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.
22 #include "linux/kernel.h"
23 #include "linux/module.h"
24 #include "linux/blkdev.h"
25 #include "linux/ata.h"
26 #include "linux/hdreg.h"
27 #include "linux/init.h"
28 #include "linux/cdrom.h"
29 #include "linux/proc_fs.h"
30 #include "linux/seq_file.h"
31 #include "linux/ctype.h"
32 #include "linux/capability.h"
34 #include "linux/slab.h"
35 #include "linux/vmalloc.h"
36 #include "linux/smp_lock.h"
37 #include "linux/blkpg.h"
38 #include "linux/genhd.h"
39 #include "linux/spinlock.h"
40 #include "linux/platform_device.h"
41 #include "linux/scatterlist.h"
42 #include "asm/segment.h"
43 #include "asm/uaccess.h"
45 #include "asm/types.h"
46 #include "asm/tlbflush.h"
48 #include "kern_util.h"
50 #include "mconsole_kern.h"
60 enum ubd_req
{ UBD_READ
, UBD_WRITE
};
62 struct io_thread_req
{
66 unsigned long offsets
[2];
67 unsigned long long offset
;
71 unsigned long sector_mask
;
72 unsigned long long cow_offset
;
73 unsigned long bitmap_words
[2];
77 static inline int ubd_test_bit(__u64 bit
, unsigned char *data
)
82 bits
= sizeof(data
[0]) * 8;
85 return (data
[n
] & (1 << off
)) != 0;
88 static inline void ubd_set_bit(__u64 bit
, unsigned char *data
)
93 bits
= sizeof(data
[0]) * 8;
96 data
[n
] |= (1 << off
);
98 /*End stuff from ubd_user.h*/
100 #define DRIVER_NAME "uml-blkdev"
102 static DEFINE_MUTEX(ubd_lock
);
104 static int ubd_open(struct block_device
*bdev
, fmode_t mode
);
105 static int ubd_release(struct gendisk
*disk
, fmode_t mode
);
106 static int ubd_ioctl(struct block_device
*bdev
, fmode_t mode
,
107 unsigned int cmd
, unsigned long arg
);
108 static int ubd_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
);
112 static const struct block_device_operations ubd_blops
= {
113 .owner
= THIS_MODULE
,
115 .release
= ubd_release
,
117 .getgeo
= ubd_getgeo
,
120 /* Protected by ubd_lock */
121 static int fake_major
= UBD_MAJOR
;
122 static struct gendisk
*ubd_gendisk
[MAX_DEV
];
123 static struct gendisk
*fake_gendisk
[MAX_DEV
];
125 #ifdef CONFIG_BLK_DEV_UBD_SYNC
126 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
129 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
132 static struct openflags global_openflags
= OPEN_FLAGS
;
135 /* backing file name */
137 /* backing file fd */
139 unsigned long *bitmap
;
140 unsigned long bitmap_len
;
148 struct list_head restart
;
149 /* name (and fd, below) of the file opened for writing, either the
150 * backing or the cow file. */
155 struct openflags boot_openflags
;
156 struct openflags openflags
;
160 struct platform_device pdev
;
161 struct request_queue
*queue
;
163 struct scatterlist sg
[MAX_SG
];
164 struct request
*request
;
165 int start_sg
, end_sg
;
168 #define DEFAULT_COW { \
172 .bitmap_offset = 0, \
176 #define DEFAULT_UBD { \
181 .boot_openflags = OPEN_FLAGS, \
182 .openflags = OPEN_FLAGS, \
185 .cow = DEFAULT_COW, \
186 .lock = SPIN_LOCK_UNLOCKED, \
192 /* Protected by ubd_lock */
193 static struct ubd ubd_devs
[MAX_DEV
] = { [0 ... MAX_DEV
- 1] = DEFAULT_UBD
};
195 /* Only changed by fake_ide_setup which is a setup */
196 static int fake_ide
= 0;
197 static struct proc_dir_entry
*proc_ide_root
= NULL
;
198 static struct proc_dir_entry
*proc_ide
= NULL
;
200 static void make_proc_ide(void)
202 proc_ide_root
= proc_mkdir("ide", NULL
);
203 proc_ide
= proc_mkdir("ide0", proc_ide_root
);
206 static int fake_ide_media_proc_show(struct seq_file
*m
, void *v
)
208 seq_puts(m
, "disk\n");
212 static int fake_ide_media_proc_open(struct inode
*inode
, struct file
*file
)
214 return single_open(file
, fake_ide_media_proc_show
, NULL
);
217 static const struct file_operations fake_ide_media_proc_fops
= {
218 .owner
= THIS_MODULE
,
219 .open
= fake_ide_media_proc_open
,
222 .release
= single_release
,
225 static void make_ide_entries(const char *dev_name
)
227 struct proc_dir_entry
*dir
, *ent
;
230 if(proc_ide_root
== NULL
) make_proc_ide();
232 dir
= proc_mkdir(dev_name
, proc_ide
);
235 ent
= proc_create("media", S_IRUGO
, dir
, &fake_ide_media_proc_fops
);
237 snprintf(name
, sizeof(name
), "ide0/%s", dev_name
);
238 proc_symlink(dev_name
, proc_ide_root
, name
);
241 static int fake_ide_setup(char *str
)
247 __setup("fake_ide", fake_ide_setup
);
249 __uml_help(fake_ide_setup
,
251 " Create ide0 entries that map onto ubd devices.\n\n"
254 static int parse_unit(char **ptr
)
256 char *str
= *ptr
, *end
;
260 n
= simple_strtoul(str
, &end
, 0);
265 else if (('a' <= *str
) && (*str
<= 'z')) {
273 /* If *index_out == -1 at exit, the passed option was a general one;
274 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
275 * should not be freed on exit.
277 static int ubd_setup_common(char *str
, int *index_out
, char **error_out
)
280 struct openflags flags
= global_openflags
;
284 if(index_out
) *index_out
= -1;
291 if(!strcmp(str
, "sync")){
292 global_openflags
= of_sync(global_openflags
);
297 major
= simple_strtoul(str
, &end
, 0);
298 if((*end
!= '\0') || (end
== str
)){
299 *error_out
= "Didn't parse major number";
303 mutex_lock(&ubd_lock
);
304 if (fake_major
!= UBD_MAJOR
) {
305 *error_out
= "Can't assign a fake major twice";
311 printk(KERN_INFO
"Setting extra ubd major number to %d\n",
315 mutex_unlock(&ubd_lock
);
319 n
= parse_unit(&str
);
321 *error_out
= "Couldn't parse device number";
325 *error_out
= "Device number out of range";
330 mutex_lock(&ubd_lock
);
332 ubd_dev
= &ubd_devs
[n
];
333 if(ubd_dev
->file
!= NULL
){
334 *error_out
= "Device is already configured";
342 for (i
= 0; i
< sizeof("rscd="); i
++) {
360 *error_out
= "Expected '=' or flag letter "
368 *error_out
= "Too many flags specified";
370 *error_out
= "Missing '='";
374 backing_file
= strchr(str
, ',');
376 if (backing_file
== NULL
)
377 backing_file
= strchr(str
, ':');
379 if(backing_file
!= NULL
){
381 *error_out
= "Can't specify both 'd' and a cow file";
385 *backing_file
= '\0';
391 ubd_dev
->cow
.file
= backing_file
;
392 ubd_dev
->boot_openflags
= flags
;
394 mutex_unlock(&ubd_lock
);
398 static int ubd_setup(char *str
)
403 err
= ubd_setup_common(str
, NULL
, &error
);
405 printk(KERN_ERR
"Failed to initialize device with \"%s\" : "
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 treat 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"
431 " 'c' will cause the device to be treated as being shared between multiple\n"
432 " UMLs and file locking will be turned off - this is appropriate for a\n"
433 " cluster filesystem and inappropriate at almost all other times.\n\n"
436 static int udb_setup(char *str
)
438 printk("udb%s specified on command line is almost certainly a ubd -> "
443 __setup("udb", udb_setup
);
444 __uml_help(udb_setup
,
446 " This option is here solely to catch ubd -> udb typos, which can be\n"
447 " to impossible to catch visually unless you specifically look for\n"
448 " them. The only result of any option starting with 'udb' is an error\n"
449 " in the boot output.\n\n"
452 static void do_ubd_request(struct request_queue
* q
);
454 /* Only changed by ubd_init, which is an initcall. */
455 static int thread_fd
= -1;
456 static LIST_HEAD(restart
);
458 /* XXX - move this inside ubd_intr. */
459 /* Called without dev->lock held, and only in interrupt context. */
460 static void ubd_handler(void)
462 struct io_thread_req
*req
;
464 struct list_head
*list
, *next_ele
;
469 n
= os_read_file(thread_fd
, &req
,
470 sizeof(struct io_thread_req
*));
471 if(n
!= sizeof(req
)){
474 printk(KERN_ERR
"spurious interrupt in ubd_handler, "
479 blk_end_request(req
->req
, 0, req
->length
);
482 reactivate_fd(thread_fd
, UBD_IRQ
);
484 list_for_each_safe(list
, next_ele
, &restart
){
485 ubd
= container_of(list
, struct ubd
, restart
);
486 list_del_init(&ubd
->restart
);
487 spin_lock_irqsave(&ubd
->lock
, flags
);
488 do_ubd_request(ubd
->queue
);
489 spin_unlock_irqrestore(&ubd
->lock
, flags
);
493 static irqreturn_t
ubd_intr(int irq
, void *dev
)
499 /* Only changed by ubd_init, which is an initcall. */
500 static int io_pid
= -1;
502 static void kill_io_thread(void)
505 os_kill_process(io_pid
, 1);
508 __uml_exitcall(kill_io_thread
);
510 static inline int ubd_file_size(struct ubd
*ubd_dev
, __u64
*size_out
)
514 file
= ubd_dev
->cow
.file
? ubd_dev
->cow
.file
: ubd_dev
->file
;
515 return os_file_size(file
, size_out
);
518 static int read_cow_bitmap(int fd
, void *buf
, int offset
, int len
)
522 err
= os_seek_file(fd
, offset
);
526 err
= os_read_file(fd
, buf
, len
);
533 static int backing_file_mismatch(char *file
, __u64 size
, time_t mtime
)
535 unsigned long modtime
;
536 unsigned long long actual
;
539 err
= os_file_modtime(file
, &modtime
);
541 printk(KERN_ERR
"Failed to get modification time of backing "
542 "file \"%s\", err = %d\n", file
, -err
);
546 err
= os_file_size(file
, &actual
);
548 printk(KERN_ERR
"Failed to get size of backing file \"%s\", "
549 "err = %d\n", file
, -err
);
553 if (actual
!= size
) {
554 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
556 printk(KERN_ERR
"Size mismatch (%llu vs %llu) of COW header "
557 "vs backing file\n", (unsigned long long) size
, actual
);
560 if (modtime
!= mtime
) {
561 printk(KERN_ERR
"mtime mismatch (%ld vs %ld) of COW header vs "
562 "backing file\n", mtime
, modtime
);
568 static int path_requires_switch(char *from_cmdline
, char *from_cow
, char *cow
)
570 struct uml_stat buf1
, buf2
;
573 if (from_cmdline
== NULL
)
575 if (!strcmp(from_cmdline
, from_cow
))
578 err
= os_stat_file(from_cmdline
, &buf1
);
580 printk(KERN_ERR
"Couldn't stat '%s', err = %d\n", from_cmdline
,
584 err
= os_stat_file(from_cow
, &buf2
);
586 printk(KERN_ERR
"Couldn't stat '%s', err = %d\n", from_cow
,
590 if ((buf1
.ust_dev
== buf2
.ust_dev
) && (buf1
.ust_ino
== buf2
.ust_ino
))
593 printk(KERN_ERR
"Backing file mismatch - \"%s\" requested, "
594 "\"%s\" specified in COW header of \"%s\"\n",
595 from_cmdline
, from_cow
, cow
);
599 static int open_ubd_file(char *file
, struct openflags
*openflags
, int shared
,
600 char **backing_file_out
, int *bitmap_offset_out
,
601 unsigned long *bitmap_len_out
, int *data_offset_out
,
605 unsigned long long size
;
606 __u32 version
, align
;
608 int fd
, err
, sectorsize
, asked_switch
, mode
= 0644;
610 fd
= os_open_file(file
, *openflags
, mode
);
612 if ((fd
== -ENOENT
) && (create_cow_out
!= NULL
))
615 ((fd
!= -EROFS
) && (fd
!= -EACCES
)))
618 fd
= os_open_file(file
, *openflags
, mode
);
624 printk(KERN_INFO
"Not locking \"%s\" on the host\n", file
);
626 err
= os_lock_file(fd
, openflags
->w
);
628 printk(KERN_ERR
"Failed to lock '%s', err = %d\n",
634 /* Successful return case! */
635 if (backing_file_out
== NULL
)
638 err
= read_cow_header(file_reader
, &fd
, &version
, &backing_file
, &mtime
,
639 &size
, §orsize
, &align
, bitmap_offset_out
);
640 if (err
&& (*backing_file_out
!= NULL
)) {
641 printk(KERN_ERR
"Failed to read COW header from COW file "
642 "\"%s\", errno = %d\n", file
, -err
);
648 asked_switch
= path_requires_switch(*backing_file_out
, backing_file
,
651 /* Allow switching only if no mismatch. */
652 if (asked_switch
&& !backing_file_mismatch(*backing_file_out
, size
,
654 printk(KERN_ERR
"Switching backing file to '%s'\n",
656 err
= write_cow_header(file
, fd
, *backing_file_out
,
657 sectorsize
, align
, &size
);
659 printk(KERN_ERR
"Switch failed, errno = %d\n", -err
);
663 *backing_file_out
= backing_file
;
664 err
= backing_file_mismatch(*backing_file_out
, size
, mtime
);
669 cow_sizes(version
, size
, sectorsize
, align
, *bitmap_offset_out
,
670 bitmap_len_out
, data_offset_out
);
678 static int create_cow_file(char *cow_file
, char *backing_file
,
679 struct openflags flags
,
680 int sectorsize
, int alignment
, int *bitmap_offset_out
,
681 unsigned long *bitmap_len_out
, int *data_offset_out
)
686 fd
= open_ubd_file(cow_file
, &flags
, 0, NULL
, NULL
, NULL
, NULL
, NULL
);
689 printk(KERN_ERR
"Open of COW file '%s' failed, errno = %d\n",
694 err
= init_cow_file(fd
, cow_file
, backing_file
, sectorsize
, alignment
,
695 bitmap_offset_out
, bitmap_len_out
,
704 static void ubd_close_dev(struct ubd
*ubd_dev
)
706 os_close_file(ubd_dev
->fd
);
707 if(ubd_dev
->cow
.file
== NULL
)
710 os_close_file(ubd_dev
->cow
.fd
);
711 vfree(ubd_dev
->cow
.bitmap
);
712 ubd_dev
->cow
.bitmap
= NULL
;
715 static int ubd_open_dev(struct ubd
*ubd_dev
)
717 struct openflags flags
;
719 int err
, create_cow
, *create_ptr
;
722 ubd_dev
->openflags
= ubd_dev
->boot_openflags
;
724 create_ptr
= (ubd_dev
->cow
.file
!= NULL
) ? &create_cow
: NULL
;
725 back_ptr
= ubd_dev
->no_cow
? NULL
: &ubd_dev
->cow
.file
;
727 fd
= open_ubd_file(ubd_dev
->file
, &ubd_dev
->openflags
, ubd_dev
->shared
,
728 back_ptr
, &ubd_dev
->cow
.bitmap_offset
,
729 &ubd_dev
->cow
.bitmap_len
, &ubd_dev
->cow
.data_offset
,
732 if((fd
== -ENOENT
) && create_cow
){
733 fd
= create_cow_file(ubd_dev
->file
, ubd_dev
->cow
.file
,
734 ubd_dev
->openflags
, 1 << 9, PAGE_SIZE
,
735 &ubd_dev
->cow
.bitmap_offset
,
736 &ubd_dev
->cow
.bitmap_len
,
737 &ubd_dev
->cow
.data_offset
);
739 printk(KERN_INFO
"Creating \"%s\" as COW file for "
740 "\"%s\"\n", ubd_dev
->file
, ubd_dev
->cow
.file
);
745 printk("Failed to open '%s', errno = %d\n", ubd_dev
->file
,
751 if(ubd_dev
->cow
.file
!= NULL
){
752 blk_queue_max_hw_sectors(ubd_dev
->queue
, 8 * sizeof(long));
755 ubd_dev
->cow
.bitmap
= vmalloc(ubd_dev
->cow
.bitmap_len
);
756 if(ubd_dev
->cow
.bitmap
== NULL
){
757 printk(KERN_ERR
"Failed to vmalloc COW bitmap\n");
760 flush_tlb_kernel_vm();
762 err
= read_cow_bitmap(ubd_dev
->fd
, ubd_dev
->cow
.bitmap
,
763 ubd_dev
->cow
.bitmap_offset
,
764 ubd_dev
->cow
.bitmap_len
);
768 flags
= ubd_dev
->openflags
;
770 err
= open_ubd_file(ubd_dev
->cow
.file
, &flags
, ubd_dev
->shared
, NULL
,
771 NULL
, NULL
, NULL
, NULL
);
772 if(err
< 0) goto error
;
773 ubd_dev
->cow
.fd
= err
;
777 os_close_file(ubd_dev
->fd
);
781 static void ubd_device_release(struct device
*dev
)
783 struct ubd
*ubd_dev
= dev_get_drvdata(dev
);
785 blk_cleanup_queue(ubd_dev
->queue
);
786 *ubd_dev
= ((struct ubd
) DEFAULT_UBD
);
789 static int ubd_disk_register(int major
, u64 size
, int unit
,
790 struct gendisk
**disk_out
)
792 struct gendisk
*disk
;
794 disk
= alloc_disk(1 << UBD_SHIFT
);
799 disk
->first_minor
= unit
<< UBD_SHIFT
;
800 disk
->fops
= &ubd_blops
;
801 set_capacity(disk
, size
/ 512);
802 if (major
== UBD_MAJOR
)
803 sprintf(disk
->disk_name
, "ubd%c", 'a' + unit
);
805 sprintf(disk
->disk_name
, "ubd_fake%d", unit
);
807 /* sysfs register (not for ide fake devices) */
808 if (major
== UBD_MAJOR
) {
809 ubd_devs
[unit
].pdev
.id
= unit
;
810 ubd_devs
[unit
].pdev
.name
= DRIVER_NAME
;
811 ubd_devs
[unit
].pdev
.dev
.release
= ubd_device_release
;
812 dev_set_drvdata(&ubd_devs
[unit
].pdev
.dev
, &ubd_devs
[unit
]);
813 platform_device_register(&ubd_devs
[unit
].pdev
);
814 disk
->driverfs_dev
= &ubd_devs
[unit
].pdev
.dev
;
817 disk
->private_data
= &ubd_devs
[unit
];
818 disk
->queue
= ubd_devs
[unit
].queue
;
825 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
827 static int ubd_add(int n
, char **error_out
)
829 struct ubd
*ubd_dev
= &ubd_devs
[n
];
832 if(ubd_dev
->file
== NULL
)
835 err
= ubd_file_size(ubd_dev
, &ubd_dev
->size
);
837 *error_out
= "Couldn't determine size of device's file";
841 ubd_dev
->size
= ROUND_BLOCK(ubd_dev
->size
);
843 INIT_LIST_HEAD(&ubd_dev
->restart
);
844 sg_init_table(ubd_dev
->sg
, MAX_SG
);
847 ubd_dev
->queue
= blk_init_queue(do_ubd_request
, &ubd_dev
->lock
);
848 if (ubd_dev
->queue
== NULL
) {
849 *error_out
= "Failed to initialize device queue";
852 ubd_dev
->queue
->queuedata
= ubd_dev
;
854 blk_queue_max_segments(ubd_dev
->queue
, MAX_SG
);
855 err
= ubd_disk_register(UBD_MAJOR
, ubd_dev
->size
, n
, &ubd_gendisk
[n
]);
857 *error_out
= "Failed to register device";
861 if (fake_major
!= UBD_MAJOR
)
862 ubd_disk_register(fake_major
, ubd_dev
->size
, n
,
866 * Perhaps this should also be under the "if (fake_major)" above
867 * using the fake_disk->disk_name
870 make_ide_entries(ubd_gendisk
[n
]->disk_name
);
877 blk_cleanup_queue(ubd_dev
->queue
);
881 static int ubd_config(char *str
, char **error_out
)
885 /* This string is possibly broken up and stored, so it's only
886 * freed if ubd_setup_common fails, or if only general options
889 str
= kstrdup(str
, GFP_KERNEL
);
891 *error_out
= "Failed to allocate memory";
895 ret
= ubd_setup_common(str
, &n
, error_out
);
904 mutex_lock(&ubd_lock
);
905 ret
= ubd_add(n
, error_out
);
907 ubd_devs
[n
].file
= NULL
;
908 mutex_unlock(&ubd_lock
);
918 static int ubd_get_config(char *name
, char *str
, int size
, char **error_out
)
923 n
= parse_unit(&name
);
924 if((n
>= MAX_DEV
) || (n
< 0)){
925 *error_out
= "ubd_get_config : device number out of range";
929 ubd_dev
= &ubd_devs
[n
];
930 mutex_lock(&ubd_lock
);
932 if(ubd_dev
->file
== NULL
){
933 CONFIG_CHUNK(str
, size
, len
, "", 1);
937 CONFIG_CHUNK(str
, size
, len
, ubd_dev
->file
, 0);
939 if(ubd_dev
->cow
.file
!= NULL
){
940 CONFIG_CHUNK(str
, size
, len
, ",", 0);
941 CONFIG_CHUNK(str
, size
, len
, ubd_dev
->cow
.file
, 1);
943 else CONFIG_CHUNK(str
, size
, len
, "", 1);
946 mutex_unlock(&ubd_lock
);
950 static int ubd_id(char **str
, int *start_out
, int *end_out
)
956 *end_out
= MAX_DEV
- 1;
960 static int ubd_remove(int n
, char **error_out
)
962 struct gendisk
*disk
= ubd_gendisk
[n
];
966 mutex_lock(&ubd_lock
);
968 ubd_dev
= &ubd_devs
[n
];
970 if(ubd_dev
->file
== NULL
)
973 /* you cannot remove a open disk */
975 if(ubd_dev
->count
> 0)
978 ubd_gendisk
[n
] = NULL
;
984 if(fake_gendisk
[n
] != NULL
){
985 del_gendisk(fake_gendisk
[n
]);
986 put_disk(fake_gendisk
[n
]);
987 fake_gendisk
[n
] = NULL
;
991 platform_device_unregister(&ubd_dev
->pdev
);
993 mutex_unlock(&ubd_lock
);
997 /* All these are called by mconsole in process context and without
998 * ubd-specific locks. The structure itself is const except for .list.
1000 static struct mc_device ubd_mc
= {
1001 .list
= LIST_HEAD_INIT(ubd_mc
.list
),
1003 .config
= ubd_config
,
1004 .get_config
= ubd_get_config
,
1006 .remove
= ubd_remove
,
1009 static int __init
ubd_mc_init(void)
1011 mconsole_register_dev(&ubd_mc
);
1015 __initcall(ubd_mc_init
);
1017 static int __init
ubd0_init(void)
1019 struct ubd
*ubd_dev
= &ubd_devs
[0];
1021 mutex_lock(&ubd_lock
);
1022 if(ubd_dev
->file
== NULL
)
1023 ubd_dev
->file
= "root_fs";
1024 mutex_unlock(&ubd_lock
);
1029 __initcall(ubd0_init
);
1031 /* Used in ubd_init, which is an initcall */
1032 static struct platform_driver ubd_driver
= {
1034 .name
= DRIVER_NAME
,
1038 static int __init
ubd_init(void)
1043 if (register_blkdev(UBD_MAJOR
, "ubd"))
1046 if (fake_major
!= UBD_MAJOR
) {
1047 char name
[sizeof("ubd_nnn\0")];
1049 snprintf(name
, sizeof(name
), "ubd_%d", fake_major
);
1050 if (register_blkdev(fake_major
, "ubd"))
1053 platform_driver_register(&ubd_driver
);
1054 mutex_lock(&ubd_lock
);
1055 for (i
= 0; i
< MAX_DEV
; i
++){
1056 err
= ubd_add(i
, &error
);
1058 printk(KERN_ERR
"Failed to initialize ubd device %d :"
1061 mutex_unlock(&ubd_lock
);
1065 late_initcall(ubd_init
);
1067 static int __init
ubd_driver_init(void){
1068 unsigned long stack
;
1071 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1072 if(global_openflags
.s
){
1073 printk(KERN_INFO
"ubd: Synchronous mode\n");
1074 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1075 * enough. So use anyway the io thread. */
1077 stack
= alloc_stack(0, 0);
1078 io_pid
= start_io_thread(stack
+ PAGE_SIZE
- sizeof(void *),
1082 "ubd : Failed to start I/O thread (errno = %d) - "
1083 "falling back to synchronous I/O\n", -io_pid
);
1087 err
= um_request_irq(UBD_IRQ
, thread_fd
, IRQ_READ
, ubd_intr
,
1088 IRQF_DISABLED
, "ubd", ubd_devs
);
1090 printk(KERN_ERR
"um_request_irq failed - errno = %d\n", -err
);
1094 device_initcall(ubd_driver_init
);
1096 static int ubd_open(struct block_device
*bdev
, fmode_t mode
)
1098 struct gendisk
*disk
= bdev
->bd_disk
;
1099 struct ubd
*ubd_dev
= disk
->private_data
;
1103 if(ubd_dev
->count
== 0){
1104 err
= ubd_open_dev(ubd_dev
);
1106 printk(KERN_ERR
"%s: Can't open \"%s\": errno = %d\n",
1107 disk
->disk_name
, ubd_dev
->file
, -err
);
1112 set_disk_ro(disk
, !ubd_dev
->openflags
.w
);
1114 /* This should no more be needed. And it didn't work anyway to exclude
1115 * read-write remounting of filesystems.*/
1116 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1117 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1125 static int ubd_release(struct gendisk
*disk
, fmode_t mode
)
1127 struct ubd
*ubd_dev
= disk
->private_data
;
1130 if(--ubd_dev
->count
== 0)
1131 ubd_close_dev(ubd_dev
);
1136 static void cowify_bitmap(__u64 io_offset
, int length
, unsigned long *cow_mask
,
1137 __u64
*cow_offset
, unsigned long *bitmap
,
1138 __u64 bitmap_offset
, unsigned long *bitmap_words
,
1141 __u64 sector
= io_offset
>> 9;
1142 int i
, update_bitmap
= 0;
1144 for(i
= 0; i
< length
>> 9; i
++){
1145 if(cow_mask
!= NULL
)
1146 ubd_set_bit(i
, (unsigned char *) cow_mask
);
1147 if(ubd_test_bit(sector
+ i
, (unsigned char *) bitmap
))
1151 ubd_set_bit(sector
+ i
, (unsigned char *) bitmap
);
1157 *cow_offset
= sector
/ (sizeof(unsigned long) * 8);
1159 /* This takes care of the case where we're exactly at the end of the
1160 * device, and *cow_offset + 1 is off the end. So, just back it up
1161 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1162 * for the original diagnosis.
1164 if (*cow_offset
== (DIV_ROUND_UP(bitmap_len
,
1165 sizeof(unsigned long)) - 1))
1168 bitmap_words
[0] = bitmap
[*cow_offset
];
1169 bitmap_words
[1] = bitmap
[*cow_offset
+ 1];
1171 *cow_offset
*= sizeof(unsigned long);
1172 *cow_offset
+= bitmap_offset
;
1175 static void cowify_req(struct io_thread_req
*req
, unsigned long *bitmap
,
1176 __u64 bitmap_offset
, __u64 bitmap_len
)
1178 __u64 sector
= req
->offset
>> 9;
1181 if(req
->length
> (sizeof(req
->sector_mask
) * 8) << 9)
1182 panic("Operation too long");
1184 if(req
->op
== UBD_READ
) {
1185 for(i
= 0; i
< req
->length
>> 9; i
++){
1186 if(ubd_test_bit(sector
+ i
, (unsigned char *) bitmap
))
1187 ubd_set_bit(i
, (unsigned char *)
1191 else cowify_bitmap(req
->offset
, req
->length
, &req
->sector_mask
,
1192 &req
->cow_offset
, bitmap
, bitmap_offset
,
1193 req
->bitmap_words
, bitmap_len
);
1196 /* Called with dev->lock held */
1197 static void prepare_request(struct request
*req
, struct io_thread_req
*io_req
,
1198 unsigned long long offset
, int page_offset
,
1199 int len
, struct page
*page
)
1201 struct gendisk
*disk
= req
->rq_disk
;
1202 struct ubd
*ubd_dev
= disk
->private_data
;
1205 io_req
->fds
[0] = (ubd_dev
->cow
.file
!= NULL
) ? ubd_dev
->cow
.fd
:
1207 io_req
->fds
[1] = ubd_dev
->fd
;
1208 io_req
->cow_offset
= -1;
1209 io_req
->offset
= offset
;
1210 io_req
->length
= len
;
1212 io_req
->sector_mask
= 0;
1214 io_req
->op
= (rq_data_dir(req
) == READ
) ? UBD_READ
: UBD_WRITE
;
1215 io_req
->offsets
[0] = 0;
1216 io_req
->offsets
[1] = ubd_dev
->cow
.data_offset
;
1217 io_req
->buffer
= page_address(page
) + page_offset
;
1218 io_req
->sectorsize
= 1 << 9;
1220 if(ubd_dev
->cow
.file
!= NULL
)
1221 cowify_req(io_req
, ubd_dev
->cow
.bitmap
,
1222 ubd_dev
->cow
.bitmap_offset
, ubd_dev
->cow
.bitmap_len
);
1226 /* Called with dev->lock held */
1227 static void do_ubd_request(struct request_queue
*q
)
1229 struct io_thread_req
*io_req
;
1230 struct request
*req
;
1235 struct ubd
*dev
= q
->queuedata
;
1236 if(dev
->end_sg
== 0){
1237 struct request
*req
= blk_fetch_request(q
);
1243 dev
->end_sg
= blk_rq_map_sg(q
, req
, dev
->sg
);
1247 sector
= blk_rq_pos(req
);
1248 while(dev
->start_sg
< dev
->end_sg
){
1249 struct scatterlist
*sg
= &dev
->sg
[dev
->start_sg
];
1251 io_req
= kmalloc(sizeof(struct io_thread_req
),
1254 if(list_empty(&dev
->restart
))
1255 list_add(&dev
->restart
, &restart
);
1258 prepare_request(req
, io_req
,
1259 (unsigned long long)sector
<< 9,
1260 sg
->offset
, sg
->length
, sg_page(sg
));
1262 sector
+= sg
->length
>> 9;
1263 n
= os_write_file(thread_fd
, &io_req
,
1264 sizeof(struct io_thread_req
*));
1265 if(n
!= sizeof(struct io_thread_req
*)){
1267 printk("write to io thread failed, "
1268 "errno = %d\n", -n
);
1269 else if(list_empty(&dev
->restart
))
1270 list_add(&dev
->restart
, &restart
);
1278 dev
->request
= NULL
;
1282 static int ubd_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1284 struct ubd
*ubd_dev
= bdev
->bd_disk
->private_data
;
1288 geo
->cylinders
= ubd_dev
->size
/ (128 * 32 * 512);
1292 static int ubd_ioctl(struct block_device
*bdev
, fmode_t mode
,
1293 unsigned int cmd
, unsigned long arg
)
1295 struct ubd
*ubd_dev
= bdev
->bd_disk
->private_data
;
1296 u16 ubd_id
[ATA_ID_WORDS
];
1299 struct cdrom_volctrl volume
;
1300 case HDIO_GET_IDENTITY
:
1301 memset(&ubd_id
, 0, ATA_ID_WORDS
* 2);
1302 ubd_id
[ATA_ID_CYLS
] = ubd_dev
->size
/ (128 * 32 * 512);
1303 ubd_id
[ATA_ID_HEADS
] = 128;
1304 ubd_id
[ATA_ID_SECTORS
] = 32;
1305 if(copy_to_user((char __user
*) arg
, (char *) &ubd_id
,
1311 if(copy_from_user(&volume
, (char __user
*) arg
, sizeof(volume
)))
1313 volume
.channel0
= 255;
1314 volume
.channel1
= 255;
1315 volume
.channel2
= 255;
1316 volume
.channel3
= 255;
1317 if(copy_to_user((char __user
*) arg
, &volume
, sizeof(volume
)))
1324 static int update_bitmap(struct io_thread_req
*req
)
1328 if(req
->cow_offset
== -1)
1331 n
= os_seek_file(req
->fds
[1], req
->cow_offset
);
1333 printk("do_io - bitmap lseek failed : err = %d\n", -n
);
1337 n
= os_write_file(req
->fds
[1], &req
->bitmap_words
,
1338 sizeof(req
->bitmap_words
));
1339 if(n
!= sizeof(req
->bitmap_words
)){
1340 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n
,
1348 static void do_io(struct io_thread_req
*req
)
1352 int n
, nsectors
, start
, end
, bit
;
1356 nsectors
= req
->length
/ req
->sectorsize
;
1359 bit
= ubd_test_bit(start
, (unsigned char *) &req
->sector_mask
);
1361 while((end
< nsectors
) &&
1362 (ubd_test_bit(end
, (unsigned char *)
1363 &req
->sector_mask
) == bit
))
1366 off
= req
->offset
+ req
->offsets
[bit
] +
1367 start
* req
->sectorsize
;
1368 len
= (end
- start
) * req
->sectorsize
;
1369 buf
= &req
->buffer
[start
* req
->sectorsize
];
1371 err
= os_seek_file(req
->fds
[bit
], off
);
1373 printk("do_io - lseek failed : err = %d\n", -err
);
1377 if(req
->op
== UBD_READ
){
1382 n
= os_read_file(req
->fds
[bit
], buf
, len
);
1384 printk("do_io - read failed, err = %d "
1385 "fd = %d\n", -n
, req
->fds
[bit
]);
1389 } while((n
< len
) && (n
!= 0));
1390 if (n
< len
) memset(&buf
[n
], 0, len
- n
);
1392 n
= os_write_file(req
->fds
[bit
], buf
, len
);
1394 printk("do_io - write failed err = %d "
1395 "fd = %d\n", -n
, req
->fds
[bit
]);
1402 } while(start
< nsectors
);
1404 req
->error
= update_bitmap(req
);
1407 /* Changed in start_io_thread, which is serialized by being called only
1408 * from ubd_init, which is an initcall.
1412 /* Only changed by the io thread. XXX: currently unused. */
1413 static int io_count
= 0;
1415 int io_thread(void *arg
)
1417 struct io_thread_req
*req
;
1420 ignore_sigwinch_sig();
1422 n
= os_read_file(kernel_fd
, &req
,
1423 sizeof(struct io_thread_req
*));
1424 if(n
!= sizeof(struct io_thread_req
*)){
1426 printk("io_thread - read failed, fd = %d, "
1427 "err = %d\n", kernel_fd
, -n
);
1429 printk("io_thread - short read, fd = %d, "
1430 "length = %d\n", kernel_fd
, n
);
1436 n
= os_write_file(kernel_fd
, &req
,
1437 sizeof(struct io_thread_req
*));
1438 if(n
!= sizeof(struct io_thread_req
*))
1439 printk("io_thread - write failed, fd = %d, err = %d\n",