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/kernel.h"
24 #include "linux/module.h"
25 #include "linux/blkdev.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/ctype.h"
31 #include "linux/capability.h"
33 #include "linux/vmalloc.h"
34 #include "linux/blkpg.h"
35 #include "linux/genhd.h"
36 #include "linux/spinlock.h"
37 #include "linux/platform_device.h"
38 #include "linux/scatterlist.h"
39 #include "asm/segment.h"
40 #include "asm/uaccess.h"
42 #include "asm/types.h"
43 #include "asm/tlbflush.h"
45 #include "kern_util.h"
47 #include "mconsole_kern.h"
52 #include "kern_util.h"
58 enum ubd_req
{ UBD_READ
, UBD_WRITE
};
60 struct io_thread_req
{
64 unsigned long offsets
[2];
65 unsigned long long offset
;
69 unsigned long sector_mask
;
70 unsigned long long cow_offset
;
71 unsigned long bitmap_words
[2];
75 extern int open_ubd_file(char *file
, struct openflags
*openflags
, int shared
,
76 char **backing_file_out
, int *bitmap_offset_out
,
77 unsigned long *bitmap_len_out
, int *data_offset_out
,
79 extern int create_cow_file(char *cow_file
, char *backing_file
,
80 struct openflags flags
, int sectorsize
,
81 int alignment
, int *bitmap_offset_out
,
82 unsigned long *bitmap_len_out
,
83 int *data_offset_out
);
84 extern int read_cow_bitmap(int fd
, void *buf
, int offset
, int len
);
85 extern void do_io(struct io_thread_req
*req
);
87 static inline int ubd_test_bit(__u64 bit
, unsigned char *data
)
92 bits
= sizeof(data
[0]) * 8;
95 return (data
[n
] & (1 << off
)) != 0;
98 static inline void ubd_set_bit(__u64 bit
, unsigned char *data
)
103 bits
= sizeof(data
[0]) * 8;
106 data
[n
] |= (1 << off
);
108 /*End stuff from ubd_user.h*/
110 #define DRIVER_NAME "uml-blkdev"
112 static DEFINE_MUTEX(ubd_lock
);
114 static int ubd_open(struct inode
* inode
, struct file
* filp
);
115 static int ubd_release(struct inode
* inode
, struct file
* file
);
116 static int ubd_ioctl(struct inode
* inode
, struct file
* file
,
117 unsigned int cmd
, unsigned long arg
);
118 static int ubd_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
);
122 static struct block_device_operations ubd_blops
= {
123 .owner
= THIS_MODULE
,
125 .release
= ubd_release
,
127 .getgeo
= ubd_getgeo
,
130 /* Protected by ubd_lock */
131 static int fake_major
= MAJOR_NR
;
132 static struct gendisk
*ubd_gendisk
[MAX_DEV
];
133 static struct gendisk
*fake_gendisk
[MAX_DEV
];
135 #ifdef CONFIG_BLK_DEV_UBD_SYNC
136 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
139 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
142 static struct openflags global_openflags
= OPEN_FLAGS
;
145 /* backing file name */
147 /* backing file fd */
149 unsigned long *bitmap
;
150 unsigned long bitmap_len
;
158 struct list_head restart
;
159 /* name (and fd, below) of the file opened for writing, either the
160 * backing or the cow file. */
165 struct openflags boot_openflags
;
166 struct openflags openflags
;
170 struct platform_device pdev
;
171 struct request_queue
*queue
;
173 struct scatterlist sg
[MAX_SG
];
174 struct request
*request
;
175 int start_sg
, end_sg
;
178 #define DEFAULT_COW { \
182 .bitmap_offset = 0, \
186 #define DEFAULT_UBD { \
191 .boot_openflags = OPEN_FLAGS, \
192 .openflags = OPEN_FLAGS, \
195 .cow = DEFAULT_COW, \
196 .lock = SPIN_LOCK_UNLOCKED, \
202 /* Protected by ubd_lock */
203 struct ubd ubd_devs
[MAX_DEV
] = { [ 0 ... MAX_DEV
- 1 ] = DEFAULT_UBD
};
205 /* Only changed by fake_ide_setup which is a setup */
206 static int fake_ide
= 0;
207 static struct proc_dir_entry
*proc_ide_root
= NULL
;
208 static struct proc_dir_entry
*proc_ide
= NULL
;
210 static void make_proc_ide(void)
212 proc_ide_root
= proc_mkdir("ide", NULL
);
213 proc_ide
= proc_mkdir("ide0", proc_ide_root
);
216 static int proc_ide_read_media(char *page
, char **start
, off_t off
, int count
,
217 int *eof
, void *data
)
221 strcpy(page
, "disk\n");
222 len
= strlen("disk\n");
226 if (len
<= 0) return 0;
233 static void make_ide_entries(const char *dev_name
)
235 struct proc_dir_entry
*dir
, *ent
;
238 if(proc_ide_root
== NULL
) make_proc_ide();
240 dir
= proc_mkdir(dev_name
, proc_ide
);
243 ent
= create_proc_entry("media", S_IFREG
|S_IRUGO
, dir
);
246 ent
->read_proc
= proc_ide_read_media
;
247 ent
->write_proc
= NULL
;
248 snprintf(name
, sizeof(name
), "ide0/%s", dev_name
);
249 proc_symlink(dev_name
, proc_ide_root
, name
);
252 static int fake_ide_setup(char *str
)
258 __setup("fake_ide", fake_ide_setup
);
260 __uml_help(fake_ide_setup
,
262 " Create ide0 entries that map onto ubd devices.\n\n"
265 static int parse_unit(char **ptr
)
267 char *str
= *ptr
, *end
;
271 n
= simple_strtoul(str
, &end
, 0);
276 else if (('a' <= *str
) && (*str
<= 'z')) {
284 /* If *index_out == -1 at exit, the passed option was a general one;
285 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
286 * should not be freed on exit.
288 static int ubd_setup_common(char *str
, int *index_out
, char **error_out
)
291 struct openflags flags
= global_openflags
;
295 if(index_out
) *index_out
= -1;
302 if(!strcmp(str
, "sync")){
303 global_openflags
= of_sync(global_openflags
);
308 major
= simple_strtoul(str
, &end
, 0);
309 if((*end
!= '\0') || (end
== str
)){
310 *error_out
= "Didn't parse major number";
314 mutex_lock(&ubd_lock
);
315 if(fake_major
!= MAJOR_NR
){
316 *error_out
= "Can't assign a fake major twice";
322 printk(KERN_INFO
"Setting extra ubd major number to %d\n",
326 mutex_unlock(&ubd_lock
);
330 n
= parse_unit(&str
);
332 *error_out
= "Couldn't parse device number";
336 *error_out
= "Device number out of range";
341 mutex_lock(&ubd_lock
);
343 ubd_dev
= &ubd_devs
[n
];
344 if(ubd_dev
->file
!= NULL
){
345 *error_out
= "Device is already configured";
353 for (i
= 0; i
< sizeof("rscd="); i
++) {
371 *error_out
= "Expected '=' or flag letter "
379 *error_out
= "Too many flags specified";
381 *error_out
= "Missing '='";
385 backing_file
= strchr(str
, ',');
387 if (backing_file
== NULL
)
388 backing_file
= strchr(str
, ':');
390 if(backing_file
!= NULL
){
392 *error_out
= "Can't specify both 'd' and a cow file";
396 *backing_file
= '\0';
402 ubd_dev
->cow
.file
= backing_file
;
403 ubd_dev
->boot_openflags
= flags
;
405 mutex_unlock(&ubd_lock
);
409 static int ubd_setup(char *str
)
414 err
= ubd_setup_common(str
, NULL
, &error
);
416 printk(KERN_ERR
"Failed to initialize device with \"%s\" : "
421 __setup("ubd", ubd_setup
);
422 __uml_help(ubd_setup
,
423 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
424 " This is used to associate a device with a file in the underlying\n"
425 " filesystem. When specifying two filenames, the first one is the\n"
426 " COW name and the second is the backing file name. As separator you can\n"
427 " use either a ':' or a ',': the first one allows writing things like;\n"
428 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
429 " while with a ',' the shell would not expand the 2nd '~'.\n"
430 " When using only one filename, UML will detect whether to treat it like\n"
431 " a COW file or a backing file. To override this detection, add the 'd'\n"
433 " ubd0d=BackingFile\n"
434 " Usually, there is a filesystem in the file, but \n"
435 " that's not required. Swap devices containing swap files can be\n"
436 " specified like this. Also, a file which doesn't contain a\n"
437 " filesystem can have its contents read in the virtual \n"
438 " machine by running 'dd' on the device. <n> must be in the range\n"
439 " 0 to 7. Appending an 'r' to the number will cause that device\n"
440 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
441 " an 's' will cause data to be written to disk on the host immediately.\n"
442 " 'c' will cause the device to be treated as being shared between multiple\n"
443 " UMLs and file locking will be turned off - this is appropriate for a\n"
444 " cluster filesystem and inappropriate at almost all other times.\n\n"
447 static int udb_setup(char *str
)
449 printk("udb%s specified on command line is almost certainly a ubd -> "
454 __setup("udb", udb_setup
);
455 __uml_help(udb_setup
,
457 " This option is here solely to catch ubd -> udb typos, which can be\n"
458 " to impossible to catch visually unless you specifically look for\n"
459 " them. The only result of any option starting with 'udb' is an error\n"
460 " in the boot output.\n\n"
463 static void do_ubd_request(struct request_queue
* q
);
465 /* Only changed by ubd_init, which is an initcall. */
468 static void ubd_end_request(struct request
*req
, int bytes
, int error
)
470 blk_end_request(req
, error
, bytes
);
473 /* Callable only from interrupt context - otherwise you need to do
474 * spin_lock_irq()/spin_lock_irqsave() */
475 static inline void ubd_finish(struct request
*req
, int bytes
)
478 ubd_end_request(req
, 0, -EIO
);
481 ubd_end_request(req
, bytes
, 0);
484 static LIST_HEAD(restart
);
486 /* XXX - move this inside ubd_intr. */
487 /* Called without dev->lock held, and only in interrupt context. */
488 static void ubd_handler(void)
490 struct io_thread_req
*req
;
493 struct list_head
*list
, *next_ele
;
498 n
= os_read_file(thread_fd
, &req
,
499 sizeof(struct io_thread_req
*));
500 if(n
!= sizeof(req
)){
503 printk(KERN_ERR
"spurious interrupt in ubd_handler, "
509 rq
->nr_sectors
-= req
->length
>> 9;
510 if(rq
->nr_sectors
== 0)
511 ubd_finish(rq
, rq
->hard_nr_sectors
<< 9);
514 reactivate_fd(thread_fd
, UBD_IRQ
);
516 list_for_each_safe(list
, next_ele
, &restart
){
517 ubd
= container_of(list
, struct ubd
, restart
);
518 list_del_init(&ubd
->restart
);
519 spin_lock_irqsave(&ubd
->lock
, flags
);
520 do_ubd_request(ubd
->queue
);
521 spin_unlock_irqrestore(&ubd
->lock
, flags
);
525 static irqreturn_t
ubd_intr(int irq
, void *dev
)
531 /* Only changed by ubd_init, which is an initcall. */
532 static int io_pid
= -1;
534 void kill_io_thread(void)
537 os_kill_process(io_pid
, 1);
540 __uml_exitcall(kill_io_thread
);
542 static inline int ubd_file_size(struct ubd
*ubd_dev
, __u64
*size_out
)
546 file
= ubd_dev
->cow
.file
? ubd_dev
->cow
.file
: ubd_dev
->file
;
547 return os_file_size(file
, size_out
);
550 static void ubd_close_dev(struct ubd
*ubd_dev
)
552 os_close_file(ubd_dev
->fd
);
553 if(ubd_dev
->cow
.file
== NULL
)
556 os_close_file(ubd_dev
->cow
.fd
);
557 vfree(ubd_dev
->cow
.bitmap
);
558 ubd_dev
->cow
.bitmap
= NULL
;
561 static int ubd_open_dev(struct ubd
*ubd_dev
)
563 struct openflags flags
;
565 int err
, create_cow
, *create_ptr
;
568 ubd_dev
->openflags
= ubd_dev
->boot_openflags
;
570 create_ptr
= (ubd_dev
->cow
.file
!= NULL
) ? &create_cow
: NULL
;
571 back_ptr
= ubd_dev
->no_cow
? NULL
: &ubd_dev
->cow
.file
;
573 fd
= open_ubd_file(ubd_dev
->file
, &ubd_dev
->openflags
, ubd_dev
->shared
,
574 back_ptr
, &ubd_dev
->cow
.bitmap_offset
,
575 &ubd_dev
->cow
.bitmap_len
, &ubd_dev
->cow
.data_offset
,
578 if((fd
== -ENOENT
) && create_cow
){
579 fd
= create_cow_file(ubd_dev
->file
, ubd_dev
->cow
.file
,
580 ubd_dev
->openflags
, 1 << 9, PAGE_SIZE
,
581 &ubd_dev
->cow
.bitmap_offset
,
582 &ubd_dev
->cow
.bitmap_len
,
583 &ubd_dev
->cow
.data_offset
);
585 printk(KERN_INFO
"Creating \"%s\" as COW file for "
586 "\"%s\"\n", ubd_dev
->file
, ubd_dev
->cow
.file
);
591 printk("Failed to open '%s', errno = %d\n", ubd_dev
->file
,
597 if(ubd_dev
->cow
.file
!= NULL
){
598 blk_queue_max_sectors(ubd_dev
->queue
, 8 * sizeof(long));
601 ubd_dev
->cow
.bitmap
= vmalloc(ubd_dev
->cow
.bitmap_len
);
602 if(ubd_dev
->cow
.bitmap
== NULL
){
603 printk(KERN_ERR
"Failed to vmalloc COW bitmap\n");
606 flush_tlb_kernel_vm();
608 err
= read_cow_bitmap(ubd_dev
->fd
, ubd_dev
->cow
.bitmap
,
609 ubd_dev
->cow
.bitmap_offset
,
610 ubd_dev
->cow
.bitmap_len
);
614 flags
= ubd_dev
->openflags
;
616 err
= open_ubd_file(ubd_dev
->cow
.file
, &flags
, ubd_dev
->shared
, NULL
,
617 NULL
, NULL
, NULL
, NULL
);
618 if(err
< 0) goto error
;
619 ubd_dev
->cow
.fd
= err
;
623 os_close_file(ubd_dev
->fd
);
627 static void ubd_device_release(struct device
*dev
)
629 struct ubd
*ubd_dev
= dev
->driver_data
;
631 blk_cleanup_queue(ubd_dev
->queue
);
632 *ubd_dev
= ((struct ubd
) DEFAULT_UBD
);
635 static int ubd_disk_register(int major
, u64 size
, int unit
,
636 struct gendisk
**disk_out
)
638 struct gendisk
*disk
;
640 disk
= alloc_disk(1 << UBD_SHIFT
);
645 disk
->first_minor
= unit
<< UBD_SHIFT
;
646 disk
->fops
= &ubd_blops
;
647 set_capacity(disk
, size
/ 512);
648 if(major
== MAJOR_NR
)
649 sprintf(disk
->disk_name
, "ubd%c", 'a' + unit
);
651 sprintf(disk
->disk_name
, "ubd_fake%d", unit
);
653 /* sysfs register (not for ide fake devices) */
654 if (major
== MAJOR_NR
) {
655 ubd_devs
[unit
].pdev
.id
= unit
;
656 ubd_devs
[unit
].pdev
.name
= DRIVER_NAME
;
657 ubd_devs
[unit
].pdev
.dev
.release
= ubd_device_release
;
658 ubd_devs
[unit
].pdev
.dev
.driver_data
= &ubd_devs
[unit
];
659 platform_device_register(&ubd_devs
[unit
].pdev
);
660 disk
->driverfs_dev
= &ubd_devs
[unit
].pdev
.dev
;
663 disk
->private_data
= &ubd_devs
[unit
];
664 disk
->queue
= ubd_devs
[unit
].queue
;
671 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
673 static int ubd_add(int n
, char **error_out
)
675 struct ubd
*ubd_dev
= &ubd_devs
[n
];
678 if(ubd_dev
->file
== NULL
)
681 err
= ubd_file_size(ubd_dev
, &ubd_dev
->size
);
683 *error_out
= "Couldn't determine size of device's file";
687 ubd_dev
->size
= ROUND_BLOCK(ubd_dev
->size
);
689 INIT_LIST_HEAD(&ubd_dev
->restart
);
690 sg_init_table(ubd_dev
->sg
, MAX_SG
);
693 ubd_dev
->queue
= blk_init_queue(do_ubd_request
, &ubd_dev
->lock
);
694 if (ubd_dev
->queue
== NULL
) {
695 *error_out
= "Failed to initialize device queue";
698 ubd_dev
->queue
->queuedata
= ubd_dev
;
700 blk_queue_max_hw_segments(ubd_dev
->queue
, MAX_SG
);
701 err
= ubd_disk_register(MAJOR_NR
, ubd_dev
->size
, n
, &ubd_gendisk
[n
]);
703 *error_out
= "Failed to register device";
707 if(fake_major
!= MAJOR_NR
)
708 ubd_disk_register(fake_major
, ubd_dev
->size
, n
,
712 * Perhaps this should also be under the "if (fake_major)" above
713 * using the fake_disk->disk_name
716 make_ide_entries(ubd_gendisk
[n
]->disk_name
);
723 blk_cleanup_queue(ubd_dev
->queue
);
727 static int ubd_config(char *str
, char **error_out
)
731 /* This string is possibly broken up and stored, so it's only
732 * freed if ubd_setup_common fails, or if only general options
735 str
= kstrdup(str
, GFP_KERNEL
);
737 *error_out
= "Failed to allocate memory";
741 ret
= ubd_setup_common(str
, &n
, error_out
);
750 mutex_lock(&ubd_lock
);
751 ret
= ubd_add(n
, error_out
);
753 ubd_devs
[n
].file
= NULL
;
754 mutex_unlock(&ubd_lock
);
764 static int ubd_get_config(char *name
, char *str
, int size
, char **error_out
)
769 n
= parse_unit(&name
);
770 if((n
>= MAX_DEV
) || (n
< 0)){
771 *error_out
= "ubd_get_config : device number out of range";
775 ubd_dev
= &ubd_devs
[n
];
776 mutex_lock(&ubd_lock
);
778 if(ubd_dev
->file
== NULL
){
779 CONFIG_CHUNK(str
, size
, len
, "", 1);
783 CONFIG_CHUNK(str
, size
, len
, ubd_dev
->file
, 0);
785 if(ubd_dev
->cow
.file
!= NULL
){
786 CONFIG_CHUNK(str
, size
, len
, ",", 0);
787 CONFIG_CHUNK(str
, size
, len
, ubd_dev
->cow
.file
, 1);
789 else CONFIG_CHUNK(str
, size
, len
, "", 1);
792 mutex_unlock(&ubd_lock
);
796 static int ubd_id(char **str
, int *start_out
, int *end_out
)
802 *end_out
= MAX_DEV
- 1;
806 static int ubd_remove(int n
, char **error_out
)
808 struct gendisk
*disk
= ubd_gendisk
[n
];
812 mutex_lock(&ubd_lock
);
814 ubd_dev
= &ubd_devs
[n
];
816 if(ubd_dev
->file
== NULL
)
819 /* you cannot remove a open disk */
821 if(ubd_dev
->count
> 0)
824 ubd_gendisk
[n
] = NULL
;
830 if(fake_gendisk
[n
] != NULL
){
831 del_gendisk(fake_gendisk
[n
]);
832 put_disk(fake_gendisk
[n
]);
833 fake_gendisk
[n
] = NULL
;
837 platform_device_unregister(&ubd_dev
->pdev
);
839 mutex_unlock(&ubd_lock
);
843 /* All these are called by mconsole in process context and without
844 * ubd-specific locks. The structure itself is const except for .list.
846 static struct mc_device ubd_mc
= {
847 .list
= LIST_HEAD_INIT(ubd_mc
.list
),
849 .config
= ubd_config
,
850 .get_config
= ubd_get_config
,
852 .remove
= ubd_remove
,
855 static int __init
ubd_mc_init(void)
857 mconsole_register_dev(&ubd_mc
);
861 __initcall(ubd_mc_init
);
863 static int __init
ubd0_init(void)
865 struct ubd
*ubd_dev
= &ubd_devs
[0];
867 mutex_lock(&ubd_lock
);
868 if(ubd_dev
->file
== NULL
)
869 ubd_dev
->file
= "root_fs";
870 mutex_unlock(&ubd_lock
);
875 __initcall(ubd0_init
);
877 /* Used in ubd_init, which is an initcall */
878 static struct platform_driver ubd_driver
= {
884 static int __init
ubd_init(void)
889 if (register_blkdev(MAJOR_NR
, "ubd"))
892 if (fake_major
!= MAJOR_NR
) {
893 char name
[sizeof("ubd_nnn\0")];
895 snprintf(name
, sizeof(name
), "ubd_%d", fake_major
);
896 if (register_blkdev(fake_major
, "ubd"))
899 platform_driver_register(&ubd_driver
);
900 mutex_lock(&ubd_lock
);
901 for (i
= 0; i
< MAX_DEV
; i
++){
902 err
= ubd_add(i
, &error
);
904 printk(KERN_ERR
"Failed to initialize ubd device %d :"
907 mutex_unlock(&ubd_lock
);
911 late_initcall(ubd_init
);
913 static int __init
ubd_driver_init(void){
917 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
918 if(global_openflags
.s
){
919 printk(KERN_INFO
"ubd: Synchronous mode\n");
920 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
921 * enough. So use anyway the io thread. */
923 stack
= alloc_stack(0, 0);
924 io_pid
= start_io_thread(stack
+ PAGE_SIZE
- sizeof(void *),
928 "ubd : Failed to start I/O thread (errno = %d) - "
929 "falling back to synchronous I/O\n", -io_pid
);
933 err
= um_request_irq(UBD_IRQ
, thread_fd
, IRQ_READ
, ubd_intr
,
934 IRQF_DISABLED
, "ubd", ubd_devs
);
936 printk(KERN_ERR
"um_request_irq failed - errno = %d\n", -err
);
940 device_initcall(ubd_driver_init
);
942 static int ubd_open(struct inode
*inode
, struct file
*filp
)
944 struct gendisk
*disk
= inode
->i_bdev
->bd_disk
;
945 struct ubd
*ubd_dev
= disk
->private_data
;
948 if(ubd_dev
->count
== 0){
949 err
= ubd_open_dev(ubd_dev
);
951 printk(KERN_ERR
"%s: Can't open \"%s\": errno = %d\n",
952 disk
->disk_name
, ubd_dev
->file
, -err
);
957 set_disk_ro(disk
, !ubd_dev
->openflags
.w
);
959 /* This should no more be needed. And it didn't work anyway to exclude
960 * read-write remounting of filesystems.*/
961 /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
962 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
969 static int ubd_release(struct inode
* inode
, struct file
* file
)
971 struct gendisk
*disk
= inode
->i_bdev
->bd_disk
;
972 struct ubd
*ubd_dev
= disk
->private_data
;
974 if(--ubd_dev
->count
== 0)
975 ubd_close_dev(ubd_dev
);
979 static void cowify_bitmap(__u64 io_offset
, int length
, unsigned long *cow_mask
,
980 __u64
*cow_offset
, unsigned long *bitmap
,
981 __u64 bitmap_offset
, unsigned long *bitmap_words
,
984 __u64 sector
= io_offset
>> 9;
985 int i
, update_bitmap
= 0;
987 for(i
= 0; i
< length
>> 9; i
++){
989 ubd_set_bit(i
, (unsigned char *) cow_mask
);
990 if(ubd_test_bit(sector
+ i
, (unsigned char *) bitmap
))
994 ubd_set_bit(sector
+ i
, (unsigned char *) bitmap
);
1000 *cow_offset
= sector
/ (sizeof(unsigned long) * 8);
1002 /* This takes care of the case where we're exactly at the end of the
1003 * device, and *cow_offset + 1 is off the end. So, just back it up
1004 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1005 * for the original diagnosis.
1007 if(*cow_offset
== ((bitmap_len
+ sizeof(unsigned long) - 1) /
1008 sizeof(unsigned long) - 1))
1011 bitmap_words
[0] = bitmap
[*cow_offset
];
1012 bitmap_words
[1] = bitmap
[*cow_offset
+ 1];
1014 *cow_offset
*= sizeof(unsigned long);
1015 *cow_offset
+= bitmap_offset
;
1018 static void cowify_req(struct io_thread_req
*req
, unsigned long *bitmap
,
1019 __u64 bitmap_offset
, __u64 bitmap_len
)
1021 __u64 sector
= req
->offset
>> 9;
1024 if(req
->length
> (sizeof(req
->sector_mask
) * 8) << 9)
1025 panic("Operation too long");
1027 if(req
->op
== UBD_READ
) {
1028 for(i
= 0; i
< req
->length
>> 9; i
++){
1029 if(ubd_test_bit(sector
+ i
, (unsigned char *) bitmap
))
1030 ubd_set_bit(i
, (unsigned char *)
1034 else cowify_bitmap(req
->offset
, req
->length
, &req
->sector_mask
,
1035 &req
->cow_offset
, bitmap
, bitmap_offset
,
1036 req
->bitmap_words
, bitmap_len
);
1039 /* Called with dev->lock held */
1040 static void prepare_request(struct request
*req
, struct io_thread_req
*io_req
,
1041 unsigned long long offset
, int page_offset
,
1042 int len
, struct page
*page
)
1044 struct gendisk
*disk
= req
->rq_disk
;
1045 struct ubd
*ubd_dev
= disk
->private_data
;
1048 io_req
->fds
[0] = (ubd_dev
->cow
.file
!= NULL
) ? ubd_dev
->cow
.fd
:
1050 io_req
->fds
[1] = ubd_dev
->fd
;
1051 io_req
->cow_offset
= -1;
1052 io_req
->offset
= offset
;
1053 io_req
->length
= len
;
1055 io_req
->sector_mask
= 0;
1057 io_req
->op
= (rq_data_dir(req
) == READ
) ? UBD_READ
: UBD_WRITE
;
1058 io_req
->offsets
[0] = 0;
1059 io_req
->offsets
[1] = ubd_dev
->cow
.data_offset
;
1060 io_req
->buffer
= page_address(page
) + page_offset
;
1061 io_req
->sectorsize
= 1 << 9;
1063 if(ubd_dev
->cow
.file
!= NULL
)
1064 cowify_req(io_req
, ubd_dev
->cow
.bitmap
,
1065 ubd_dev
->cow
.bitmap_offset
, ubd_dev
->cow
.bitmap_len
);
1069 /* Called with dev->lock held */
1070 static void do_ubd_request(struct request_queue
*q
)
1072 struct io_thread_req
*io_req
;
1073 struct request
*req
;
1074 int n
, last_sectors
;
1077 struct ubd
*dev
= q
->queuedata
;
1078 if(dev
->end_sg
== 0){
1079 struct request
*req
= elv_next_request(q
);
1084 blkdev_dequeue_request(req
);
1086 dev
->end_sg
= blk_rq_map_sg(q
, req
, dev
->sg
);
1091 while(dev
->start_sg
< dev
->end_sg
){
1092 struct scatterlist
*sg
= &dev
->sg
[dev
->start_sg
];
1094 req
->sector
+= last_sectors
;
1095 io_req
= kmalloc(sizeof(struct io_thread_req
),
1098 if(list_empty(&dev
->restart
))
1099 list_add(&dev
->restart
, &restart
);
1102 prepare_request(req
, io_req
,
1103 (unsigned long long) req
->sector
<< 9,
1104 sg
->offset
, sg
->length
, sg_page(sg
));
1106 last_sectors
= sg
->length
>> 9;
1107 n
= os_write_file(thread_fd
, &io_req
,
1108 sizeof(struct io_thread_req
*));
1109 if(n
!= sizeof(struct io_thread_req
*)){
1111 printk("write to io thread failed, "
1112 "errno = %d\n", -n
);
1113 else if(list_empty(&dev
->restart
))
1114 list_add(&dev
->restart
, &restart
);
1122 dev
->request
= NULL
;
1126 static int ubd_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1128 struct ubd
*ubd_dev
= bdev
->bd_disk
->private_data
;
1132 geo
->cylinders
= ubd_dev
->size
/ (128 * 32 * 512);
1136 static int ubd_ioctl(struct inode
* inode
, struct file
* file
,
1137 unsigned int cmd
, unsigned long arg
)
1139 struct ubd
*ubd_dev
= inode
->i_bdev
->bd_disk
->private_data
;
1140 struct hd_driveid ubd_id
= {
1147 struct cdrom_volctrl volume
;
1148 case HDIO_GET_IDENTITY
:
1149 ubd_id
.cyls
= ubd_dev
->size
/ (128 * 32 * 512);
1150 if(copy_to_user((char __user
*) arg
, (char *) &ubd_id
,
1156 if(copy_from_user(&volume
, (char __user
*) arg
, sizeof(volume
)))
1158 volume
.channel0
= 255;
1159 volume
.channel1
= 255;
1160 volume
.channel2
= 255;
1161 volume
.channel3
= 255;
1162 if(copy_to_user((char __user
*) arg
, &volume
, sizeof(volume
)))
1169 static int path_requires_switch(char *from_cmdline
, char *from_cow
, char *cow
)
1171 struct uml_stat buf1
, buf2
;
1174 if(from_cmdline
== NULL
)
1176 if(!strcmp(from_cmdline
, from_cow
))
1179 err
= os_stat_file(from_cmdline
, &buf1
);
1181 printk("Couldn't stat '%s', err = %d\n", from_cmdline
, -err
);
1184 err
= os_stat_file(from_cow
, &buf2
);
1186 printk("Couldn't stat '%s', err = %d\n", from_cow
, -err
);
1189 if((buf1
.ust_dev
== buf2
.ust_dev
) && (buf1
.ust_ino
== buf2
.ust_ino
))
1192 printk("Backing file mismatch - \"%s\" requested,\n"
1193 "\"%s\" specified in COW header of \"%s\"\n",
1194 from_cmdline
, from_cow
, cow
);
1198 static int backing_file_mismatch(char *file
, __u64 size
, time_t mtime
)
1200 unsigned long modtime
;
1201 unsigned long long actual
;
1204 err
= os_file_modtime(file
, &modtime
);
1206 printk("Failed to get modification time of backing file "
1207 "\"%s\", err = %d\n", file
, -err
);
1211 err
= os_file_size(file
, &actual
);
1213 printk("Failed to get size of backing file \"%s\", "
1214 "err = %d\n", file
, -err
);
1219 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1221 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1222 "file\n", (unsigned long long) size
, actual
);
1225 if(modtime
!= mtime
){
1226 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1227 "file\n", mtime
, modtime
);
1233 int read_cow_bitmap(int fd
, void *buf
, int offset
, int len
)
1237 err
= os_seek_file(fd
, offset
);
1241 err
= os_read_file(fd
, buf
, len
);
1248 int open_ubd_file(char *file
, struct openflags
*openflags
, int shared
,
1249 char **backing_file_out
, int *bitmap_offset_out
,
1250 unsigned long *bitmap_len_out
, int *data_offset_out
,
1251 int *create_cow_out
)
1254 unsigned long long size
;
1255 __u32 version
, align
;
1257 int fd
, err
, sectorsize
, asked_switch
, mode
= 0644;
1259 fd
= os_open_file(file
, *openflags
, mode
);
1261 if ((fd
== -ENOENT
) && (create_cow_out
!= NULL
))
1262 *create_cow_out
= 1;
1263 if (!openflags
->w
||
1264 ((fd
!= -EROFS
) && (fd
!= -EACCES
)))
1267 fd
= os_open_file(file
, *openflags
, mode
);
1273 printk("Not locking \"%s\" on the host\n", file
);
1275 err
= os_lock_file(fd
, openflags
->w
);
1277 printk("Failed to lock '%s', err = %d\n", file
, -err
);
1282 /* Successful return case! */
1283 if(backing_file_out
== NULL
)
1286 err
= read_cow_header(file_reader
, &fd
, &version
, &backing_file
, &mtime
,
1287 &size
, §orsize
, &align
, bitmap_offset_out
);
1288 if(err
&& (*backing_file_out
!= NULL
)){
1289 printk("Failed to read COW header from COW file \"%s\", "
1290 "errno = %d\n", file
, -err
);
1296 asked_switch
= path_requires_switch(*backing_file_out
, backing_file
, file
);
1298 /* Allow switching only if no mismatch. */
1299 if (asked_switch
&& !backing_file_mismatch(*backing_file_out
, size
, mtime
)) {
1300 printk("Switching backing file to '%s'\n", *backing_file_out
);
1301 err
= write_cow_header(file
, fd
, *backing_file_out
,
1302 sectorsize
, align
, &size
);
1304 printk("Switch failed, errno = %d\n", -err
);
1308 *backing_file_out
= backing_file
;
1309 err
= backing_file_mismatch(*backing_file_out
, size
, mtime
);
1314 cow_sizes(version
, size
, sectorsize
, align
, *bitmap_offset_out
,
1315 bitmap_len_out
, data_offset_out
);
1323 int create_cow_file(char *cow_file
, char *backing_file
, struct openflags flags
,
1324 int sectorsize
, int alignment
, int *bitmap_offset_out
,
1325 unsigned long *bitmap_len_out
, int *data_offset_out
)
1330 fd
= open_ubd_file(cow_file
, &flags
, 0, NULL
, NULL
, NULL
, NULL
, NULL
);
1333 printk("Open of COW file '%s' failed, errno = %d\n", cow_file
,
1338 err
= init_cow_file(fd
, cow_file
, backing_file
, sectorsize
, alignment
,
1339 bitmap_offset_out
, bitmap_len_out
,
1348 static int update_bitmap(struct io_thread_req
*req
)
1352 if(req
->cow_offset
== -1)
1355 n
= os_seek_file(req
->fds
[1], req
->cow_offset
);
1357 printk("do_io - bitmap lseek failed : err = %d\n", -n
);
1361 n
= os_write_file(req
->fds
[1], &req
->bitmap_words
,
1362 sizeof(req
->bitmap_words
));
1363 if(n
!= sizeof(req
->bitmap_words
)){
1364 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n
,
1372 void do_io(struct io_thread_req
*req
)
1376 int n
, nsectors
, start
, end
, bit
;
1380 nsectors
= req
->length
/ req
->sectorsize
;
1383 bit
= ubd_test_bit(start
, (unsigned char *) &req
->sector_mask
);
1385 while((end
< nsectors
) &&
1386 (ubd_test_bit(end
, (unsigned char *)
1387 &req
->sector_mask
) == bit
))
1390 off
= req
->offset
+ req
->offsets
[bit
] +
1391 start
* req
->sectorsize
;
1392 len
= (end
- start
) * req
->sectorsize
;
1393 buf
= &req
->buffer
[start
* req
->sectorsize
];
1395 err
= os_seek_file(req
->fds
[bit
], off
);
1397 printk("do_io - lseek failed : err = %d\n", -err
);
1401 if(req
->op
== UBD_READ
){
1406 n
= os_read_file(req
->fds
[bit
], buf
, len
);
1408 printk("do_io - read failed, err = %d "
1409 "fd = %d\n", -n
, req
->fds
[bit
]);
1413 } while((n
< len
) && (n
!= 0));
1414 if (n
< len
) memset(&buf
[n
], 0, len
- n
);
1416 n
= os_write_file(req
->fds
[bit
], buf
, len
);
1418 printk("do_io - write failed err = %d "
1419 "fd = %d\n", -n
, req
->fds
[bit
]);
1426 } while(start
< nsectors
);
1428 req
->error
= update_bitmap(req
);
1431 /* Changed in start_io_thread, which is serialized by being called only
1432 * from ubd_init, which is an initcall.
1436 /* Only changed by the io thread. XXX: currently unused. */
1437 static int io_count
= 0;
1439 int io_thread(void *arg
)
1441 struct io_thread_req
*req
;
1444 ignore_sigwinch_sig();
1446 n
= os_read_file(kernel_fd
, &req
,
1447 sizeof(struct io_thread_req
*));
1448 if(n
!= sizeof(struct io_thread_req
*)){
1450 printk("io_thread - read failed, fd = %d, "
1451 "err = %d\n", kernel_fd
, -n
);
1453 printk("io_thread - short read, fd = %d, "
1454 "length = %d\n", kernel_fd
, n
);
1460 n
= os_write_file(kernel_fd
, &req
,
1461 sizeof(struct io_thread_req
*));
1462 if(n
!= sizeof(struct io_thread_req
*))
1463 printk("io_thread - write failed, fd = %d, err = %d\n",