2 * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3 * Licensed under the GPL
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.
20 #define MAJOR_NR UBD_MAJOR
23 #include "linux/config.h"
24 #include "linux/module.h"
25 #include "linux/blkdev.h"
26 #include "linux/hdreg.h"
27 #include "linux/init.h"
28 #include "linux/devfs_fs_kernel.h"
29 #include "linux/cdrom.h"
30 #include "linux/proc_fs.h"
31 #include "linux/ctype.h"
32 #include "linux/capability.h"
34 #include "linux/vmalloc.h"
35 #include "linux/blkpg.h"
36 #include "linux/genhd.h"
37 #include "linux/spinlock.h"
38 #include "linux/platform_device.h"
39 #include "asm/segment.h"
40 #include "asm/uaccess.h"
42 #include "asm/types.h"
43 #include "asm/tlbflush.h"
44 #include "user_util.h"
46 #include "kern_util.h"
48 #include "mconsole_kern.h"
58 enum ubd_req
{ UBD_READ
, UBD_WRITE
};
60 struct io_thread_req
{
63 unsigned long offsets
[2];
64 unsigned long long offset
;
68 unsigned long sector_mask
;
69 unsigned long long cow_offset
;
70 unsigned long bitmap_words
[2];
74 extern int open_ubd_file(char *file
, struct openflags
*openflags
,
75 char **backing_file_out
, int *bitmap_offset_out
,
76 unsigned long *bitmap_len_out
, int *data_offset_out
,
78 extern int create_cow_file(char *cow_file
, char *backing_file
,
79 struct openflags flags
, int sectorsize
,
80 int alignment
, int *bitmap_offset_out
,
81 unsigned long *bitmap_len_out
,
82 int *data_offset_out
);
83 extern int read_cow_bitmap(int fd
, void *buf
, int offset
, int len
);
84 extern void do_io(struct io_thread_req
*req
);
86 static inline int ubd_test_bit(__u64 bit
, unsigned char *data
)
91 bits
= sizeof(data
[0]) * 8;
94 return((data
[n
] & (1 << off
)) != 0);
97 static inline void ubd_set_bit(__u64 bit
, unsigned char *data
)
102 bits
= sizeof(data
[0]) * 8;
105 data
[n
] |= (1 << off
);
107 /*End stuff from ubd_user.h*/
109 #define DRIVER_NAME "uml-blkdev"
111 static DEFINE_SPINLOCK(ubd_io_lock
);
112 static DEFINE_SPINLOCK(ubd_lock
);
114 static void (*do_ubd
)(void);
116 static int ubd_open(struct inode
* inode
, struct file
* filp
);
117 static int ubd_release(struct inode
* inode
, struct file
* file
);
118 static int ubd_ioctl(struct inode
* inode
, struct file
* file
,
119 unsigned int cmd
, unsigned long arg
);
123 static struct block_device_operations ubd_blops
= {
124 .owner
= THIS_MODULE
,
126 .release
= ubd_release
,
130 /* Protected by the queue_lock */
131 static request_queue_t
*ubd_queue
;
133 /* Protected by ubd_lock */
134 static int fake_major
= MAJOR_NR
;
136 static struct gendisk
*ubd_gendisk
[MAX_DEV
];
137 static struct gendisk
*fake_gendisk
[MAX_DEV
];
139 #ifdef CONFIG_BLK_DEV_UBD_SYNC
140 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
143 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
147 /* Not protected - changed only in ubd_setup_common and then only to
150 static struct openflags global_openflags
= OPEN_FLAGS
;
153 /* This is the backing file, actually */
156 unsigned long *bitmap
;
157 unsigned long bitmap_len
;
167 struct openflags boot_openflags
;
168 struct openflags openflags
;
171 struct platform_device pdev
;
174 #define DEFAULT_COW { \
178 .bitmap_offset = 0, \
182 #define DEFAULT_UBD { \
187 .boot_openflags = OPEN_FLAGS, \
188 .openflags = OPEN_FLAGS, \
190 .cow = DEFAULT_COW, \
193 struct ubd ubd_dev
[MAX_DEV
] = { [ 0 ... MAX_DEV
- 1 ] = DEFAULT_UBD
};
195 static int ubd0_init(void)
197 struct ubd
*dev
= &ubd_dev
[0];
199 if(dev
->file
== NULL
)
200 dev
->file
= "root_fs";
204 __initcall(ubd0_init
);
206 /* Only changed by fake_ide_setup which is a setup */
207 static int fake_ide
= 0;
208 static struct proc_dir_entry
*proc_ide_root
= NULL
;
209 static struct proc_dir_entry
*proc_ide
= NULL
;
211 static void make_proc_ide(void)
213 proc_ide_root
= proc_mkdir("ide", NULL
);
214 proc_ide
= proc_mkdir("ide0", proc_ide_root
);
217 static int proc_ide_read_media(char *page
, char **start
, off_t off
, int count
,
218 int *eof
, void *data
)
222 strcpy(page
, "disk\n");
223 len
= strlen("disk\n");
227 if (len
<= 0) return 0;
234 static void make_ide_entries(char *dev_name
)
236 struct proc_dir_entry
*dir
, *ent
;
239 if(proc_ide_root
== NULL
) make_proc_ide();
241 dir
= proc_mkdir(dev_name
, proc_ide
);
244 ent
= create_proc_entry("media", S_IFREG
|S_IRUGO
, dir
);
248 ent
->read_proc
= proc_ide_read_media
;
249 ent
->write_proc
= NULL
;
250 sprintf(name
,"ide0/%s", dev_name
);
251 proc_symlink(dev_name
, proc_ide_root
, name
);
254 static int fake_ide_setup(char *str
)
260 __setup("fake_ide", fake_ide_setup
);
262 __uml_help(fake_ide_setup
,
264 " Create ide0 entries that map onto ubd devices.\n\n"
267 static int parse_unit(char **ptr
)
269 char *str
= *ptr
, *end
;
273 n
= simple_strtoul(str
, &end
, 0);
278 else if (('a' <= *str
) && (*str
<= 'h')) {
286 static int ubd_setup_common(char *str
, int *index_out
)
289 struct openflags flags
= global_openflags
;
293 if(index_out
) *index_out
= -1;
300 if(!strcmp(str
, "sync")){
301 global_openflags
= of_sync(global_openflags
);
304 major
= simple_strtoul(str
, &end
, 0);
305 if((*end
!= '\0') || (end
== str
)){
307 "ubd_setup : didn't parse major number\n");
312 spin_lock(&ubd_lock
);
313 if(fake_major
!= MAJOR_NR
){
314 printk(KERN_ERR
"Can't assign a fake major twice\n");
320 printk(KERN_INFO
"Setting extra ubd major number to %d\n",
324 spin_unlock(&ubd_lock
);
328 n
= parse_unit(&str
);
330 printk(KERN_ERR
"ubd_setup : couldn't parse unit number "
335 printk(KERN_ERR
"ubd_setup : index %d out of range "
336 "(%d devices, from 0 to %d)\n", n
, MAX_DEV
, MAX_DEV
- 1);
341 spin_lock(&ubd_lock
);
344 if(dev
->file
!= NULL
){
345 printk(KERN_ERR
"ubd_setup : device already configured\n");
352 for (i
= 0; i
< 4; i
++) {
367 printk(KERN_ERR
"ubd_setup : Expected '=' or flag letter (r,s or d)\n");
374 printk(KERN_ERR
"ubd_setup : Too many flags specified\n");
376 printk(KERN_ERR
"ubd_setup : Expected '='\n");
381 backing_file
= strchr(str
, ',');
384 backing_file
= strchr(str
, ':');
389 printk(KERN_ERR
"Can't specify both 'd' and a "
392 *backing_file
= '\0';
397 dev
->cow
.file
= backing_file
;
398 dev
->boot_openflags
= flags
;
400 spin_unlock(&ubd_lock
);
404 static int ubd_setup(char *str
)
406 ubd_setup_common(str
, NULL
);
410 __setup("ubd", ubd_setup
);
411 __uml_help(ubd_setup
,
412 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
413 " This is used to associate a device with a file in the underlying\n"
414 " filesystem. When specifying two filenames, the first one is the\n"
415 " COW name and the second is the backing file name. As separator you can\n"
416 " use either a ':' or a ',': the first one allows writing things like;\n"
417 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
418 " while with a ',' the shell would not expand the 2nd '~'.\n"
419 " When using only one filename, UML will detect whether to thread it like\n"
420 " a COW file or a backing file. To override this detection, add the 'd'\n"
422 " ubd0d=BackingFile\n"
423 " Usually, there is a filesystem in the file, but \n"
424 " that's not required. Swap devices containing swap files can be\n"
425 " specified like this. Also, a file which doesn't contain a\n"
426 " filesystem can have its contents read in the virtual \n"
427 " machine by running 'dd' on the device. <n> must be in the range\n"
428 " 0 to 7. Appending an 'r' to the number will cause that device\n"
429 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
430 " an 's' will cause data to be written to disk on the host immediately.\n\n"
433 static int udb_setup(char *str
)
435 printk("udb%s specified on command line is almost certainly a ubd -> "
440 __setup("udb", udb_setup
);
441 __uml_help(udb_setup
,
443 " This option is here solely to catch ubd -> udb typos, which can be\n"
444 " to impossible to catch visually unless you specifically look for\n"
445 " them. The only result of any option starting with 'udb' is an error\n"
446 " in the boot output.\n\n"
449 static int fakehd_set
= 0;
450 static int fakehd(char *str
)
452 printk(KERN_INFO
"fakehd : Changing ubd name to \"hd\".\n");
457 __setup("fakehd", fakehd
);
460 " Change the ubd device name to \"hd\".\n\n"
463 static void do_ubd_request(request_queue_t
* q
);
465 /* Only changed by ubd_init, which is an initcall. */
468 /* Changed by ubd_handler, which is serialized because interrupts only
473 /* call ubd_finish if you need to serialize */
474 static void __ubd_finish(struct request
*req
, int error
)
482 nsect
= req
->current_nr_sectors
;
483 req
->sector
+= nsect
;
484 req
->buffer
+= nsect
<< 9;
486 req
->nr_sectors
-= nsect
;
487 req
->current_nr_sectors
= 0;
491 static inline void ubd_finish(struct request
*req
, int error
)
493 spin_lock(&ubd_io_lock
);
494 __ubd_finish(req
, error
);
495 spin_unlock(&ubd_io_lock
);
498 /* Called without ubd_io_lock held */
499 static void ubd_handler(void)
501 struct io_thread_req req
;
502 struct request
*rq
= elv_next_request(ubd_queue
);
507 n
= os_read_file(thread_fd
, &req
, sizeof(req
));
508 if(n
!= sizeof(req
)){
509 printk(KERN_ERR
"Pid %d - spurious interrupt in ubd_handler, "
510 "err = %d\n", os_getpid(), -n
);
511 spin_lock(&ubd_io_lock
);
513 spin_unlock(&ubd_io_lock
);
517 ubd_finish(rq
, req
.error
);
518 reactivate_fd(thread_fd
, UBD_IRQ
);
519 do_ubd_request(ubd_queue
);
522 static irqreturn_t
ubd_intr(int irq
, void *dev
, struct pt_regs
*unused
)
528 /* Only changed by ubd_init, which is an initcall. */
529 static int io_pid
= -1;
531 void kill_io_thread(void)
534 os_kill_process(io_pid
, 1);
537 __uml_exitcall(kill_io_thread
);
539 static int ubd_file_size(struct ubd
*dev
, __u64
*size_out
)
543 file
= dev
->cow
.file
? dev
->cow
.file
: dev
->file
;
544 return(os_file_size(file
, size_out
));
547 static void ubd_close(struct ubd
*dev
)
549 os_close_file(dev
->fd
);
550 if(dev
->cow
.file
== NULL
)
553 os_close_file(dev
->cow
.fd
);
554 vfree(dev
->cow
.bitmap
);
555 dev
->cow
.bitmap
= NULL
;
558 static int ubd_open_dev(struct ubd
*dev
)
560 struct openflags flags
;
562 int err
, create_cow
, *create_ptr
;
564 dev
->openflags
= dev
->boot_openflags
;
566 create_ptr
= (dev
->cow
.file
!= NULL
) ? &create_cow
: NULL
;
567 back_ptr
= dev
->no_cow
? NULL
: &dev
->cow
.file
;
568 dev
->fd
= open_ubd_file(dev
->file
, &dev
->openflags
, back_ptr
,
569 &dev
->cow
.bitmap_offset
, &dev
->cow
.bitmap_len
,
570 &dev
->cow
.data_offset
, create_ptr
);
572 if((dev
->fd
== -ENOENT
) && create_cow
){
573 dev
->fd
= create_cow_file(dev
->file
, dev
->cow
.file
,
574 dev
->openflags
, 1 << 9, PAGE_SIZE
,
575 &dev
->cow
.bitmap_offset
,
576 &dev
->cow
.bitmap_len
,
577 &dev
->cow
.data_offset
);
579 printk(KERN_INFO
"Creating \"%s\" as COW file for "
580 "\"%s\"\n", dev
->file
, dev
->cow
.file
);
585 printk("Failed to open '%s', errno = %d\n", dev
->file
,
590 if(dev
->cow
.file
!= NULL
){
592 dev
->cow
.bitmap
= (void *) vmalloc(dev
->cow
.bitmap_len
);
593 if(dev
->cow
.bitmap
== NULL
){
594 printk(KERN_ERR
"Failed to vmalloc COW bitmap\n");
597 flush_tlb_kernel_vm();
599 err
= read_cow_bitmap(dev
->fd
, dev
->cow
.bitmap
,
600 dev
->cow
.bitmap_offset
,
601 dev
->cow
.bitmap_len
);
605 flags
= dev
->openflags
;
607 err
= open_ubd_file(dev
->cow
.file
, &flags
, NULL
, NULL
, NULL
,
609 if(err
< 0) goto error
;
614 os_close_file(dev
->fd
);
618 static int ubd_new_disk(int major
, u64 size
, int unit
,
619 struct gendisk
**disk_out
)
622 struct gendisk
*disk
;
623 char from
[sizeof("ubd/nnnnn\0")], to
[sizeof("discnnnnn/disc\0")];
626 disk
= alloc_disk(1 << UBD_SHIFT
);
631 disk
->first_minor
= unit
<< UBD_SHIFT
;
632 disk
->fops
= &ubd_blops
;
633 set_capacity(disk
, size
/ 512);
634 if(major
== MAJOR_NR
){
635 sprintf(disk
->disk_name
, "ubd%c", 'a' + unit
);
636 sprintf(disk
->devfs_name
, "ubd/disc%d", unit
);
637 sprintf(from
, "ubd/%d", unit
);
638 sprintf(to
, "disc%d/disc", unit
);
639 err
= devfs_mk_symlink(from
, to
);
641 printk("ubd_new_disk failed to make link from %s to "
642 "%s, error = %d\n", from
, to
, err
);
645 sprintf(disk
->disk_name
, "ubd_fake%d", unit
);
646 sprintf(disk
->devfs_name
, "ubd_fake/disc%d", unit
);
649 /* sysfs register (not for ide fake devices) */
650 if (major
== MAJOR_NR
) {
651 ubd_dev
[unit
].pdev
.id
= unit
;
652 ubd_dev
[unit
].pdev
.name
= DRIVER_NAME
;
653 platform_device_register(&ubd_dev
[unit
].pdev
);
654 disk
->driverfs_dev
= &ubd_dev
[unit
].pdev
.dev
;
657 disk
->private_data
= &ubd_dev
[unit
];
658 disk
->queue
= ubd_queue
;
665 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
667 static int ubd_add(int n
)
669 struct ubd
*dev
= &ubd_dev
[n
];
673 if(dev
->file
== NULL
)
676 if (ubd_open_dev(dev
))
679 err
= ubd_file_size(dev
, &dev
->size
);
683 dev
->size
= ROUND_BLOCK(dev
->size
);
685 err
= ubd_new_disk(MAJOR_NR
, dev
->size
, n
, &ubd_gendisk
[n
]);
689 if(fake_major
!= MAJOR_NR
)
690 ubd_new_disk(fake_major
, dev
->size
, n
,
693 /* perhaps this should also be under the "if (fake_major)" above */
694 /* using the fake_disk->disk_name and also the fakehd_set name */
696 make_ide_entries(ubd_gendisk
[n
]->disk_name
);
705 static int ubd_config(char *str
)
709 str
= kstrdup(str
, GFP_KERNEL
);
711 printk(KERN_ERR
"ubd_config failed to strdup string\n");
714 err
= ubd_setup_common(str
, &n
);
719 if(n
== -1) return(0);
721 spin_lock(&ubd_lock
);
724 ubd_dev
[n
].file
= NULL
;
725 spin_unlock(&ubd_lock
);
730 static int ubd_get_config(char *name
, char *str
, int size
, char **error_out
)
735 n
= parse_unit(&name
);
736 if((n
>= MAX_DEV
) || (n
< 0)){
737 *error_out
= "ubd_get_config : device number out of range";
742 spin_lock(&ubd_lock
);
744 if(dev
->file
== NULL
){
745 CONFIG_CHUNK(str
, size
, len
, "", 1);
749 CONFIG_CHUNK(str
, size
, len
, dev
->file
, 0);
751 if(dev
->cow
.file
!= NULL
){
752 CONFIG_CHUNK(str
, size
, len
, ",", 0);
753 CONFIG_CHUNK(str
, size
, len
, dev
->cow
.file
, 1);
755 else CONFIG_CHUNK(str
, size
, len
, "", 1);
758 spin_unlock(&ubd_lock
);
762 static int ubd_id(char **str
, int *start_out
, int *end_out
)
768 *end_out
= MAX_DEV
- 1;
772 static int ubd_remove(int n
)
777 spin_lock(&ubd_lock
);
779 if(ubd_gendisk
[n
] == NULL
)
784 if(dev
->file
== NULL
)
787 /* you cannot remove a open disk */
792 del_gendisk(ubd_gendisk
[n
]);
793 put_disk(ubd_gendisk
[n
]);
794 ubd_gendisk
[n
] = NULL
;
796 if(fake_gendisk
[n
] != NULL
){
797 del_gendisk(fake_gendisk
[n
]);
798 put_disk(fake_gendisk
[n
]);
799 fake_gendisk
[n
] = NULL
;
802 platform_device_unregister(&dev
->pdev
);
803 *dev
= ((struct ubd
) DEFAULT_UBD
);
806 spin_unlock(&ubd_lock
);
810 static struct mc_device ubd_mc
= {
812 .config
= ubd_config
,
813 .get_config
= ubd_get_config
,
815 .remove
= ubd_remove
,
818 static int ubd_mc_init(void)
820 mconsole_register_dev(&ubd_mc
);
824 __initcall(ubd_mc_init
);
826 static struct platform_driver ubd_driver
= {
837 if (register_blkdev(MAJOR_NR
, "ubd"))
840 ubd_queue
= blk_init_queue(do_ubd_request
, &ubd_io_lock
);
842 unregister_blkdev(MAJOR_NR
, "ubd");
846 if (fake_major
!= MAJOR_NR
) {
847 char name
[sizeof("ubd_nnn\0")];
849 snprintf(name
, sizeof(name
), "ubd_%d", fake_major
);
851 if (register_blkdev(fake_major
, "ubd"))
854 platform_driver_register(&ubd_driver
);
855 for (i
= 0; i
< MAX_DEV
; i
++)
860 late_initcall(ubd_init
);
862 int ubd_driver_init(void){
866 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
867 if(global_openflags
.s
){
868 printk(KERN_INFO
"ubd: Synchronous mode\n");
869 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
870 * enough. So use anyway the io thread. */
872 stack
= alloc_stack(0, 0);
873 io_pid
= start_io_thread(stack
+ PAGE_SIZE
- sizeof(void *),
877 "ubd : Failed to start I/O thread (errno = %d) - "
878 "falling back to synchronous I/O\n", -io_pid
);
882 err
= um_request_irq(UBD_IRQ
, thread_fd
, IRQ_READ
, ubd_intr
,
883 SA_INTERRUPT
, "ubd", ubd_dev
);
885 printk(KERN_ERR
"um_request_irq failed - errno = %d\n", -err
);
889 device_initcall(ubd_driver_init
);
891 static int ubd_open(struct inode
*inode
, struct file
*filp
)
893 struct gendisk
*disk
= inode
->i_bdev
->bd_disk
;
894 struct ubd
*dev
= disk
->private_data
;
898 err
= ubd_open_dev(dev
);
900 printk(KERN_ERR
"%s: Can't open \"%s\": errno = %d\n",
901 disk
->disk_name
, dev
->file
, -err
);
906 set_disk_ro(disk
, !dev
->openflags
.w
);
908 /* This should no more be needed. And it didn't work anyway to exclude
909 * read-write remounting of filesystems.*/
910 /*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
911 if(--dev->count == 0) ubd_close(dev);
918 static int ubd_release(struct inode
* inode
, struct file
* file
)
920 struct gendisk
*disk
= inode
->i_bdev
->bd_disk
;
921 struct ubd
*dev
= disk
->private_data
;
923 if(--dev
->count
== 0)
928 static void cowify_bitmap(__u64 io_offset
, int length
, unsigned long *cow_mask
,
929 __u64
*cow_offset
, unsigned long *bitmap
,
930 __u64 bitmap_offset
, unsigned long *bitmap_words
,
933 __u64 sector
= io_offset
>> 9;
934 int i
, update_bitmap
= 0;
936 for(i
= 0; i
< length
>> 9; i
++){
938 ubd_set_bit(i
, (unsigned char *) cow_mask
);
939 if(ubd_test_bit(sector
+ i
, (unsigned char *) bitmap
))
943 ubd_set_bit(sector
+ i
, (unsigned char *) bitmap
);
949 *cow_offset
= sector
/ (sizeof(unsigned long) * 8);
951 /* This takes care of the case where we're exactly at the end of the
952 * device, and *cow_offset + 1 is off the end. So, just back it up
953 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
954 * for the original diagnosis.
956 if(*cow_offset
== ((bitmap_len
+ sizeof(unsigned long) - 1) /
957 sizeof(unsigned long) - 1))
960 bitmap_words
[0] = bitmap
[*cow_offset
];
961 bitmap_words
[1] = bitmap
[*cow_offset
+ 1];
963 *cow_offset
*= sizeof(unsigned long);
964 *cow_offset
+= bitmap_offset
;
967 static void cowify_req(struct io_thread_req
*req
, unsigned long *bitmap
,
968 __u64 bitmap_offset
, __u64 bitmap_len
)
970 __u64 sector
= req
->offset
>> 9;
973 if(req
->length
> (sizeof(req
->sector_mask
) * 8) << 9)
974 panic("Operation too long");
976 if(req
->op
== UBD_READ
) {
977 for(i
= 0; i
< req
->length
>> 9; i
++){
978 if(ubd_test_bit(sector
+ i
, (unsigned char *) bitmap
))
979 ubd_set_bit(i
, (unsigned char *)
983 else cowify_bitmap(req
->offset
, req
->length
, &req
->sector_mask
,
984 &req
->cow_offset
, bitmap
, bitmap_offset
,
985 req
->bitmap_words
, bitmap_len
);
988 /* Called with ubd_io_lock held */
989 static int prepare_request(struct request
*req
, struct io_thread_req
*io_req
)
991 struct gendisk
*disk
= req
->rq_disk
;
992 struct ubd
*dev
= disk
->private_data
;
996 if(req
->rq_status
== RQ_INACTIVE
) return(1);
998 /* This should be impossible now */
999 if((rq_data_dir(req
) == WRITE
) && !dev
->openflags
.w
){
1000 printk("Write attempted on readonly ubd device %s\n",
1002 end_request(req
, 0);
1006 offset
= ((__u64
) req
->sector
) << 9;
1007 len
= req
->current_nr_sectors
<< 9;
1009 io_req
->fds
[0] = (dev
->cow
.file
!= NULL
) ? dev
->cow
.fd
: dev
->fd
;
1010 io_req
->fds
[1] = dev
->fd
;
1011 io_req
->cow_offset
= -1;
1012 io_req
->offset
= offset
;
1013 io_req
->length
= len
;
1015 io_req
->sector_mask
= 0;
1017 io_req
->op
= (rq_data_dir(req
) == READ
) ? UBD_READ
: UBD_WRITE
;
1018 io_req
->offsets
[0] = 0;
1019 io_req
->offsets
[1] = dev
->cow
.data_offset
;
1020 io_req
->buffer
= req
->buffer
;
1021 io_req
->sectorsize
= 1 << 9;
1023 if(dev
->cow
.file
!= NULL
)
1024 cowify_req(io_req
, dev
->cow
.bitmap
, dev
->cow
.bitmap_offset
,
1025 dev
->cow
.bitmap_len
);
1030 /* Called with ubd_io_lock held */
1031 static void do_ubd_request(request_queue_t
*q
)
1033 struct io_thread_req io_req
;
1034 struct request
*req
;
1037 if(thread_fd
== -1){
1038 while((req
= elv_next_request(q
)) != NULL
){
1039 err
= prepare_request(req
, &io_req
);
1042 __ubd_finish(req
, io_req
.error
);
1047 if(do_ubd
|| (req
= elv_next_request(q
)) == NULL
)
1049 err
= prepare_request(req
, &io_req
);
1051 do_ubd
= ubd_handler
;
1052 n
= os_write_file(thread_fd
, (char *) &io_req
,
1054 if(n
!= sizeof(io_req
))
1055 printk("write to io thread failed, "
1056 "errno = %d\n", -n
);
1061 static int ubd_ioctl(struct inode
* inode
, struct file
* file
,
1062 unsigned int cmd
, unsigned long arg
)
1064 struct hd_geometry __user
*loc
= (struct hd_geometry __user
*) arg
;
1065 struct ubd
*dev
= inode
->i_bdev
->bd_disk
->private_data
;
1066 struct hd_driveid ubd_id
= {
1073 struct hd_geometry g
;
1074 struct cdrom_volctrl volume
;
1076 if(!loc
) return(-EINVAL
);
1079 g
.cylinders
= dev
->size
/ (128 * 32 * 512);
1080 g
.start
= get_start_sect(inode
->i_bdev
);
1081 return(copy_to_user(loc
, &g
, sizeof(g
)) ? -EFAULT
: 0);
1083 case HDIO_GET_IDENTITY
:
1084 ubd_id
.cyls
= dev
->size
/ (128 * 32 * 512);
1085 if(copy_to_user((char __user
*) arg
, (char *) &ubd_id
,
1091 if(copy_from_user(&volume
, (char __user
*) arg
, sizeof(volume
)))
1093 volume
.channel0
= 255;
1094 volume
.channel1
= 255;
1095 volume
.channel2
= 255;
1096 volume
.channel3
= 255;
1097 if(copy_to_user((char __user
*) arg
, &volume
, sizeof(volume
)))
1104 static int same_backing_files(char *from_cmdline
, char *from_cow
, char *cow
)
1106 struct uml_stat buf1
, buf2
;
1109 if(from_cmdline
== NULL
) return(1);
1110 if(!strcmp(from_cmdline
, from_cow
)) return(1);
1112 err
= os_stat_file(from_cmdline
, &buf1
);
1114 printk("Couldn't stat '%s', err = %d\n", from_cmdline
, -err
);
1117 err
= os_stat_file(from_cow
, &buf2
);
1119 printk("Couldn't stat '%s', err = %d\n", from_cow
, -err
);
1122 if((buf1
.ust_dev
== buf2
.ust_dev
) && (buf1
.ust_ino
== buf2
.ust_ino
))
1125 printk("Backing file mismatch - \"%s\" requested,\n"
1126 "\"%s\" specified in COW header of \"%s\"\n",
1127 from_cmdline
, from_cow
, cow
);
1131 static int backing_file_mismatch(char *file
, __u64 size
, time_t mtime
)
1133 unsigned long modtime
;
1137 err
= os_file_modtime(file
, &modtime
);
1139 printk("Failed to get modification time of backing file "
1140 "\"%s\", err = %d\n", file
, -err
);
1144 err
= os_file_size(file
, &actual
);
1146 printk("Failed to get size of backing file \"%s\", "
1147 "err = %d\n", file
, -err
);
1152 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1154 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1155 "file\n", (unsigned long long) size
, actual
);
1158 if(modtime
!= mtime
){
1159 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1160 "file\n", mtime
, modtime
);
1166 int read_cow_bitmap(int fd
, void *buf
, int offset
, int len
)
1170 err
= os_seek_file(fd
, offset
);
1174 err
= os_read_file(fd
, buf
, len
);
1181 int open_ubd_file(char *file
, struct openflags
*openflags
,
1182 char **backing_file_out
, int *bitmap_offset_out
,
1183 unsigned long *bitmap_len_out
, int *data_offset_out
,
1184 int *create_cow_out
)
1187 unsigned long long size
;
1188 __u32 version
, align
;
1190 int fd
, err
, sectorsize
, same
, mode
= 0644;
1192 fd
= os_open_file(file
, *openflags
, mode
);
1194 if((fd
== -ENOENT
) && (create_cow_out
!= NULL
))
1195 *create_cow_out
= 1;
1197 ((fd
!= -EROFS
) && (fd
!= -EACCES
))) return(fd
);
1199 fd
= os_open_file(file
, *openflags
, mode
);
1204 err
= os_lock_file(fd
, openflags
->w
);
1206 printk("Failed to lock '%s', err = %d\n", file
, -err
);
1210 if(backing_file_out
== NULL
) return(fd
);
1212 err
= read_cow_header(file_reader
, &fd
, &version
, &backing_file
, &mtime
,
1213 &size
, §orsize
, &align
, bitmap_offset_out
);
1214 if(err
&& (*backing_file_out
!= NULL
)){
1215 printk("Failed to read COW header from COW file \"%s\", "
1216 "errno = %d\n", file
, -err
);
1221 if(backing_file_out
== NULL
) return(fd
);
1223 same
= same_backing_files(*backing_file_out
, backing_file
, file
);
1225 if(!same
&& !backing_file_mismatch(*backing_file_out
, size
, mtime
)){
1226 printk("Switching backing file to '%s'\n", *backing_file_out
);
1227 err
= write_cow_header(file
, fd
, *backing_file_out
,
1228 sectorsize
, align
, &size
);
1230 printk("Switch failed, errno = %d\n", -err
);
1235 *backing_file_out
= backing_file
;
1236 err
= backing_file_mismatch(*backing_file_out
, size
, mtime
);
1237 if(err
) goto out_close
;
1240 cow_sizes(version
, size
, sectorsize
, align
, *bitmap_offset_out
,
1241 bitmap_len_out
, data_offset_out
);
1249 int create_cow_file(char *cow_file
, char *backing_file
, struct openflags flags
,
1250 int sectorsize
, int alignment
, int *bitmap_offset_out
,
1251 unsigned long *bitmap_len_out
, int *data_offset_out
)
1256 fd
= open_ubd_file(cow_file
, &flags
, NULL
, NULL
, NULL
, NULL
, NULL
);
1259 printk("Open of COW file '%s' failed, errno = %d\n", cow_file
,
1264 err
= init_cow_file(fd
, cow_file
, backing_file
, sectorsize
, alignment
,
1265 bitmap_offset_out
, bitmap_len_out
,
1274 static int update_bitmap(struct io_thread_req
*req
)
1278 if(req
->cow_offset
== -1)
1281 n
= os_seek_file(req
->fds
[1], req
->cow_offset
);
1283 printk("do_io - bitmap lseek failed : err = %d\n", -n
);
1287 n
= os_write_file(req
->fds
[1], &req
->bitmap_words
,
1288 sizeof(req
->bitmap_words
));
1289 if(n
!= sizeof(req
->bitmap_words
)){
1290 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n
,
1298 void do_io(struct io_thread_req
*req
)
1302 int n
, nsectors
, start
, end
, bit
;
1306 nsectors
= req
->length
/ req
->sectorsize
;
1309 bit
= ubd_test_bit(start
, (unsigned char *) &req
->sector_mask
);
1311 while((end
< nsectors
) &&
1312 (ubd_test_bit(end
, (unsigned char *)
1313 &req
->sector_mask
) == bit
))
1316 off
= req
->offset
+ req
->offsets
[bit
] +
1317 start
* req
->sectorsize
;
1318 len
= (end
- start
) * req
->sectorsize
;
1319 buf
= &req
->buffer
[start
* req
->sectorsize
];
1321 err
= os_seek_file(req
->fds
[bit
], off
);
1323 printk("do_io - lseek failed : err = %d\n", -err
);
1327 if(req
->op
== UBD_READ
){
1332 n
= os_read_file(req
->fds
[bit
], buf
, len
);
1334 printk("do_io - read failed, err = %d "
1335 "fd = %d\n", -n
, req
->fds
[bit
]);
1339 } while((n
< len
) && (n
!= 0));
1340 if (n
< len
) memset(&buf
[n
], 0, len
- n
);
1342 n
= os_write_file(req
->fds
[bit
], buf
, len
);
1344 printk("do_io - write failed err = %d "
1345 "fd = %d\n", -n
, req
->fds
[bit
]);
1352 } while(start
< nsectors
);
1354 req
->error
= update_bitmap(req
);
1357 /* Changed in start_io_thread, which is serialized by being called only
1358 * from ubd_init, which is an initcall.
1362 /* Only changed by the io thread */
1365 int io_thread(void *arg
)
1367 struct io_thread_req req
;
1370 ignore_sigwinch_sig();
1372 n
= os_read_file(kernel_fd
, &req
, sizeof(req
));
1373 if(n
!= sizeof(req
)){
1375 printk("io_thread - read failed, fd = %d, "
1376 "err = %d\n", kernel_fd
, -n
);
1378 printk("io_thread - short read, fd = %d, "
1379 "length = %d\n", kernel_fd
, n
);
1385 n
= os_write_file(kernel_fd
, &req
, sizeof(req
));
1386 if(n
!= sizeof(req
))
1387 printk("io_thread - write failed, fd = %d, err = %d\n",
1393 * Overrides for Emacs so that we follow Linus's tabbing style.
1394 * Emacs will notice this stuff at the end of the file and automatically
1395 * adjust the settings for this buffer only. This must remain at the end
1397 * ---------------------------------------------------------------------------
1399 * c-file-style: "linux"