bitmap_offset_out,
706 unsigned long *bitmap_len_out, int *data_offset_out)
707{
708 int err, fd;
709
710 flags.c = 1;
711 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
712 if (fd < 0) {
713 err = fd;
714 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
715 cow_file, -err);
716 goto out;
717 }
718
719 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
720 bitmap_offset_out, bitmap_len_out,
721 data_offset_out);
722 if (!err)
723 return fd;
724 os_close_file(fd);
725 out:
726 return err;
727}
728
729static void ubd_close_dev(struct ubd *ubd_dev)
730{
731 os_close_file(ubd_dev->fd);
732 if(ubd_dev->cow.file == NULL)
733 return;
734
735 os_close_file(ubd_dev->cow.fd);
736 vfree(ubd_dev->cow.bitmap);
737 ubd_dev->cow.bitmap = NULL;
738}
739
740static int ubd_open_dev(struct ubd *ubd_dev)
741{
742 struct openflags flags;
743 char **back_ptr;
744 int err, create_cow, *create_ptr;
745 int fd;
746
747 ubd_dev->openflags = ubd_dev->boot_openflags;
748 create_cow = 0;
749 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
750 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
751
752 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
753 back_ptr, &ubd_dev->cow.bitmap_offset,
754 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
755 create_ptr);
756
757 if((fd == -ENOENT) && create_cow){
758 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
759 ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
760 &ubd_dev->cow.bitmap_offset,
761 &ubd_dev->cow.bitmap_len,
762 &ubd_dev->cow.data_offset);
763 if(fd >= 0){
764 printk(KERN_INFO "Creating \"%s\" as COW file for "
765 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
766 }
767 }
768
769 if(fd < 0){
770 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
771 -fd);
772 return fd;
773 }
774 ubd_dev->fd = fd;
775
776 if(ubd_dev->cow.file != NULL){
777 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
778
779 err = -ENOMEM;
780 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
781 if(ubd_dev->cow.bitmap == NULL){
782 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
783 goto error;
784 }
785 flush_tlb_kernel_vm();
786
787 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
788 ubd_dev->cow.bitmap_offset,
789 ubd_dev->cow.bitmap_len);
790 if(err < 0)
791 goto error;
792
793 flags = ubd_dev->openflags;
794 flags.w = 0;
795 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
796 NULL, NULL, NULL, NULL);
797 if(err < 0) goto error;
798 ubd_dev->cow.fd = err;
799 }
800 if (ubd_dev->no_trim == 0) {
801 ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE;
802 blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
803 blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
804 }
805 blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue);
806 return 0;
807 error:
808 os_close_file(ubd_dev->fd);
809 return err;
810}
811
812static void ubd_device_release(struct device *dev)
813{
814 struct ubd *ubd_dev = dev_get_drvdata(dev);
815
816 blk_mq_free_tag_set(&ubd_dev->tag_set);
817 *ubd_dev = ((struct ubd) DEFAULT_UBD);
818}
819
820static ssize_t serial_show(struct device *dev,
821 struct device_attribute *attr, char *buf)
822{
823 struct gendisk *disk = dev_to_disk(dev);
824 struct ubd *ubd_dev = disk->private_data;
825
826 if (!ubd_dev)
827 return 0;
828
829 return sprintf(buf, "%s", ubd_dev->serial);
830}
831
832static DEVICE_ATTR_RO(serial);
833
834static struct attribute *ubd_attrs[] = {
835 &dev_attr_serial.attr,
836 NULL,
837};
838
839static umode_t ubd_attrs_are_visible(struct kobject *kobj,
840 struct attribute *a, int n)
841{
842 return a->mode;
843}
844
845static const struct attribute_group ubd_attr_group = {
846 .attrs = ubd_attrs,
847 .is_visible = ubd_attrs_are_visible,
848};
849
850static const struct attribute_group *ubd_attr_groups[] = {
851 &ubd_attr_group,
852 NULL,
853};
854
855static int ubd_disk_register(int major, u64 size, int unit,
856 struct gendisk *disk)
857{
858 disk->major = major;
859 disk->first_minor = unit << UBD_SHIFT;
860 disk->minors = 1 << UBD_SHIFT;
861 disk->fops = &ubd_blops;
862 set_capacity(disk, size / 512);
863 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
864
865 ubd_devs[unit].pdev.id = unit;
866 ubd_devs[unit].pdev.name = DRIVER_NAME;
867 ubd_devs[unit].pdev.dev.release = ubd_device_release;
868 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
869 platform_device_register(&ubd_devs[unit].pdev);
870
871 disk->private_data = &ubd_devs[unit];
872 disk->queue = ubd_devs[unit].queue;
873 return device_add_disk(&ubd_devs[unit].pdev.dev, disk, ubd_attr_groups);
874}
875
876#define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
877
878static const struct blk_mq_ops ubd_mq_ops = {
879 .queue_rq = ubd_queue_rq,
880};
881
882static int ubd_add(int n, char **error_out)
883{
884 struct ubd *ubd_dev = &ubd_devs[n];
885 struct gendisk *disk;
886 int err = 0;
887
888 if(ubd_dev->file == NULL)
889 goto out;
890
891 err = ubd_file_size(ubd_dev, &ubd_dev->size);
892 if(err < 0){
893 *error_out = "Couldn't determine size of device's file";
894 goto out;
895 }
896
897 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
898
899 ubd_dev->tag_set.ops = &ubd_mq_ops;
900 ubd_dev->tag_set.queue_depth = 64;
901 ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
902 ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
903 ubd_dev->tag_set.driver_data = ubd_dev;
904 ubd_dev->tag_set.nr_hw_queues = 1;
905
906 err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
907 if (err)
908 goto out;
909
910 disk = blk_mq_alloc_disk(&ubd_dev->tag_set, ubd_dev);
911 if (IS_ERR(disk)) {
912 err = PTR_ERR(disk);
913 goto out_cleanup_tags;
914 }
915 ubd_dev->queue = disk->queue;
916
917 blk_queue_write_cache(ubd_dev->queue, true, false);
918 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
919 blk_queue_segment_boundary(ubd_dev->queue, PAGE_SIZE - 1);
920 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, disk);
921 if (err)
922 goto out_cleanup_disk;
923
924 ubd_gendisk[n] = disk;
925 return 0;
926
927out_cleanup_disk:
928 put_disk(disk);
929out_cleanup_tags:
930 blk_mq_free_tag_set(&ubd_dev->tag_set);
931out:
932 return err;
933}
934
935static int ubd_config(char *str, char **error_out)
936{
937 int n, ret;
938
939 /* This string is possibly broken up and stored, so it's only
940 * freed if ubd_setup_common fails, or if only general options
941 * were set.
942 */
943 str = kstrdup(str, GFP_KERNEL);
944 if (str == NULL) {
945 *error_out = "Failed to allocate memory";
946 return -ENOMEM;
947 }
948
949 ret = ubd_setup_common(str, &n, error_out);
950 if (ret)
951 goto err_free;
952
953 if (n == -1) {
954 ret = 0;
955 goto err_free;
956 }
957
958 mutex_lock(&ubd_lock);
959 ret = ubd_add(n, error_out);
960 if (ret)
961 ubd_devs[n].file = NULL;
962 mutex_unlock(&ubd_lock);
963
964out:
965 return ret;
966
967err_free:
968 kfree(str);
969 goto out;
970}
971
972static int ubd_get_config(char *name, char *str, int size, char **error_out)
973{
974 struct ubd *ubd_dev;
975 int n, len = 0;
976
977 n = parse_unit(&name);
978 if((n >= MAX_DEV) || (n < 0)){
979 *error_out = "ubd_get_config : device number out of range";
980 return -1;
981 }
982
983 ubd_dev = &ubd_devs[n];
984 mutex_lock(&ubd_lock);
985
986 if(ubd_dev->file == NULL){
987 CONFIG_CHUNK(str, size, len, "", 1);
988 goto out;
989 }
990
991 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
992
993 if(ubd_dev->cow.file != NULL){
994 CONFIG_CHUNK(str, size, len, ",", 0);
995 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
996 }
997 else CONFIG_CHUNK(str, size, len, "", 1);
998
999 out:
1000 mutex_unlock(&ubd_lock);
1001 return len;
1002}
1003
1004static int ubd_id(char **str, int *start_out, int *end_out)
1005{
1006 int n;
1007
1008 n = parse_unit(str);
1009 *start_out = 0;
1010 *end_out = MAX_DEV - 1;
1011 return n;
1012}
1013
1014static int ubd_remove(int n, char **error_out)
1015{
1016 struct gendisk *disk = ubd_gendisk[n];
1017 struct ubd *ubd_dev;
1018 int err = -ENODEV;
1019
1020 mutex_lock(&ubd_lock);
1021
1022 ubd_dev = &ubd_devs[n];
1023
1024 if(ubd_dev->file == NULL)
1025 goto out;
1026
1027 /* you cannot remove a open disk */
1028 err = -EBUSY;
1029 if(ubd_dev->count > 0)
1030 goto out;
1031
1032 ubd_gendisk[n] = NULL;
1033 if(disk != NULL){
1034 del_gendisk(disk);
1035 put_disk(disk);
1036 }
1037
1038 err = 0;
1039 platform_device_unregister(&ubd_dev->pdev);
1040out:
1041 mutex_unlock(&ubd_lock);
1042 return err;
1043}
1044
1045/* All these are called by mconsole in process context and without
1046 * ubd-specific locks. The structure itself is const except for .list.
1047 */
1048static struct mc_device ubd_mc = {
1049 .list = LIST_HEAD_INIT(ubd_mc.list),
1050 .name = "ubd",
1051 .config = ubd_config,
1052 .get_config = ubd_get_config,
1053 .id = ubd_id,
1054 .remove = ubd_remove,
1055};
1056
1057static int __init ubd_mc_init(void)
1058{
1059 mconsole_register_dev(&ubd_mc);
1060 return 0;
1061}
1062
1063__initcall(ubd_mc_init);
1064
1065static int __init ubd0_init(void)
1066{
1067 struct ubd *ubd_dev = &ubd_devs[0];
1068
1069 mutex_lock(&ubd_lock);
1070 if(ubd_dev->file == NULL)
1071 ubd_dev->file = "root_fs";
1072 mutex_unlock(&ubd_lock);
1073
1074 return 0;
1075}
1076
1077__initcall(ubd0_init);
1078
1079/* Used in ubd_init, which is an initcall */
1080static struct platform_driver ubd_driver = {
1081 .driver = {
1082 .name = DRIVER_NAME,
1083 },
1084};
1085
1086static int __init ubd_init(void)
1087{
1088 char *error;
1089 int i, err;
1090
1091 if (register_blkdev(UBD_MAJOR, "ubd"))
1092 return -1;
1093
1094 irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1095 sizeof(struct io_thread_req *),
1096 GFP_KERNEL
1097 );
1098 irq_remainder = 0;
1099
1100 if (irq_req_buffer == NULL) {
1101 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1102 return -1;
1103 }
1104 io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1105 sizeof(struct io_thread_req *),
1106 GFP_KERNEL
1107 );
1108
1109 io_remainder = 0;
1110
1111 if (io_req_buffer == NULL) {
1112 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1113 return -1;
1114 }
1115 platform_driver_register(&ubd_driver);
1116 mutex_lock(&ubd_lock);
1117 for (i = 0; i < MAX_DEV; i++){
1118 err = ubd_add(i, &error);
1119 if(err)
1120 printk(KERN_ERR "Failed to initialize ubd device %d :"
1121 "%s\n", i, error);
1122 }
1123 mutex_unlock(&ubd_lock);
1124 return 0;
1125}
1126
1127late_initcall(ubd_init);
1128
1129static int __init ubd_driver_init(void){
1130 unsigned long stack;
1131 int err;
1132
1133 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1134 if(global_openflags.s){
1135 printk(KERN_INFO "ubd: Synchronous mode\n");
1136 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1137 * enough. So use anyway the io thread. */
1138 }
1139 stack = alloc_stack(0, 0);
1140 io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd);
1141 if(io_pid < 0){
1142 printk(KERN_ERR
1143 "ubd : Failed to start I/O thread (errno = %d) - "
1144 "falling back to synchronous I/O\n", -io_pid);
1145 io_pid = -1;
1146 return 0;
1147 }
1148 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1149 0, "ubd", ubd_devs);
1150 if(err < 0)
1151 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1152 return 0;
1153}
1154
1155device_initcall(ubd_driver_init);
1156
1157static int ubd_open(struct block_device *bdev, fmode_t mode)
1158{
1159 struct gendisk *disk = bdev->bd_disk;
1160 struct ubd *ubd_dev = disk->private_data;
1161 int err = 0;
1162
1163 mutex_lock(&ubd_mutex);
1164 if(ubd_dev->count == 0){
1165 err = ubd_open_dev(ubd_dev);
1166 if(err){
1167 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1168 disk->disk_name, ubd_dev->file, -err);
1169 goto out;
1170 }
1171 }
1172 ubd_dev->count++;
1173 set_disk_ro(disk, !ubd_dev->openflags.w);
1174
1175 /* This should no more be needed. And it didn't work anyway to exclude
1176 * read-write remounting of filesystems.*/
1177 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1178 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1179 err = -EROFS;
1180 }*/
1181out:
1182 mutex_unlock(&ubd_mutex);
1183 return err;
1184}
1185
1186static void ubd_release(struct gendisk *disk, fmode_t mode)
1187{
1188 struct ubd *ubd_dev = disk->private_data;
1189
1190 mutex_lock(&ubd_mutex);
1191 if(--ubd_dev->count == 0)
1192 ubd_close_dev(ubd_dev);
1193 mutex_unlock(&ubd_mutex);
1194}
1195
1196static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1197 __u64 *cow_offset, unsigned long *bitmap,
1198 __u64 bitmap_offset, unsigned long *bitmap_words,
1199 __u64 bitmap_len)
1200{
1201 __u64 sector = io_offset >> SECTOR_SHIFT;
1202 int i, update_bitmap = 0;
1203
1204 for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1205 if(cow_mask != NULL)
1206 ubd_set_bit(i, (unsigned char *) cow_mask);
1207 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1208 continue;
1209
1210 update_bitmap = 1;
1211 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1212 }
1213
1214 if(!update_bitmap)
1215 return;
1216
1217 *cow_offset = sector / (sizeof(unsigned long) * 8);
1218
1219 /* This takes care of the case where we're exactly at the end of the
1220 * device, and *cow_offset + 1 is off the end. So, just back it up
1221 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1222 * for the original diagnosis.
1223 */
1224 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1225 sizeof(unsigned long)) - 1))
1226 (*cow_offset)--;
1227
1228 bitmap_words[0] = bitmap[*cow_offset];
1229 bitmap_words[1] = bitmap[*cow_offset + 1];
1230
1231 *cow_offset *= sizeof(unsigned long);
1232 *cow_offset += bitmap_offset;
1233}
1234
1235static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1236 unsigned long offset, unsigned long *bitmap,
1237 __u64 bitmap_offset, __u64 bitmap_len)
1238{
1239 __u64 sector = offset >> SECTOR_SHIFT;
1240 int i;
1241
1242 if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1243 panic("Operation too long");
1244
1245 if (req_op(req->req) == REQ_OP_READ) {
1246 for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1247 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1248 ubd_set_bit(i, (unsigned char *)
1249 &segment->sector_mask);
1250 }
1251 } else {
1252 cowify_bitmap(offset, segment->length, &segment->sector_mask,
1253 &segment->cow_offset, bitmap, bitmap_offset,
1254 segment->bitmap_words, bitmap_len);
1255 }
1256}
1257
1258static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1259 struct request *req)
1260{
1261 struct bio_vec bvec;
1262 struct req_iterator iter;
1263 int i = 0;
1264 unsigned long byte_offset = io_req->offset;
1265 enum req_op op = req_op(req);
1266
1267 if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1268 io_req->io_desc[0].buffer = NULL;
1269 io_req->io_desc[0].length = blk_rq_bytes(req);
1270 } else {
1271 rq_for_each_segment(bvec, req, iter) {
1272 BUG_ON(i >= io_req->desc_cnt);
1273
1274 io_req->io_desc[i].buffer = bvec_virt(&bvec);
1275 io_req->io_desc[i].length = bvec.bv_len;
1276 i++;
1277 }
1278 }
1279
1280 if (dev->cow.file) {
1281 for (i = 0; i < io_req->desc_cnt; i++) {
1282 cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1283 dev->cow.bitmap, dev->cow.bitmap_offset,
1284 dev->cow.bitmap_len);
1285 byte_offset += io_req->io_desc[i].length;
1286 }
1287
1288 }
1289}
1290
1291static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1292 int desc_cnt)
1293{
1294 struct io_thread_req *io_req;
1295 int i;
1296
1297 io_req = kmalloc(sizeof(*io_req) +
1298 (desc_cnt * sizeof(struct io_desc)),
1299 GFP_ATOMIC);
1300 if (!io_req)
1301 return NULL;
1302
1303 io_req->req = req;
1304 if (dev->cow.file)
1305 io_req->fds[0] = dev->cow.fd;
1306 else
1307 io_req->fds[0] = dev->fd;
1308 io_req->error = 0;
1309 io_req->sectorsize = SECTOR_SIZE;
1310 io_req->fds[1] = dev->fd;
1311 io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1312 io_req->offsets[0] = 0;
1313 io_req->offsets[1] = dev->cow.data_offset;
1314
1315 for (i = 0 ; i < desc_cnt; i++) {
1316 io_req->io_desc[i].sector_mask = 0;
1317 io_req->io_desc[i].cow_offset = -1;
1318 }
1319
1320 return io_req;
1321}
1322
1323static int ubd_submit_request(struct ubd *dev, struct request *req)
1324{
1325 int segs = 0;
1326 struct io_thread_req *io_req;
1327 int ret;
1328 enum req_op op = req_op(req);
1329
1330 if (op == REQ_OP_FLUSH)
1331 segs = 0;
1332 else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1333 segs = 1;
1334 else
1335 segs = blk_rq_nr_phys_segments(req);
1336
1337 io_req = ubd_alloc_req(dev, req, segs);
1338 if (!io_req)
1339 return -ENOMEM;
1340
1341 io_req->desc_cnt = segs;
1342 if (segs)
1343 ubd_map_req(dev, io_req, req);
1344
1345 ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1346 if (ret != sizeof(io_req)) {
1347 if (ret != -EAGAIN)
1348 pr_err("write to io thread failed: %d\n", -ret);
1349 kfree(io_req);
1350 }
1351 return ret;
1352}
1353
1354static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1355 const struct blk_mq_queue_data *bd)
1356{
1357 struct ubd *ubd_dev = hctx->queue->queuedata;
1358 struct request *req = bd->rq;
1359 int ret = 0, res = BLK_STS_OK;
1360
1361 blk_mq_start_request(req);
1362
1363 spin_lock_irq(&ubd_dev->lock);
1364
1365 switch (req_op(req)) {
1366 case REQ_OP_FLUSH:
1367 case REQ_OP_READ:
1368 case REQ_OP_WRITE:
1369 case REQ_OP_DISCARD:
1370 case REQ_OP_WRITE_ZEROES:
1371 ret = ubd_submit_request(ubd_dev, req);
1372 break;
1373 default:
1374 WARN_ON_ONCE(1);
1375 res = BLK_STS_NOTSUPP;
1376 }
1377
1378 spin_unlock_irq(&ubd_dev->lock);
1379
1380 if (ret < 0) {
1381 if (ret == -ENOMEM)
1382 res = BLK_STS_RESOURCE;
1383 else
1384 res = BLK_STS_DEV_RESOURCE;
1385 }
1386
1387 return res;
1388}
1389
1390static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1391{
1392 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1393
1394 geo->heads = 128;
1395 geo->sectors = 32;
1396 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1397 return 0;
1398}
1399
1400static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1401 unsigned int cmd, unsigned long arg)
1402{
1403 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1404 u16 ubd_id[ATA_ID_WORDS];
1405
1406 switch (cmd) {
1407 struct cdrom_volctrl volume;
1408 case HDIO_GET_IDENTITY:
1409 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1410 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1411 ubd_id[ATA_ID_HEADS] = 128;
1412 ubd_id[ATA_ID_SECTORS] = 32;
1413 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1414 sizeof(ubd_id)))
1415 return -EFAULT;
1416 return 0;
1417
1418 case CDROMVOLREAD:
1419 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1420 return -EFAULT;
1421 volume.channel0 = 255;
1422 volume.channel1 = 255;
1423 volume.channel2 = 255;
1424 volume.channel3 = 255;
1425 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1426 return -EFAULT;
1427 return 0;
1428 }
1429 return -EINVAL;
1430}
1431
1432static int map_error(int error_code)
1433{
1434 switch (error_code) {
1435 case 0:
1436 return BLK_STS_OK;
1437 case ENOSYS:
1438 case EOPNOTSUPP:
1439 return BLK_STS_NOTSUPP;
1440 case ENOSPC:
1441 return BLK_STS_NOSPC;
1442 }
1443 return BLK_STS_IOERR;
1444}
1445
1446/*
1447 * Everything from here onwards *IS NOT PART OF THE KERNEL*
1448 *
1449 * The following functions are part of UML hypervisor code.
1450 * All functions from here onwards are executed as a helper
1451 * thread and are not allowed to execute any kernel functions.
1452 *
1453 * Any communication must occur strictly via shared memory and IPC.
1454 *
1455 * Do not add printks, locks, kernel memory operations, etc - it
1456 * will result in unpredictable behaviour and/or crashes.
1457 */
1458
1459static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1460{
1461 int n;
1462
1463 if (segment->cow_offset == -1)
1464 return map_error(0);
1465
1466 n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1467 sizeof(segment->bitmap_words), segment->cow_offset);
1468 if (n != sizeof(segment->bitmap_words))
1469 return map_error(-n);
1470
1471 return map_error(0);
1472}
1473
1474static void do_io(struct io_thread_req *req, struct io_desc *desc)
1475{
1476 char *buf = NULL;
1477 unsigned long len;
1478 int n, nsectors, start, end, bit;
1479 __u64 off;
1480
1481 /* FLUSH is really a special case, we cannot "case" it with others */
1482
1483 if (req_op(req->req) == REQ_OP_FLUSH) {
1484 /* fds[0] is always either the rw image or our cow file */
1485 req->error = map_error(-os_sync_file(req->fds[0]));
1486 return;
1487 }
1488
1489 nsectors = desc->length / req->sectorsize;
1490 start = 0;
1491 do {
1492 bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1493 end = start;
1494 while((end < nsectors) &&
1495 (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1496 end++;
1497
1498 off = req->offset + req->offsets[bit] +
1499 start * req->sectorsize;
1500 len = (end - start) * req->sectorsize;
1501 if (desc->buffer != NULL)
1502 buf = &desc->buffer[start * req->sectorsize];
1503
1504 switch (req_op(req->req)) {
1505 case REQ_OP_READ:
1506 n = 0;
1507 do {
1508 buf = &buf[n];
1509 len -= n;
1510 n = os_pread_file(req->fds[bit], buf, len, off);
1511 if (n < 0) {
1512 req->error = map_error(-n);
1513 return;
1514 }
1515 } while((n < len) && (n != 0));
1516 if (n < len) memset(&buf[n], 0, len - n);
1517 break;
1518 case REQ_OP_WRITE:
1519 n = os_pwrite_file(req->fds[bit], buf, len, off);
1520 if(n != len){
1521 req->error = map_error(-n);
1522 return;
1523 }
1524 break;
1525 case REQ_OP_DISCARD:
1526 n = os_falloc_punch(req->fds[bit], off, len);
1527 if (n) {
1528 req->error = map_error(-n);
1529 return;
1530 }
1531 break;
1532 case REQ_OP_WRITE_ZEROES:
1533 n = os_falloc_zeroes(req->fds[bit], off, len);
1534 if (n) {
1535 req->error = map_error(-n);
1536 return;
1537 }
1538 break;
1539 default:
1540 WARN_ON_ONCE(1);
1541 req->error = BLK_STS_NOTSUPP;
1542 return;
1543 }
1544
1545 start = end;
1546 } while(start < nsectors);
1547
1548 req->offset += len;
1549 req->error = update_bitmap(req, desc);
1550}
1551
1552/* Changed in start_io_thread, which is serialized by being called only
1553 * from ubd_init, which is an initcall.
1554 */
1555int kernel_fd = -1;
1556
1557/* Only changed by the io thread. XXX: currently unused. */
1558static int io_count;
1559
1560int io_thread(void *arg)
1561{
1562 int n, count, written, res;
1563
1564 os_fix_helper_signals();
1565
1566 while(1){
1567 n = bulk_req_safe_read(
1568 kernel_fd,
1569 io_req_buffer,
1570 &io_remainder,
1571 &io_remainder_size,
1572 UBD_REQ_BUFFER_SIZE
1573 );
1574 if (n <= 0) {
1575 if (n == -EAGAIN)
1576 ubd_read_poll(-1);
1577
1578 continue;
1579 }
1580
1581 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1582 struct io_thread_req *req = (*io_req_buffer)[count];
1583 int i;
1584
1585 io_count++;
1586 for (i = 0; !req->error && i < req->desc_cnt; i++)
1587 do_io(req, &(req->io_desc[i]));
1588
1589 }
1590
1591 written = 0;
1592
1593 do {
1594 res = os_write_file(kernel_fd,
1595 ((char *) io_req_buffer) + written,
1596 n - written);
1597 if (res >= 0) {
1598 written += res;
1599 }
1600 if (written < n) {
1601 ubd_write_poll(-1);
1602 }
1603 } while (written < n);
1604 }
1605
1606 return 0;
1607}
1/*
2 * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3 * Licensed under the GPL
4 */
5
6/* 2001-09-28...2002-04-17
7 * Partition stuff by James_McMechan@hotmail.com
8 * old style ubd by setting UBD_SHIFT to 0
9 * 2002-09-27...2002-10-18 massive tinkering for 2.5
10 * partitions have changed in 2.5
11 * 2003-01-29 more tinkering for 2.5.59-1
12 * This should now address the sysfs problems and has
13 * the symlink for devfs to allow for booting with
14 * the common /dev/ubd/discX/... names rather than
15 * only /dev/ubdN/discN this version also has lots of
16 * clean ups preparing for ubd-many.
17 * James McMechan
18 */
19
20#define UBD_SHIFT 4
21
22#include <linux/module.h>
23#include <linux/init.h>
24#include <linux/blkdev.h>
25#include <linux/ata.h>
26#include <linux/hdreg.h>
27#include <linux/cdrom.h>
28#include <linux/proc_fs.h>
29#include <linux/seq_file.h>
30#include <linux/ctype.h>
31#include <linux/slab.h>
32#include <linux/vmalloc.h>
33#include <linux/platform_device.h>
34#include <linux/scatterlist.h>
35#include <asm/tlbflush.h>
36#include <kern_util.h>
37#include "mconsole_kern.h"
38#include <init.h>
39#include <irq_kern.h>
40#include "ubd.h"
41#include <os.h>
42#include "cow.h"
43
44enum ubd_req { UBD_READ, UBD_WRITE, UBD_FLUSH };
45
46struct io_thread_req {
47 struct request *req;
48 enum ubd_req op;
49 int fds[2];
50 unsigned long offsets[2];
51 unsigned long long offset;
52 unsigned long length;
53 char *buffer;
54 int sectorsize;
55 unsigned long sector_mask;
56 unsigned long long cow_offset;
57 unsigned long bitmap_words[2];
58 int error;
59};
60
61static inline int ubd_test_bit(__u64 bit, unsigned char *data)
62{
63 __u64 n;
64 int bits, off;
65
66 bits = sizeof(data[0]) * 8;
67 n = bit / bits;
68 off = bit % bits;
69 return (data[n] & (1 << off)) != 0;
70}
71
72static inline void ubd_set_bit(__u64 bit, unsigned char *data)
73{
74 __u64 n;
75 int bits, off;
76
77 bits = sizeof(data[0]) * 8;
78 n = bit / bits;
79 off = bit % bits;
80 data[n] |= (1 << off);
81}
82/*End stuff from ubd_user.h*/
83
84#define DRIVER_NAME "uml-blkdev"
85
86static DEFINE_MUTEX(ubd_lock);
87static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
88
89static int ubd_open(struct block_device *bdev, fmode_t mode);
90static void ubd_release(struct gendisk *disk, fmode_t mode);
91static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
92 unsigned int cmd, unsigned long arg);
93static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
94
95#define MAX_DEV (16)
96
97static const struct block_device_operations ubd_blops = {
98 .owner = THIS_MODULE,
99 .open = ubd_open,
100 .release = ubd_release,
101 .ioctl = ubd_ioctl,
102 .getgeo = ubd_getgeo,
103};
104
105/* Protected by ubd_lock */
106static int fake_major = UBD_MAJOR;
107static struct gendisk *ubd_gendisk[MAX_DEV];
108static struct gendisk *fake_gendisk[MAX_DEV];
109
110#ifdef CONFIG_BLK_DEV_UBD_SYNC
111#define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
112 .cl = 1 })
113#else
114#define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
115 .cl = 1 })
116#endif
117static struct openflags global_openflags = OPEN_FLAGS;
118
119struct cow {
120 /* backing file name */
121 char *file;
122 /* backing file fd */
123 int fd;
124 unsigned long *bitmap;
125 unsigned long bitmap_len;
126 int bitmap_offset;
127 int data_offset;
128};
129
130#define MAX_SG 64
131
132struct ubd {
133 struct list_head restart;
134 /* name (and fd, below) of the file opened for writing, either the
135 * backing or the cow file. */
136 char *file;
137 int count;
138 int fd;
139 __u64 size;
140 struct openflags boot_openflags;
141 struct openflags openflags;
142 unsigned shared:1;
143 unsigned no_cow:1;
144 struct cow cow;
145 struct platform_device pdev;
146 struct request_queue *queue;
147 spinlock_t lock;
148 struct scatterlist sg[MAX_SG];
149 struct request *request;
150 int start_sg, end_sg;
151 sector_t rq_pos;
152};
153
154#define DEFAULT_COW { \
155 .file = NULL, \
156 .fd = -1, \
157 .bitmap = NULL, \
158 .bitmap_offset = 0, \
159 .data_offset = 0, \
160}
161
162#define DEFAULT_UBD { \
163 .file = NULL, \
164 .count = 0, \
165 .fd = -1, \
166 .size = -1, \
167 .boot_openflags = OPEN_FLAGS, \
168 .openflags = OPEN_FLAGS, \
169 .no_cow = 0, \
170 .shared = 0, \
171 .cow = DEFAULT_COW, \
172 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
173 .request = NULL, \
174 .start_sg = 0, \
175 .end_sg = 0, \
176 .rq_pos = 0, \
177}
178
179/* Protected by ubd_lock */
180static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
181
182/* Only changed by fake_ide_setup which is a setup */
183static int fake_ide = 0;
184static struct proc_dir_entry *proc_ide_root = NULL;
185static struct proc_dir_entry *proc_ide = NULL;
186
187static void make_proc_ide(void)
188{
189 proc_ide_root = proc_mkdir("ide", NULL);
190 proc_ide = proc_mkdir("ide0", proc_ide_root);
191}
192
193static int fake_ide_media_proc_show(struct seq_file *m, void *v)
194{
195 seq_puts(m, "disk\n");
196 return 0;
197}
198
199static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
200{
201 return single_open(file, fake_ide_media_proc_show, NULL);
202}
203
204static const struct file_operations fake_ide_media_proc_fops = {
205 .owner = THIS_MODULE,
206 .open = fake_ide_media_proc_open,
207 .read = seq_read,
208 .llseek = seq_lseek,
209 .release = single_release,
210};
211
212static void make_ide_entries(const char *dev_name)
213{
214 struct proc_dir_entry *dir, *ent;
215 char name[64];
216
217 if(proc_ide_root == NULL) make_proc_ide();
218
219 dir = proc_mkdir(dev_name, proc_ide);
220 if(!dir) return;
221
222 ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
223 if(!ent) return;
224 snprintf(name, sizeof(name), "ide0/%s", dev_name);
225 proc_symlink(dev_name, proc_ide_root, name);
226}
227
228static int fake_ide_setup(char *str)
229{
230 fake_ide = 1;
231 return 1;
232}
233
234__setup("fake_ide", fake_ide_setup);
235
236__uml_help(fake_ide_setup,
237"fake_ide\n"
238" Create ide0 entries that map onto ubd devices.\n\n"
239);
240
241static int parse_unit(char **ptr)
242{
243 char *str = *ptr, *end;
244 int n = -1;
245
246 if(isdigit(*str)) {
247 n = simple_strtoul(str, &end, 0);
248 if(end == str)
249 return -1;
250 *ptr = end;
251 }
252 else if (('a' <= *str) && (*str <= 'z')) {
253 n = *str - 'a';
254 str++;
255 *ptr = str;
256 }
257 return n;
258}
259
260/* If *index_out == -1 at exit, the passed option was a general one;
261 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
262 * should not be freed on exit.
263 */
264static int ubd_setup_common(char *str, int *index_out, char **error_out)
265{
266 struct ubd *ubd_dev;
267 struct openflags flags = global_openflags;
268 char *backing_file;
269 int n, err = 0, i;
270
271 if(index_out) *index_out = -1;
272 n = *str;
273 if(n == '='){
274 char *end;
275 int major;
276
277 str++;
278 if(!strcmp(str, "sync")){
279 global_openflags = of_sync(global_openflags);
280 goto out1;
281 }
282
283 err = -EINVAL;
284 major = simple_strtoul(str, &end, 0);
285 if((*end != '\0') || (end == str)){
286 *error_out = "Didn't parse major number";
287 goto out1;
288 }
289
290 mutex_lock(&ubd_lock);
291 if (fake_major != UBD_MAJOR) {
292 *error_out = "Can't assign a fake major twice";
293 goto out1;
294 }
295
296 fake_major = major;
297
298 printk(KERN_INFO "Setting extra ubd major number to %d\n",
299 major);
300 err = 0;
301 out1:
302 mutex_unlock(&ubd_lock);
303 return err;
304 }
305
306 n = parse_unit(&str);
307 if(n < 0){
308 *error_out = "Couldn't parse device number";
309 return -EINVAL;
310 }
311 if(n >= MAX_DEV){
312 *error_out = "Device number out of range";
313 return 1;
314 }
315
316 err = -EBUSY;
317 mutex_lock(&ubd_lock);
318
319 ubd_dev = &ubd_devs[n];
320 if(ubd_dev->file != NULL){
321 *error_out = "Device is already configured";
322 goto out;
323 }
324
325 if (index_out)
326 *index_out = n;
327
328 err = -EINVAL;
329 for (i = 0; i < sizeof("rscd="); i++) {
330 switch (*str) {
331 case 'r':
332 flags.w = 0;
333 break;
334 case 's':
335 flags.s = 1;
336 break;
337 case 'd':
338 ubd_dev->no_cow = 1;
339 break;
340 case 'c':
341 ubd_dev->shared = 1;
342 break;
343 case '=':
344 str++;
345 goto break_loop;
346 default:
347 *error_out = "Expected '=' or flag letter "
348 "(r, s, c, or d)";
349 goto out;
350 }
351 str++;
352 }
353
354 if (*str == '=')
355 *error_out = "Too many flags specified";
356 else
357 *error_out = "Missing '='";
358 goto out;
359
360break_loop:
361 backing_file = strchr(str, ',');
362
363 if (backing_file == NULL)
364 backing_file = strchr(str, ':');
365
366 if(backing_file != NULL){
367 if(ubd_dev->no_cow){
368 *error_out = "Can't specify both 'd' and a cow file";
369 goto out;
370 }
371 else {
372 *backing_file = '\0';
373 backing_file++;
374 }
375 }
376 err = 0;
377 ubd_dev->file = str;
378 ubd_dev->cow.file = backing_file;
379 ubd_dev->boot_openflags = flags;
380out:
381 mutex_unlock(&ubd_lock);
382 return err;
383}
384
385static int ubd_setup(char *str)
386{
387 char *error;
388 int err;
389
390 err = ubd_setup_common(str, NULL, &error);
391 if(err)
392 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
393 "%s\n", str, error);
394 return 1;
395}
396
397__setup("ubd", ubd_setup);
398__uml_help(ubd_setup,
399"ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
400" This is used to associate a device with a file in the underlying\n"
401" filesystem. When specifying two filenames, the first one is the\n"
402" COW name and the second is the backing file name. As separator you can\n"
403" use either a ':' or a ',': the first one allows writing things like;\n"
404" ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
405" while with a ',' the shell would not expand the 2nd '~'.\n"
406" When using only one filename, UML will detect whether to treat it like\n"
407" a COW file or a backing file. To override this detection, add the 'd'\n"
408" flag:\n"
409" ubd0d=BackingFile\n"
410" Usually, there is a filesystem in the file, but \n"
411" that's not required. Swap devices containing swap files can be\n"
412" specified like this. Also, a file which doesn't contain a\n"
413" filesystem can have its contents read in the virtual \n"
414" machine by running 'dd' on the device. <n> must be in the range\n"
415" 0 to 7. Appending an 'r' to the number will cause that device\n"
416" to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
417" an 's' will cause data to be written to disk on the host immediately.\n"
418" 'c' will cause the device to be treated as being shared between multiple\n"
419" UMLs and file locking will be turned off - this is appropriate for a\n"
420" cluster filesystem and inappropriate at almost all other times.\n\n"
421);
422
423static int udb_setup(char *str)
424{
425 printk("udb%s specified on command line is almost certainly a ubd -> "
426 "udb TYPO\n", str);
427 return 1;
428}
429
430__setup("udb", udb_setup);
431__uml_help(udb_setup,
432"udb\n"
433" This option is here solely to catch ubd -> udb typos, which can be\n"
434" to impossible to catch visually unless you specifically look for\n"
435" them. The only result of any option starting with 'udb' is an error\n"
436" in the boot output.\n\n"
437);
438
439static void do_ubd_request(struct request_queue * q);
440
441/* Only changed by ubd_init, which is an initcall. */
442static int thread_fd = -1;
443static LIST_HEAD(restart);
444
445/* XXX - move this inside ubd_intr. */
446/* Called without dev->lock held, and only in interrupt context. */
447static void ubd_handler(void)
448{
449 struct io_thread_req *req;
450 struct ubd *ubd;
451 struct list_head *list, *next_ele;
452 unsigned long flags;
453 int n;
454
455 while(1){
456 n = os_read_file(thread_fd, &req,
457 sizeof(struct io_thread_req *));
458 if(n != sizeof(req)){
459 if(n == -EAGAIN)
460 break;
461 printk(KERN_ERR "spurious interrupt in ubd_handler, "
462 "err = %d\n", -n);
463 return;
464 }
465
466 blk_end_request(req->req, 0, req->length);
467 kfree(req);
468 }
469 reactivate_fd(thread_fd, UBD_IRQ);
470
471 list_for_each_safe(list, next_ele, &restart){
472 ubd = container_of(list, struct ubd, restart);
473 list_del_init(&ubd->restart);
474 spin_lock_irqsave(&ubd->lock, flags);
475 do_ubd_request(ubd->queue);
476 spin_unlock_irqrestore(&ubd->lock, flags);
477 }
478}
479
480static irqreturn_t ubd_intr(int irq, void *dev)
481{
482 ubd_handler();
483 return IRQ_HANDLED;
484}
485
486/* Only changed by ubd_init, which is an initcall. */
487static int io_pid = -1;
488
489static void kill_io_thread(void)
490{
491 if(io_pid != -1)
492 os_kill_process(io_pid, 1);
493}
494
495__uml_exitcall(kill_io_thread);
496
497static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
498{
499 char *file;
500 int fd;
501 int err;
502
503 __u32 version;
504 __u32 align;
505 char *backing_file;
506 time_t mtime;
507 unsigned long long size;
508 int sector_size;
509 int bitmap_offset;
510
511 if (ubd_dev->file && ubd_dev->cow.file) {
512 file = ubd_dev->cow.file;
513
514 goto out;
515 }
516
517 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
518 if (fd < 0)
519 return fd;
520
521 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
522 &mtime, &size, §or_size, &align, &bitmap_offset);
523 os_close_file(fd);
524
525 if(err == -EINVAL)
526 file = ubd_dev->file;
527 else
528 file = backing_file;
529
530out:
531 return os_file_size(file, size_out);
532}
533
534static int read_cow_bitmap(int fd, void *buf, int offset, int len)
535{
536 int err;
537
538 err = os_pread_file(fd, buf, len, offset);
539 if (err < 0)
540 return err;
541
542 return 0;
543}
544
545static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
546{
547 unsigned long modtime;
548 unsigned long long actual;
549 int err;
550
551 err = os_file_modtime(file, &modtime);
552 if (err < 0) {
553 printk(KERN_ERR "Failed to get modification time of backing "
554 "file \"%s\", err = %d\n", file, -err);
555 return err;
556 }
557
558 err = os_file_size(file, &actual);
559 if (err < 0) {
560 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
561 "err = %d\n", file, -err);
562 return err;
563 }
564
565 if (actual != size) {
566 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
567 * the typecast.*/
568 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
569 "vs backing file\n", (unsigned long long) size, actual);
570 return -EINVAL;
571 }
572 if (modtime != mtime) {
573 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
574 "backing file\n", mtime, modtime);
575 return -EINVAL;
576 }
577 return 0;
578}
579
580static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
581{
582 struct uml_stat buf1, buf2;
583 int err;
584
585 if (from_cmdline == NULL)
586 return 0;
587 if (!strcmp(from_cmdline, from_cow))
588 return 0;
589
590 err = os_stat_file(from_cmdline, &buf1);
591 if (err < 0) {
592 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
593 -err);
594 return 0;
595 }
596 err = os_stat_file(from_cow, &buf2);
597 if (err < 0) {
598 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
599 -err);
600 return 1;
601 }
602 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
603 return 0;
604
605 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
606 "\"%s\" specified in COW header of \"%s\"\n",
607 from_cmdline, from_cow, cow);
608 return 1;
609}
610
611static int open_ubd_file(char *file, struct openflags *openflags, int shared,
612 char **backing_file_out, int *bitmap_offset_out,
613 unsigned long *bitmap_len_out, int *data_offset_out,
614 int *create_cow_out)
615{
616 time_t mtime;
617 unsigned long long size;
618 __u32 version, align;
619 char *backing_file;
620 int fd, err, sectorsize, asked_switch, mode = 0644;
621
622 fd = os_open_file(file, *openflags, mode);
623 if (fd < 0) {
624 if ((fd == -ENOENT) && (create_cow_out != NULL))
625 *create_cow_out = 1;
626 if (!openflags->w ||
627 ((fd != -EROFS) && (fd != -EACCES)))
628 return fd;
629 openflags->w = 0;
630 fd = os_open_file(file, *openflags, mode);
631 if (fd < 0)
632 return fd;
633 }
634
635 if (shared)
636 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
637 else {
638 err = os_lock_file(fd, openflags->w);
639 if (err < 0) {
640 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
641 file, -err);
642 goto out_close;
643 }
644 }
645
646 /* Successful return case! */
647 if (backing_file_out == NULL)
648 return fd;
649
650 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
651 &size, §orsize, &align, bitmap_offset_out);
652 if (err && (*backing_file_out != NULL)) {
653 printk(KERN_ERR "Failed to read COW header from COW file "
654 "\"%s\", errno = %d\n", file, -err);
655 goto out_close;
656 }
657 if (err)
658 return fd;
659
660 asked_switch = path_requires_switch(*backing_file_out, backing_file,
661 file);
662
663 /* Allow switching only if no mismatch. */
664 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
665 mtime)) {
666 printk(KERN_ERR "Switching backing file to '%s'\n",
667 *backing_file_out);
668 err = write_cow_header(file, fd, *backing_file_out,
669 sectorsize, align, &size);
670 if (err) {
671 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
672 goto out_close;
673 }
674 } else {
675 *backing_file_out = backing_file;
676 err = backing_file_mismatch(*backing_file_out, size, mtime);
677 if (err)
678 goto out_close;
679 }
680
681 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
682 bitmap_len_out, data_offset_out);
683
684 return fd;
685 out_close:
686 os_close_file(fd);
687 return err;
688}
689
690static int create_cow_file(char *cow_file, char *backing_file,
691 struct openflags flags,
692 int sectorsize, int alignment, int *bitmap_offset_out,
693 unsigned long *bitmap_len_out, int *data_offset_out)
694{
695 int err, fd;
696
697 flags.c = 1;
698 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
699 if (fd < 0) {
700 err = fd;
701 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
702 cow_file, -err);
703 goto out;
704 }
705
706 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
707 bitmap_offset_out, bitmap_len_out,
708 data_offset_out);
709 if (!err)
710 return fd;
711 os_close_file(fd);
712 out:
713 return err;
714}
715
716static void ubd_close_dev(struct ubd *ubd_dev)
717{
718 os_close_file(ubd_dev->fd);
719 if(ubd_dev->cow.file == NULL)
720 return;
721
722 os_close_file(ubd_dev->cow.fd);
723 vfree(ubd_dev->cow.bitmap);
724 ubd_dev->cow.bitmap = NULL;
725}
726
727static int ubd_open_dev(struct ubd *ubd_dev)
728{
729 struct openflags flags;
730 char **back_ptr;
731 int err, create_cow, *create_ptr;
732 int fd;
733
734 ubd_dev->openflags = ubd_dev->boot_openflags;
735 create_cow = 0;
736 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
737 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
738
739 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
740 back_ptr, &ubd_dev->cow.bitmap_offset,
741 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
742 create_ptr);
743
744 if((fd == -ENOENT) && create_cow){
745 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
746 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
747 &ubd_dev->cow.bitmap_offset,
748 &ubd_dev->cow.bitmap_len,
749 &ubd_dev->cow.data_offset);
750 if(fd >= 0){
751 printk(KERN_INFO "Creating \"%s\" as COW file for "
752 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
753 }
754 }
755
756 if(fd < 0){
757 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
758 -fd);
759 return fd;
760 }
761 ubd_dev->fd = fd;
762
763 if(ubd_dev->cow.file != NULL){
764 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
765
766 err = -ENOMEM;
767 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
768 if(ubd_dev->cow.bitmap == NULL){
769 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
770 goto error;
771 }
772 flush_tlb_kernel_vm();
773
774 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
775 ubd_dev->cow.bitmap_offset,
776 ubd_dev->cow.bitmap_len);
777 if(err < 0)
778 goto error;
779
780 flags = ubd_dev->openflags;
781 flags.w = 0;
782 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
783 NULL, NULL, NULL, NULL);
784 if(err < 0) goto error;
785 ubd_dev->cow.fd = err;
786 }
787 return 0;
788 error:
789 os_close_file(ubd_dev->fd);
790 return err;
791}
792
793static void ubd_device_release(struct device *dev)
794{
795 struct ubd *ubd_dev = dev_get_drvdata(dev);
796
797 blk_cleanup_queue(ubd_dev->queue);
798 *ubd_dev = ((struct ubd) DEFAULT_UBD);
799}
800
801static int ubd_disk_register(int major, u64 size, int unit,
802 struct gendisk **disk_out)
803{
804 struct gendisk *disk;
805
806 disk = alloc_disk(1 << UBD_SHIFT);
807 if(disk == NULL)
808 return -ENOMEM;
809
810 disk->major = major;
811 disk->first_minor = unit << UBD_SHIFT;
812 disk->fops = &ubd_blops;
813 set_capacity(disk, size / 512);
814 if (major == UBD_MAJOR)
815 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
816 else
817 sprintf(disk->disk_name, "ubd_fake%d", unit);
818
819 /* sysfs register (not for ide fake devices) */
820 if (major == UBD_MAJOR) {
821 ubd_devs[unit].pdev.id = unit;
822 ubd_devs[unit].pdev.name = DRIVER_NAME;
823 ubd_devs[unit].pdev.dev.release = ubd_device_release;
824 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
825 platform_device_register(&ubd_devs[unit].pdev);
826 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
827 }
828
829 disk->private_data = &ubd_devs[unit];
830 disk->queue = ubd_devs[unit].queue;
831 add_disk(disk);
832
833 *disk_out = disk;
834 return 0;
835}
836
837#define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
838
839static int ubd_add(int n, char **error_out)
840{
841 struct ubd *ubd_dev = &ubd_devs[n];
842 int err = 0;
843
844 if(ubd_dev->file == NULL)
845 goto out;
846
847 err = ubd_file_size(ubd_dev, &ubd_dev->size);
848 if(err < 0){
849 *error_out = "Couldn't determine size of device's file";
850 goto out;
851 }
852
853 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
854
855 INIT_LIST_HEAD(&ubd_dev->restart);
856 sg_init_table(ubd_dev->sg, MAX_SG);
857
858 err = -ENOMEM;
859 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
860 if (ubd_dev->queue == NULL) {
861 *error_out = "Failed to initialize device queue";
862 goto out;
863 }
864 ubd_dev->queue->queuedata = ubd_dev;
865 blk_queue_flush(ubd_dev->queue, REQ_FLUSH);
866
867 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
868 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
869 if(err){
870 *error_out = "Failed to register device";
871 goto out_cleanup;
872 }
873
874 if (fake_major != UBD_MAJOR)
875 ubd_disk_register(fake_major, ubd_dev->size, n,
876 &fake_gendisk[n]);
877
878 /*
879 * Perhaps this should also be under the "if (fake_major)" above
880 * using the fake_disk->disk_name
881 */
882 if (fake_ide)
883 make_ide_entries(ubd_gendisk[n]->disk_name);
884
885 err = 0;
886out:
887 return err;
888
889out_cleanup:
890 blk_cleanup_queue(ubd_dev->queue);
891 goto out;
892}
893
894static int ubd_config(char *str, char **error_out)
895{
896 int n, ret;
897
898 /* This string is possibly broken up and stored, so it's only
899 * freed if ubd_setup_common fails, or if only general options
900 * were set.
901 */
902 str = kstrdup(str, GFP_KERNEL);
903 if (str == NULL) {
904 *error_out = "Failed to allocate memory";
905 return -ENOMEM;
906 }
907
908 ret = ubd_setup_common(str, &n, error_out);
909 if (ret)
910 goto err_free;
911
912 if (n == -1) {
913 ret = 0;
914 goto err_free;
915 }
916
917 mutex_lock(&ubd_lock);
918 ret = ubd_add(n, error_out);
919 if (ret)
920 ubd_devs[n].file = NULL;
921 mutex_unlock(&ubd_lock);
922
923out:
924 return ret;
925
926err_free:
927 kfree(str);
928 goto out;
929}
930
931static int ubd_get_config(char *name, char *str, int size, char **error_out)
932{
933 struct ubd *ubd_dev;
934 int n, len = 0;
935
936 n = parse_unit(&name);
937 if((n >= MAX_DEV) || (n < 0)){
938 *error_out = "ubd_get_config : device number out of range";
939 return -1;
940 }
941
942 ubd_dev = &ubd_devs[n];
943 mutex_lock(&ubd_lock);
944
945 if(ubd_dev->file == NULL){
946 CONFIG_CHUNK(str, size, len, "", 1);
947 goto out;
948 }
949
950 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
951
952 if(ubd_dev->cow.file != NULL){
953 CONFIG_CHUNK(str, size, len, ",", 0);
954 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
955 }
956 else CONFIG_CHUNK(str, size, len, "", 1);
957
958 out:
959 mutex_unlock(&ubd_lock);
960 return len;
961}
962
963static int ubd_id(char **str, int *start_out, int *end_out)
964{
965 int n;
966
967 n = parse_unit(str);
968 *start_out = 0;
969 *end_out = MAX_DEV - 1;
970 return n;
971}
972
973static int ubd_remove(int n, char **error_out)
974{
975 struct gendisk *disk = ubd_gendisk[n];
976 struct ubd *ubd_dev;
977 int err = -ENODEV;
978
979 mutex_lock(&ubd_lock);
980
981 ubd_dev = &ubd_devs[n];
982
983 if(ubd_dev->file == NULL)
984 goto out;
985
986 /* you cannot remove a open disk */
987 err = -EBUSY;
988 if(ubd_dev->count > 0)
989 goto out;
990
991 ubd_gendisk[n] = NULL;
992 if(disk != NULL){
993 del_gendisk(disk);
994 put_disk(disk);
995 }
996
997 if(fake_gendisk[n] != NULL){
998 del_gendisk(fake_gendisk[n]);
999 put_disk(fake_gendisk[n]);
1000 fake_gendisk[n] = NULL;
1001 }
1002
1003 err = 0;
1004 platform_device_unregister(&ubd_dev->pdev);
1005out:
1006 mutex_unlock(&ubd_lock);
1007 return err;
1008}
1009
1010/* All these are called by mconsole in process context and without
1011 * ubd-specific locks. The structure itself is const except for .list.
1012 */
1013static struct mc_device ubd_mc = {
1014 .list = LIST_HEAD_INIT(ubd_mc.list),
1015 .name = "ubd",
1016 .config = ubd_config,
1017 .get_config = ubd_get_config,
1018 .id = ubd_id,
1019 .remove = ubd_remove,
1020};
1021
1022static int __init ubd_mc_init(void)
1023{
1024 mconsole_register_dev(&ubd_mc);
1025 return 0;
1026}
1027
1028__initcall(ubd_mc_init);
1029
1030static int __init ubd0_init(void)
1031{
1032 struct ubd *ubd_dev = &ubd_devs[0];
1033
1034 mutex_lock(&ubd_lock);
1035 if(ubd_dev->file == NULL)
1036 ubd_dev->file = "root_fs";
1037 mutex_unlock(&ubd_lock);
1038
1039 return 0;
1040}
1041
1042__initcall(ubd0_init);
1043
1044/* Used in ubd_init, which is an initcall */
1045static struct platform_driver ubd_driver = {
1046 .driver = {
1047 .name = DRIVER_NAME,
1048 },
1049};
1050
1051static int __init ubd_init(void)
1052{
1053 char *error;
1054 int i, err;
1055
1056 if (register_blkdev(UBD_MAJOR, "ubd"))
1057 return -1;
1058
1059 if (fake_major != UBD_MAJOR) {
1060 char name[sizeof("ubd_nnn\0")];
1061
1062 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1063 if (register_blkdev(fake_major, "ubd"))
1064 return -1;
1065 }
1066 platform_driver_register(&ubd_driver);
1067 mutex_lock(&ubd_lock);
1068 for (i = 0; i < MAX_DEV; i++){
1069 err = ubd_add(i, &error);
1070 if(err)
1071 printk(KERN_ERR "Failed to initialize ubd device %d :"
1072 "%s\n", i, error);
1073 }
1074 mutex_unlock(&ubd_lock);
1075 return 0;
1076}
1077
1078late_initcall(ubd_init);
1079
1080static int __init ubd_driver_init(void){
1081 unsigned long stack;
1082 int err;
1083
1084 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1085 if(global_openflags.s){
1086 printk(KERN_INFO "ubd: Synchronous mode\n");
1087 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1088 * enough. So use anyway the io thread. */
1089 }
1090 stack = alloc_stack(0, 0);
1091 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1092 &thread_fd);
1093 if(io_pid < 0){
1094 printk(KERN_ERR
1095 "ubd : Failed to start I/O thread (errno = %d) - "
1096 "falling back to synchronous I/O\n", -io_pid);
1097 io_pid = -1;
1098 return 0;
1099 }
1100 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1101 0, "ubd", ubd_devs);
1102 if(err != 0)
1103 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1104 return 0;
1105}
1106
1107device_initcall(ubd_driver_init);
1108
1109static int ubd_open(struct block_device *bdev, fmode_t mode)
1110{
1111 struct gendisk *disk = bdev->bd_disk;
1112 struct ubd *ubd_dev = disk->private_data;
1113 int err = 0;
1114
1115 mutex_lock(&ubd_mutex);
1116 if(ubd_dev->count == 0){
1117 err = ubd_open_dev(ubd_dev);
1118 if(err){
1119 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1120 disk->disk_name, ubd_dev->file, -err);
1121 goto out;
1122 }
1123 }
1124 ubd_dev->count++;
1125 set_disk_ro(disk, !ubd_dev->openflags.w);
1126
1127 /* This should no more be needed. And it didn't work anyway to exclude
1128 * read-write remounting of filesystems.*/
1129 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1130 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1131 err = -EROFS;
1132 }*/
1133out:
1134 mutex_unlock(&ubd_mutex);
1135 return err;
1136}
1137
1138static void ubd_release(struct gendisk *disk, fmode_t mode)
1139{
1140 struct ubd *ubd_dev = disk->private_data;
1141
1142 mutex_lock(&ubd_mutex);
1143 if(--ubd_dev->count == 0)
1144 ubd_close_dev(ubd_dev);
1145 mutex_unlock(&ubd_mutex);
1146}
1147
1148static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1149 __u64 *cow_offset, unsigned long *bitmap,
1150 __u64 bitmap_offset, unsigned long *bitmap_words,
1151 __u64 bitmap_len)
1152{
1153 __u64 sector = io_offset >> 9;
1154 int i, update_bitmap = 0;
1155
1156 for(i = 0; i < length >> 9; i++){
1157 if(cow_mask != NULL)
1158 ubd_set_bit(i, (unsigned char *) cow_mask);
1159 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1160 continue;
1161
1162 update_bitmap = 1;
1163 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1164 }
1165
1166 if(!update_bitmap)
1167 return;
1168
1169 *cow_offset = sector / (sizeof(unsigned long) * 8);
1170
1171 /* This takes care of the case where we're exactly at the end of the
1172 * device, and *cow_offset + 1 is off the end. So, just back it up
1173 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1174 * for the original diagnosis.
1175 */
1176 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1177 sizeof(unsigned long)) - 1))
1178 (*cow_offset)--;
1179
1180 bitmap_words[0] = bitmap[*cow_offset];
1181 bitmap_words[1] = bitmap[*cow_offset + 1];
1182
1183 *cow_offset *= sizeof(unsigned long);
1184 *cow_offset += bitmap_offset;
1185}
1186
1187static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1188 __u64 bitmap_offset, __u64 bitmap_len)
1189{
1190 __u64 sector = req->offset >> 9;
1191 int i;
1192
1193 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1194 panic("Operation too long");
1195
1196 if(req->op == UBD_READ) {
1197 for(i = 0; i < req->length >> 9; i++){
1198 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1199 ubd_set_bit(i, (unsigned char *)
1200 &req->sector_mask);
1201 }
1202 }
1203 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1204 &req->cow_offset, bitmap, bitmap_offset,
1205 req->bitmap_words, bitmap_len);
1206}
1207
1208/* Called with dev->lock held */
1209static void prepare_request(struct request *req, struct io_thread_req *io_req,
1210 unsigned long long offset, int page_offset,
1211 int len, struct page *page)
1212{
1213 struct gendisk *disk = req->rq_disk;
1214 struct ubd *ubd_dev = disk->private_data;
1215
1216 io_req->req = req;
1217 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1218 ubd_dev->fd;
1219 io_req->fds[1] = ubd_dev->fd;
1220 io_req->cow_offset = -1;
1221 io_req->offset = offset;
1222 io_req->length = len;
1223 io_req->error = 0;
1224 io_req->sector_mask = 0;
1225
1226 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1227 io_req->offsets[0] = 0;
1228 io_req->offsets[1] = ubd_dev->cow.data_offset;
1229 io_req->buffer = page_address(page) + page_offset;
1230 io_req->sectorsize = 1 << 9;
1231
1232 if(ubd_dev->cow.file != NULL)
1233 cowify_req(io_req, ubd_dev->cow.bitmap,
1234 ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1235
1236}
1237
1238/* Called with dev->lock held */
1239static void prepare_flush_request(struct request *req,
1240 struct io_thread_req *io_req)
1241{
1242 struct gendisk *disk = req->rq_disk;
1243 struct ubd *ubd_dev = disk->private_data;
1244
1245 io_req->req = req;
1246 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1247 ubd_dev->fd;
1248 io_req->op = UBD_FLUSH;
1249}
1250
1251static bool submit_request(struct io_thread_req *io_req, struct ubd *dev)
1252{
1253 int n = os_write_file(thread_fd, &io_req,
1254 sizeof(io_req));
1255 if (n != sizeof(io_req)) {
1256 if (n != -EAGAIN)
1257 printk("write to io thread failed, "
1258 "errno = %d\n", -n);
1259 else if (list_empty(&dev->restart))
1260 list_add(&dev->restart, &restart);
1261
1262 kfree(io_req);
1263 return false;
1264 }
1265 return true;
1266}
1267
1268/* Called with dev->lock held */
1269static void do_ubd_request(struct request_queue *q)
1270{
1271 struct io_thread_req *io_req;
1272 struct request *req;
1273
1274 while(1){
1275 struct ubd *dev = q->queuedata;
1276 if(dev->request == NULL){
1277 struct request *req = blk_fetch_request(q);
1278 if(req == NULL)
1279 return;
1280
1281 dev->request = req;
1282 dev->rq_pos = blk_rq_pos(req);
1283 dev->start_sg = 0;
1284 dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1285 }
1286
1287 req = dev->request;
1288
1289 if (req->cmd_flags & REQ_FLUSH) {
1290 io_req = kmalloc(sizeof(struct io_thread_req),
1291 GFP_ATOMIC);
1292 if (io_req == NULL) {
1293 if (list_empty(&dev->restart))
1294 list_add(&dev->restart, &restart);
1295 return;
1296 }
1297 prepare_flush_request(req, io_req);
1298 if (submit_request(io_req, dev) == false)
1299 return;
1300 }
1301
1302 while(dev->start_sg < dev->end_sg){
1303 struct scatterlist *sg = &dev->sg[dev->start_sg];
1304
1305 io_req = kmalloc(sizeof(struct io_thread_req),
1306 GFP_ATOMIC);
1307 if(io_req == NULL){
1308 if(list_empty(&dev->restart))
1309 list_add(&dev->restart, &restart);
1310 return;
1311 }
1312 prepare_request(req, io_req,
1313 (unsigned long long)dev->rq_pos << 9,
1314 sg->offset, sg->length, sg_page(sg));
1315
1316 if (submit_request(io_req, dev) == false)
1317 return;
1318
1319 dev->rq_pos += sg->length >> 9;
1320 dev->start_sg++;
1321 }
1322 dev->end_sg = 0;
1323 dev->request = NULL;
1324 }
1325}
1326
1327static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1328{
1329 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1330
1331 geo->heads = 128;
1332 geo->sectors = 32;
1333 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1334 return 0;
1335}
1336
1337static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1338 unsigned int cmd, unsigned long arg)
1339{
1340 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1341 u16 ubd_id[ATA_ID_WORDS];
1342
1343 switch (cmd) {
1344 struct cdrom_volctrl volume;
1345 case HDIO_GET_IDENTITY:
1346 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1347 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1348 ubd_id[ATA_ID_HEADS] = 128;
1349 ubd_id[ATA_ID_SECTORS] = 32;
1350 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1351 sizeof(ubd_id)))
1352 return -EFAULT;
1353 return 0;
1354
1355 case CDROMVOLREAD:
1356 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1357 return -EFAULT;
1358 volume.channel0 = 255;
1359 volume.channel1 = 255;
1360 volume.channel2 = 255;
1361 volume.channel3 = 255;
1362 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1363 return -EFAULT;
1364 return 0;
1365 }
1366 return -EINVAL;
1367}
1368
1369static int update_bitmap(struct io_thread_req *req)
1370{
1371 int n;
1372
1373 if(req->cow_offset == -1)
1374 return 0;
1375
1376 n = os_pwrite_file(req->fds[1], &req->bitmap_words,
1377 sizeof(req->bitmap_words), req->cow_offset);
1378 if(n != sizeof(req->bitmap_words)){
1379 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1380 req->fds[1]);
1381 return 1;
1382 }
1383
1384 return 0;
1385}
1386
1387static void do_io(struct io_thread_req *req)
1388{
1389 char *buf;
1390 unsigned long len;
1391 int n, nsectors, start, end, bit;
1392 __u64 off;
1393
1394 if (req->op == UBD_FLUSH) {
1395 /* fds[0] is always either the rw image or our cow file */
1396 n = os_sync_file(req->fds[0]);
1397 if (n != 0) {
1398 printk("do_io - sync failed err = %d "
1399 "fd = %d\n", -n, req->fds[0]);
1400 req->error = 1;
1401 }
1402 return;
1403 }
1404
1405 nsectors = req->length / req->sectorsize;
1406 start = 0;
1407 do {
1408 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1409 end = start;
1410 while((end < nsectors) &&
1411 (ubd_test_bit(end, (unsigned char *)
1412 &req->sector_mask) == bit))
1413 end++;
1414
1415 off = req->offset + req->offsets[bit] +
1416 start * req->sectorsize;
1417 len = (end - start) * req->sectorsize;
1418 buf = &req->buffer[start * req->sectorsize];
1419
1420 if(req->op == UBD_READ){
1421 n = 0;
1422 do {
1423 buf = &buf[n];
1424 len -= n;
1425 n = os_pread_file(req->fds[bit], buf, len, off);
1426 if (n < 0) {
1427 printk("do_io - read failed, err = %d "
1428 "fd = %d\n", -n, req->fds[bit]);
1429 req->error = 1;
1430 return;
1431 }
1432 } while((n < len) && (n != 0));
1433 if (n < len) memset(&buf[n], 0, len - n);
1434 } else {
1435 n = os_pwrite_file(req->fds[bit], buf, len, off);
1436 if(n != len){
1437 printk("do_io - write failed err = %d "
1438 "fd = %d\n", -n, req->fds[bit]);
1439 req->error = 1;
1440 return;
1441 }
1442 }
1443
1444 start = end;
1445 } while(start < nsectors);
1446
1447 req->error = update_bitmap(req);
1448}
1449
1450/* Changed in start_io_thread, which is serialized by being called only
1451 * from ubd_init, which is an initcall.
1452 */
1453int kernel_fd = -1;
1454
1455/* Only changed by the io thread. XXX: currently unused. */
1456static int io_count = 0;
1457
1458int io_thread(void *arg)
1459{
1460 struct io_thread_req *req;
1461 int n;
1462
1463 os_fix_helper_signals();
1464
1465 while(1){
1466 n = os_read_file(kernel_fd, &req,
1467 sizeof(struct io_thread_req *));
1468 if(n != sizeof(struct io_thread_req *)){
1469 if(n < 0)
1470 printk("io_thread - read failed, fd = %d, "
1471 "err = %d\n", kernel_fd, -n);
1472 else {
1473 printk("io_thread - short read, fd = %d, "
1474 "length = %d\n", kernel_fd, n);
1475 }
1476 continue;
1477 }
1478 io_count++;
1479 do_io(req);
1480 n = os_write_file(kernel_fd, &req,
1481 sizeof(struct io_thread_req *));
1482 if(n != sizeof(struct io_thread_req *))
1483 printk("io_thread - write failed, fd = %d, err = %d\n",
1484 kernel_fd, -n);
1485 }
1486
1487 return 0;
1488}