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/module.h"
24 #include "linux/blkdev.h"
25 #include "linux/hdreg.h"
26 #include "linux/init.h"
27 #include "linux/cdrom.h"
28 #include "linux/proc_fs.h"
29 #include "linux/ctype.h"
30 #include "linux/capability.h"
32 #include "linux/vmalloc.h"
33 #include "linux/blkpg.h"
34 #include "linux/genhd.h"
35 #include "linux/spinlock.h"
36 #include "linux/platform_device.h"
37 #include "asm/segment.h"
38 #include "asm/uaccess.h"
40 #include "asm/types.h"
41 #include "asm/tlbflush.h"
43 #include "kern_util.h"
45 #include "mconsole_kern.h"
55 enum ubd_req
{ UBD_READ
, UBD_WRITE
};
57 struct io_thread_req
{
61 unsigned long offsets
[2];
62 unsigned long long offset
;
66 unsigned long sector_mask
;
67 unsigned long long cow_offset
;
68 unsigned long bitmap_words
[2];
72 extern int open_ubd_file(char *file
, struct openflags
*openflags
, int shared
,
73 char **backing_file_out
, int *bitmap_offset_out
,
74 unsigned long *bitmap_len_out
, int *data_offset_out
,
76 extern int create_cow_file(char *cow_file
, char *backing_file
,
77 struct openflags flags
, int sectorsize
,
78 int alignment
, int *bitmap_offset_out
,
79 unsigned long *bitmap_len_out
,
80 int *data_offset_out
);
81 extern int read_cow_bitmap(int fd
, void *buf
, int offset
, int len
);
82 extern void do_io(struct io_thread_req
*req
);
84 static inline int ubd_test_bit(__u64 bit
, unsigned char *data
)
89 bits
= sizeof(data
[0]) * 8;
92 return((data
[n
] & (1 << off
)) != 0);
95 static inline void ubd_set_bit(__u64 bit
, unsigned char *data
)
100 bits
= sizeof(data
[0]) * 8;
103 data
[n
] |= (1 << off
);
105 /*End stuff from ubd_user.h*/
107 #define DRIVER_NAME "uml-blkdev"
109 static DEFINE_MUTEX(ubd_lock
);
111 static int ubd_open(struct inode
* inode
, struct file
* filp
);
112 static int ubd_release(struct inode
* inode
, struct file
* file
);
113 static int ubd_ioctl(struct inode
* inode
, struct file
* file
,
114 unsigned int cmd
, unsigned long arg
);
115 static int ubd_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
);
119 static struct block_device_operations ubd_blops
= {
120 .owner
= THIS_MODULE
,
122 .release
= ubd_release
,
124 .getgeo
= ubd_getgeo
,
127 /* Protected by ubd_lock */
128 static int fake_major
= MAJOR_NR
;
129 static struct gendisk
*ubd_gendisk
[MAX_DEV
];
130 static struct gendisk
*fake_gendisk
[MAX_DEV
];
132 #ifdef CONFIG_BLK_DEV_UBD_SYNC
133 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
136 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
139 static struct openflags global_openflags
= OPEN_FLAGS
;
142 /* backing file name */
144 /* backing file fd */
146 unsigned long *bitmap
;
147 unsigned long bitmap_len
;
153 /* name (and fd, below) of the file opened for writing, either the
154 * backing or the cow file. */
159 struct openflags boot_openflags
;
160 struct openflags openflags
;
164 struct platform_device pdev
;
165 struct request_queue
*queue
;
170 #define DEFAULT_COW { \
174 .bitmap_offset = 0, \
178 #define DEFAULT_UBD { \
183 .boot_openflags = OPEN_FLAGS, \
184 .openflags = OPEN_FLAGS, \
187 .cow = DEFAULT_COW, \
188 .lock = SPIN_LOCK_UNLOCKED, \
192 /* Protected by ubd_lock */
193 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 proc_ide_read_media(char *page
, char **start
, off_t off
, int count
,
207 int *eof
, void *data
)
211 strcpy(page
, "disk\n");
212 len
= strlen("disk\n");
216 if (len
<= 0) return 0;
223 static void make_ide_entries(char *dev_name
)
225 struct proc_dir_entry
*dir
, *ent
;
228 if(proc_ide_root
== NULL
) make_proc_ide();
230 dir
= proc_mkdir(dev_name
, proc_ide
);
233 ent
= create_proc_entry("media", S_IFREG
|S_IRUGO
, dir
);
236 ent
->read_proc
= proc_ide_read_media
;
237 ent
->write_proc
= NULL
;
238 sprintf(name
,"ide0/%s", dev_name
);
239 proc_symlink(dev_name
, proc_ide_root
, name
);
242 static int fake_ide_setup(char *str
)
248 __setup("fake_ide", fake_ide_setup
);
250 __uml_help(fake_ide_setup
,
252 " Create ide0 entries that map onto ubd devices.\n\n"
255 static int parse_unit(char **ptr
)
257 char *str
= *ptr
, *end
;
261 n
= simple_strtoul(str
, &end
, 0);
266 else if (('a' <= *str
) && (*str
<= 'z')) {
274 /* If *index_out == -1 at exit, the passed option was a general one;
275 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
276 * should not be freed on exit.
278 static int ubd_setup_common(char *str
, int *index_out
, char **error_out
)
281 struct openflags flags
= global_openflags
;
285 if(index_out
) *index_out
= -1;
292 if(!strcmp(str
, "sync")){
293 global_openflags
= of_sync(global_openflags
);
298 major
= simple_strtoul(str
, &end
, 0);
299 if((*end
!= '\0') || (end
== str
)){
300 *error_out
= "Didn't parse major number";
304 mutex_lock(&ubd_lock
);
305 if(fake_major
!= MAJOR_NR
){
306 *error_out
= "Can't assign a fake major twice";
312 printk(KERN_INFO
"Setting extra ubd major number to %d\n",
316 mutex_unlock(&ubd_lock
);
320 n
= parse_unit(&str
);
322 *error_out
= "Couldn't parse device number";
326 *error_out
= "Device number out of range";
331 mutex_lock(&ubd_lock
);
333 ubd_dev
= &ubd_devs
[n
];
334 if(ubd_dev
->file
!= NULL
){
335 *error_out
= "Device is already configured";
343 for (i
= 0; i
< sizeof("rscd="); i
++) {
361 *error_out
= "Expected '=' or flag letter "
369 *error_out
= "Too many flags specified";
371 *error_out
= "Missing '='";
375 backing_file
= strchr(str
, ',');
377 if (backing_file
== NULL
)
378 backing_file
= strchr(str
, ':');
380 if(backing_file
!= NULL
){
382 *error_out
= "Can't specify both 'd' and a cow file";
386 *backing_file
= '\0';
392 ubd_dev
->cow
.file
= backing_file
;
393 ubd_dev
->boot_openflags
= flags
;
395 mutex_unlock(&ubd_lock
);
399 static int ubd_setup(char *str
)
404 err
= ubd_setup_common(str
, NULL
, &error
);
406 printk(KERN_ERR
"Failed to initialize device with \"%s\" : "
411 __setup("ubd", ubd_setup
);
412 __uml_help(ubd_setup
,
413 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
414 " This is used to associate a device with a file in the underlying\n"
415 " filesystem. When specifying two filenames, the first one is the\n"
416 " COW name and the second is the backing file name. As separator you can\n"
417 " use either a ':' or a ',': the first one allows writing things like;\n"
418 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
419 " while with a ',' the shell would not expand the 2nd '~'.\n"
420 " When using only one filename, UML will detect whether to treat it like\n"
421 " a COW file or a backing file. To override this detection, add the 'd'\n"
423 " ubd0d=BackingFile\n"
424 " Usually, there is a filesystem in the file, but \n"
425 " that's not required. Swap devices containing swap files can be\n"
426 " specified like this. Also, a file which doesn't contain a\n"
427 " filesystem can have its contents read in the virtual \n"
428 " machine by running 'dd' on the device. <n> must be in the range\n"
429 " 0 to 7. Appending an 'r' to the number will cause that device\n"
430 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
431 " an 's' will cause data to be written to disk on the host immediately.\n\n"
434 static int udb_setup(char *str
)
436 printk("udb%s specified on command line is almost certainly a ubd -> "
441 __setup("udb", udb_setup
);
442 __uml_help(udb_setup
,
444 " This option is here solely to catch ubd -> udb typos, which can be\n"
445 " to impossible to catch visually unless you specifically look for\n"
446 " them. The only result of any option starting with 'udb' is an error\n"
447 " in the boot output.\n\n"
450 static int fakehd_set
= 0;
451 static int fakehd(char *str
)
453 printk(KERN_INFO
"fakehd : Changing ubd name to \"hd\".\n");
458 __setup("fakehd", fakehd
);
461 " Change the ubd device name to \"hd\".\n\n"
464 static void do_ubd_request(request_queue_t
* q
);
466 /* Only changed by ubd_init, which is an initcall. */
469 /* call ubd_finish if you need to serialize */
470 static void __ubd_finish(struct request
*req
, int error
)
478 nsect
= req
->current_nr_sectors
;
479 req
->sector
+= nsect
;
480 req
->buffer
+= nsect
<< 9;
482 req
->nr_sectors
-= nsect
;
483 req
->current_nr_sectors
= 0;
487 /* Callable only from interrupt context - otherwise you need to do
488 * spin_lock_irq()/spin_lock_irqsave() */
489 static inline void ubd_finish(struct request
*req
, int error
)
491 struct ubd
*dev
= req
->rq_disk
->private_data
;
493 spin_lock(&dev
->lock
);
494 __ubd_finish(req
, error
);
495 spin_unlock(&dev
->lock
);
498 /* XXX - move this inside ubd_intr. */
499 /* Called without dev->lock held, and only in interrupt context. */
500 static void ubd_handler(void)
502 struct io_thread_req req
;
507 n
= os_read_file_k(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
);
515 dev
= rq
->rq_disk
->private_data
;
518 ubd_finish(rq
, req
.error
);
519 reactivate_fd(thread_fd
, UBD_IRQ
);
520 spin_lock(&dev
->lock
);
521 do_ubd_request(dev
->queue
);
522 spin_unlock(&dev
->lock
);
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
){
599 ubd_dev
->cow
.bitmap
= (void *) vmalloc(ubd_dev
->cow
.bitmap_len
);
600 if(ubd_dev
->cow
.bitmap
== NULL
){
601 printk(KERN_ERR
"Failed to vmalloc COW bitmap\n");
604 flush_tlb_kernel_vm();
606 err
= read_cow_bitmap(ubd_dev
->fd
, ubd_dev
->cow
.bitmap
,
607 ubd_dev
->cow
.bitmap_offset
,
608 ubd_dev
->cow
.bitmap_len
);
612 flags
= ubd_dev
->openflags
;
614 err
= open_ubd_file(ubd_dev
->cow
.file
, &flags
, ubd_dev
->shared
, NULL
,
615 NULL
, NULL
, NULL
, NULL
);
616 if(err
< 0) goto error
;
617 ubd_dev
->cow
.fd
= err
;
621 os_close_file(ubd_dev
->fd
);
625 static void ubd_device_release(struct device
*dev
)
627 struct ubd
*ubd_dev
= dev
->driver_data
;
629 blk_cleanup_queue(ubd_dev
->queue
);
630 *ubd_dev
= ((struct ubd
) DEFAULT_UBD
);
633 static int ubd_disk_register(int major
, u64 size
, int unit
,
634 struct gendisk
**disk_out
)
636 struct gendisk
*disk
;
638 disk
= alloc_disk(1 << UBD_SHIFT
);
643 disk
->first_minor
= unit
<< UBD_SHIFT
;
644 disk
->fops
= &ubd_blops
;
645 set_capacity(disk
, size
/ 512);
646 if(major
== MAJOR_NR
)
647 sprintf(disk
->disk_name
, "ubd%c", 'a' + unit
);
649 sprintf(disk
->disk_name
, "ubd_fake%d", unit
);
651 /* sysfs register (not for ide fake devices) */
652 if (major
== MAJOR_NR
) {
653 ubd_devs
[unit
].pdev
.id
= unit
;
654 ubd_devs
[unit
].pdev
.name
= DRIVER_NAME
;
655 ubd_devs
[unit
].pdev
.dev
.release
= ubd_device_release
;
656 ubd_devs
[unit
].pdev
.dev
.driver_data
= &ubd_devs
[unit
];
657 platform_device_register(&ubd_devs
[unit
].pdev
);
658 disk
->driverfs_dev
= &ubd_devs
[unit
].pdev
.dev
;
661 disk
->private_data
= &ubd_devs
[unit
];
662 disk
->queue
= ubd_devs
[unit
].queue
;
669 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
671 static int ubd_add(int n
, char **error_out
)
673 struct ubd
*ubd_dev
= &ubd_devs
[n
];
676 if(ubd_dev
->file
== NULL
)
679 err
= ubd_file_size(ubd_dev
, &ubd_dev
->size
);
681 *error_out
= "Couldn't determine size of device's file";
685 ubd_dev
->size
= ROUND_BLOCK(ubd_dev
->size
);
688 ubd_dev
->queue
= blk_init_queue(do_ubd_request
, &ubd_dev
->lock
);
689 if (ubd_dev
->queue
== NULL
) {
690 *error_out
= "Failed to initialize device queue";
693 ubd_dev
->queue
->queuedata
= ubd_dev
;
695 err
= ubd_disk_register(MAJOR_NR
, ubd_dev
->size
, n
, &ubd_gendisk
[n
]);
697 *error_out
= "Failed to register device";
701 if(fake_major
!= MAJOR_NR
)
702 ubd_disk_register(fake_major
, ubd_dev
->size
, n
,
705 /* perhaps this should also be under the "if (fake_major)" above */
706 /* using the fake_disk->disk_name and also the fakehd_set name */
708 make_ide_entries(ubd_gendisk
[n
]->disk_name
);
715 blk_cleanup_queue(ubd_dev
->queue
);
719 static int ubd_config(char *str
, char **error_out
)
723 /* This string is possibly broken up and stored, so it's only
724 * freed if ubd_setup_common fails, or if only general options
727 str
= kstrdup(str
, GFP_KERNEL
);
729 *error_out
= "Failed to allocate memory";
733 ret
= ubd_setup_common(str
, &n
, error_out
);
742 mutex_lock(&ubd_lock
);
743 ret
= ubd_add(n
, error_out
);
745 ubd_devs
[n
].file
= NULL
;
746 mutex_unlock(&ubd_lock
);
756 static int ubd_get_config(char *name
, char *str
, int size
, char **error_out
)
761 n
= parse_unit(&name
);
762 if((n
>= MAX_DEV
) || (n
< 0)){
763 *error_out
= "ubd_get_config : device number out of range";
767 ubd_dev
= &ubd_devs
[n
];
768 mutex_lock(&ubd_lock
);
770 if(ubd_dev
->file
== NULL
){
771 CONFIG_CHUNK(str
, size
, len
, "", 1);
775 CONFIG_CHUNK(str
, size
, len
, ubd_dev
->file
, 0);
777 if(ubd_dev
->cow
.file
!= NULL
){
778 CONFIG_CHUNK(str
, size
, len
, ",", 0);
779 CONFIG_CHUNK(str
, size
, len
, ubd_dev
->cow
.file
, 1);
781 else CONFIG_CHUNK(str
, size
, len
, "", 1);
784 mutex_unlock(&ubd_lock
);
788 static int ubd_id(char **str
, int *start_out
, int *end_out
)
794 *end_out
= MAX_DEV
- 1;
798 static int ubd_remove(int n
, char **error_out
)
800 struct gendisk
*disk
= ubd_gendisk
[n
];
804 mutex_lock(&ubd_lock
);
806 ubd_dev
= &ubd_devs
[n
];
808 if(ubd_dev
->file
== NULL
)
811 /* you cannot remove a open disk */
813 if(ubd_dev
->count
> 0)
816 ubd_gendisk
[n
] = NULL
;
822 if(fake_gendisk
[n
] != NULL
){
823 del_gendisk(fake_gendisk
[n
]);
824 put_disk(fake_gendisk
[n
]);
825 fake_gendisk
[n
] = NULL
;
829 platform_device_unregister(&ubd_dev
->pdev
);
831 mutex_unlock(&ubd_lock
);
835 /* All these are called by mconsole in process context and without
836 * ubd-specific locks. The structure itself is const except for .list.
838 static struct mc_device ubd_mc
= {
839 .list
= LIST_HEAD_INIT(ubd_mc
.list
),
841 .config
= ubd_config
,
842 .get_config
= ubd_get_config
,
844 .remove
= ubd_remove
,
847 static int __init
ubd_mc_init(void)
849 mconsole_register_dev(&ubd_mc
);
853 __initcall(ubd_mc_init
);
855 static int __init
ubd0_init(void)
857 struct ubd
*ubd_dev
= &ubd_devs
[0];
859 mutex_lock(&ubd_lock
);
860 if(ubd_dev
->file
== NULL
)
861 ubd_dev
->file
= "root_fs";
862 mutex_unlock(&ubd_lock
);
867 __initcall(ubd0_init
);
869 /* Used in ubd_init, which is an initcall */
870 static struct platform_driver ubd_driver
= {
876 static int __init
ubd_init(void)
881 if (register_blkdev(MAJOR_NR
, "ubd"))
884 if (fake_major
!= MAJOR_NR
) {
885 char name
[sizeof("ubd_nnn\0")];
887 snprintf(name
, sizeof(name
), "ubd_%d", fake_major
);
888 if (register_blkdev(fake_major
, "ubd"))
891 platform_driver_register(&ubd_driver
);
892 mutex_lock(&ubd_lock
);
893 for (i
= 0; i
< MAX_DEV
; i
++){
894 err
= ubd_add(i
, &error
);
896 printk(KERN_ERR
"Failed to initialize ubd device %d :"
899 mutex_unlock(&ubd_lock
);
903 late_initcall(ubd_init
);
905 static int __init
ubd_driver_init(void){
909 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
910 if(global_openflags
.s
){
911 printk(KERN_INFO
"ubd: Synchronous mode\n");
912 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
913 * enough. So use anyway the io thread. */
915 stack
= alloc_stack(0, 0);
916 io_pid
= start_io_thread(stack
+ PAGE_SIZE
- sizeof(void *),
920 "ubd : Failed to start I/O thread (errno = %d) - "
921 "falling back to synchronous I/O\n", -io_pid
);
925 err
= um_request_irq(UBD_IRQ
, thread_fd
, IRQ_READ
, ubd_intr
,
926 IRQF_DISABLED
, "ubd", ubd_devs
);
928 printk(KERN_ERR
"um_request_irq failed - errno = %d\n", -err
);
932 device_initcall(ubd_driver_init
);
934 static int ubd_open(struct inode
*inode
, struct file
*filp
)
936 struct gendisk
*disk
= inode
->i_bdev
->bd_disk
;
937 struct ubd
*ubd_dev
= disk
->private_data
;
940 if(ubd_dev
->count
== 0){
941 err
= ubd_open_dev(ubd_dev
);
943 printk(KERN_ERR
"%s: Can't open \"%s\": errno = %d\n",
944 disk
->disk_name
, ubd_dev
->file
, -err
);
949 set_disk_ro(disk
, !ubd_dev
->openflags
.w
);
951 /* This should no more be needed. And it didn't work anyway to exclude
952 * read-write remounting of filesystems.*/
953 /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
954 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
961 static int ubd_release(struct inode
* inode
, struct file
* file
)
963 struct gendisk
*disk
= inode
->i_bdev
->bd_disk
;
964 struct ubd
*ubd_dev
= disk
->private_data
;
966 if(--ubd_dev
->count
== 0)
967 ubd_close_dev(ubd_dev
);
971 static void cowify_bitmap(__u64 io_offset
, int length
, unsigned long *cow_mask
,
972 __u64
*cow_offset
, unsigned long *bitmap
,
973 __u64 bitmap_offset
, unsigned long *bitmap_words
,
976 __u64 sector
= io_offset
>> 9;
977 int i
, update_bitmap
= 0;
979 for(i
= 0; i
< length
>> 9; i
++){
981 ubd_set_bit(i
, (unsigned char *) cow_mask
);
982 if(ubd_test_bit(sector
+ i
, (unsigned char *) bitmap
))
986 ubd_set_bit(sector
+ i
, (unsigned char *) bitmap
);
992 *cow_offset
= sector
/ (sizeof(unsigned long) * 8);
994 /* This takes care of the case where we're exactly at the end of the
995 * device, and *cow_offset + 1 is off the end. So, just back it up
996 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
997 * for the original diagnosis.
999 if(*cow_offset
== ((bitmap_len
+ sizeof(unsigned long) - 1) /
1000 sizeof(unsigned long) - 1))
1003 bitmap_words
[0] = bitmap
[*cow_offset
];
1004 bitmap_words
[1] = bitmap
[*cow_offset
+ 1];
1006 *cow_offset
*= sizeof(unsigned long);
1007 *cow_offset
+= bitmap_offset
;
1010 static void cowify_req(struct io_thread_req
*req
, unsigned long *bitmap
,
1011 __u64 bitmap_offset
, __u64 bitmap_len
)
1013 __u64 sector
= req
->offset
>> 9;
1016 if(req
->length
> (sizeof(req
->sector_mask
) * 8) << 9)
1017 panic("Operation too long");
1019 if(req
->op
== UBD_READ
) {
1020 for(i
= 0; i
< req
->length
>> 9; i
++){
1021 if(ubd_test_bit(sector
+ i
, (unsigned char *) bitmap
))
1022 ubd_set_bit(i
, (unsigned char *)
1026 else cowify_bitmap(req
->offset
, req
->length
, &req
->sector_mask
,
1027 &req
->cow_offset
, bitmap
, bitmap_offset
,
1028 req
->bitmap_words
, bitmap_len
);
1031 /* Called with dev->lock held */
1032 static int prepare_request(struct request
*req
, struct io_thread_req
*io_req
)
1034 struct gendisk
*disk
= req
->rq_disk
;
1035 struct ubd
*ubd_dev
= disk
->private_data
;
1039 /* This should be impossible now */
1040 if((rq_data_dir(req
) == WRITE
) && !ubd_dev
->openflags
.w
){
1041 printk("Write attempted on readonly ubd device %s\n",
1043 end_request(req
, 0);
1047 offset
= ((__u64
) req
->sector
) << 9;
1048 len
= req
->current_nr_sectors
<< 9;
1051 io_req
->fds
[0] = (ubd_dev
->cow
.file
!= NULL
) ? ubd_dev
->cow
.fd
: ubd_dev
->fd
;
1052 io_req
->fds
[1] = ubd_dev
->fd
;
1053 io_req
->cow_offset
= -1;
1054 io_req
->offset
= offset
;
1055 io_req
->length
= len
;
1057 io_req
->sector_mask
= 0;
1059 io_req
->op
= (rq_data_dir(req
) == READ
) ? UBD_READ
: UBD_WRITE
;
1060 io_req
->offsets
[0] = 0;
1061 io_req
->offsets
[1] = ubd_dev
->cow
.data_offset
;
1062 io_req
->buffer
= req
->buffer
;
1063 io_req
->sectorsize
= 1 << 9;
1065 if(ubd_dev
->cow
.file
!= NULL
)
1066 cowify_req(io_req
, ubd_dev
->cow
.bitmap
, ubd_dev
->cow
.bitmap_offset
,
1067 ubd_dev
->cow
.bitmap_len
);
1072 /* Called with dev->lock held */
1073 static void do_ubd_request(request_queue_t
*q
)
1075 struct io_thread_req io_req
;
1076 struct request
*req
;
1079 if(thread_fd
== -1){
1080 while((req
= elv_next_request(q
)) != NULL
){
1081 err
= prepare_request(req
, &io_req
);
1084 __ubd_finish(req
, io_req
.error
);
1089 struct ubd
*dev
= q
->queuedata
;
1090 if(dev
->active
|| (req
= elv_next_request(q
)) == NULL
)
1092 err
= prepare_request(req
, &io_req
);
1095 n
= os_write_file_k(thread_fd
, &io_req
, sizeof(io_req
));
1096 if(n
!= sizeof(io_req
))
1097 printk("write to io thread failed, "
1098 "errno = %d\n", -n
);
1103 static int ubd_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1105 struct ubd
*ubd_dev
= bdev
->bd_disk
->private_data
;
1109 geo
->cylinders
= ubd_dev
->size
/ (128 * 32 * 512);
1113 static int ubd_ioctl(struct inode
* inode
, struct file
* file
,
1114 unsigned int cmd
, unsigned long arg
)
1116 struct ubd
*ubd_dev
= inode
->i_bdev
->bd_disk
->private_data
;
1117 struct hd_driveid ubd_id
= {
1124 struct cdrom_volctrl volume
;
1125 case HDIO_GET_IDENTITY
:
1126 ubd_id
.cyls
= ubd_dev
->size
/ (128 * 32 * 512);
1127 if(copy_to_user((char __user
*) arg
, (char *) &ubd_id
,
1133 if(copy_from_user(&volume
, (char __user
*) arg
, sizeof(volume
)))
1135 volume
.channel0
= 255;
1136 volume
.channel1
= 255;
1137 volume
.channel2
= 255;
1138 volume
.channel3
= 255;
1139 if(copy_to_user((char __user
*) arg
, &volume
, sizeof(volume
)))
1146 static int path_requires_switch(char *from_cmdline
, char *from_cow
, char *cow
)
1148 struct uml_stat buf1
, buf2
;
1151 if(from_cmdline
== NULL
)
1153 if(!strcmp(from_cmdline
, from_cow
))
1156 err
= os_stat_file(from_cmdline
, &buf1
);
1158 printk("Couldn't stat '%s', err = %d\n", from_cmdline
, -err
);
1161 err
= os_stat_file(from_cow
, &buf2
);
1163 printk("Couldn't stat '%s', err = %d\n", from_cow
, -err
);
1166 if((buf1
.ust_dev
== buf2
.ust_dev
) && (buf1
.ust_ino
== buf2
.ust_ino
))
1169 printk("Backing file mismatch - \"%s\" requested,\n"
1170 "\"%s\" specified in COW header of \"%s\"\n",
1171 from_cmdline
, from_cow
, cow
);
1175 static int backing_file_mismatch(char *file
, __u64 size
, time_t mtime
)
1177 unsigned long modtime
;
1178 unsigned long long actual
;
1181 err
= os_file_modtime(file
, &modtime
);
1183 printk("Failed to get modification time of backing file "
1184 "\"%s\", err = %d\n", file
, -err
);
1188 err
= os_file_size(file
, &actual
);
1190 printk("Failed to get size of backing file \"%s\", "
1191 "err = %d\n", file
, -err
);
1196 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1198 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1199 "file\n", (unsigned long long) size
, actual
);
1202 if(modtime
!= mtime
){
1203 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1204 "file\n", mtime
, modtime
);
1210 int read_cow_bitmap(int fd
, void *buf
, int offset
, int len
)
1214 err
= os_seek_file(fd
, offset
);
1218 err
= os_read_file(fd
, buf
, len
);
1225 int open_ubd_file(char *file
, struct openflags
*openflags
, int shared
,
1226 char **backing_file_out
, int *bitmap_offset_out
,
1227 unsigned long *bitmap_len_out
, int *data_offset_out
,
1228 int *create_cow_out
)
1231 unsigned long long size
;
1232 __u32 version
, align
;
1234 int fd
, err
, sectorsize
, asked_switch
, mode
= 0644;
1236 fd
= os_open_file(file
, *openflags
, mode
);
1238 if ((fd
== -ENOENT
) && (create_cow_out
!= NULL
))
1239 *create_cow_out
= 1;
1240 if (!openflags
->w
||
1241 ((fd
!= -EROFS
) && (fd
!= -EACCES
)))
1244 fd
= os_open_file(file
, *openflags
, mode
);
1250 printk("Not locking \"%s\" on the host\n", file
);
1252 err
= os_lock_file(fd
, openflags
->w
);
1254 printk("Failed to lock '%s', err = %d\n", file
, -err
);
1259 /* Successful return case! */
1260 if(backing_file_out
== NULL
)
1263 err
= read_cow_header(file_reader
, &fd
, &version
, &backing_file
, &mtime
,
1264 &size
, §orsize
, &align
, bitmap_offset_out
);
1265 if(err
&& (*backing_file_out
!= NULL
)){
1266 printk("Failed to read COW header from COW file \"%s\", "
1267 "errno = %d\n", file
, -err
);
1273 asked_switch
= path_requires_switch(*backing_file_out
, backing_file
, file
);
1275 /* Allow switching only if no mismatch. */
1276 if (asked_switch
&& !backing_file_mismatch(*backing_file_out
, size
, mtime
)) {
1277 printk("Switching backing file to '%s'\n", *backing_file_out
);
1278 err
= write_cow_header(file
, fd
, *backing_file_out
,
1279 sectorsize
, align
, &size
);
1281 printk("Switch failed, errno = %d\n", -err
);
1285 *backing_file_out
= backing_file
;
1286 err
= backing_file_mismatch(*backing_file_out
, size
, mtime
);
1291 cow_sizes(version
, size
, sectorsize
, align
, *bitmap_offset_out
,
1292 bitmap_len_out
, data_offset_out
);
1300 int create_cow_file(char *cow_file
, char *backing_file
, struct openflags flags
,
1301 int sectorsize
, int alignment
, int *bitmap_offset_out
,
1302 unsigned long *bitmap_len_out
, int *data_offset_out
)
1307 fd
= open_ubd_file(cow_file
, &flags
, 0, NULL
, NULL
, NULL
, NULL
, NULL
);
1310 printk("Open of COW file '%s' failed, errno = %d\n", cow_file
,
1315 err
= init_cow_file(fd
, cow_file
, backing_file
, sectorsize
, alignment
,
1316 bitmap_offset_out
, bitmap_len_out
,
1325 static int update_bitmap(struct io_thread_req
*req
)
1329 if(req
->cow_offset
== -1)
1332 n
= os_seek_file(req
->fds
[1], req
->cow_offset
);
1334 printk("do_io - bitmap lseek failed : err = %d\n", -n
);
1338 n
= os_write_file_k(req
->fds
[1], &req
->bitmap_words
,
1339 sizeof(req
->bitmap_words
));
1340 if(n
!= sizeof(req
->bitmap_words
)){
1341 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n
,
1349 void do_io(struct io_thread_req
*req
)
1353 int n
, nsectors
, start
, end
, bit
;
1357 nsectors
= req
->length
/ req
->sectorsize
;
1360 bit
= ubd_test_bit(start
, (unsigned char *) &req
->sector_mask
);
1362 while((end
< nsectors
) &&
1363 (ubd_test_bit(end
, (unsigned char *)
1364 &req
->sector_mask
) == bit
))
1367 off
= req
->offset
+ req
->offsets
[bit
] +
1368 start
* req
->sectorsize
;
1369 len
= (end
- start
) * req
->sectorsize
;
1370 buf
= &req
->buffer
[start
* req
->sectorsize
];
1372 err
= os_seek_file(req
->fds
[bit
], off
);
1374 printk("do_io - lseek failed : err = %d\n", -err
);
1378 if(req
->op
== UBD_READ
){
1383 n
= os_read_file_k(req
->fds
[bit
], buf
, len
);
1385 printk("do_io - read failed, err = %d "
1386 "fd = %d\n", -n
, req
->fds
[bit
]);
1390 } while((n
< len
) && (n
!= 0));
1391 if (n
< len
) memset(&buf
[n
], 0, len
- n
);
1393 n
= os_write_file_k(req
->fds
[bit
], buf
, len
);
1395 printk("do_io - write failed err = %d "
1396 "fd = %d\n", -n
, req
->fds
[bit
]);
1403 } while(start
< nsectors
);
1405 req
->error
= update_bitmap(req
);
1408 /* Changed in start_io_thread, which is serialized by being called only
1409 * from ubd_init, which is an initcall.
1413 /* Only changed by the io thread. XXX: currently unused. */
1414 static int io_count
= 0;
1416 int io_thread(void *arg
)
1418 struct io_thread_req req
;
1421 ignore_sigwinch_sig();
1423 n
= os_read_file_k(kernel_fd
, &req
, sizeof(req
));
1424 if(n
!= sizeof(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_k(kernel_fd
, &req
, sizeof(req
));
1437 if(n
!= sizeof(req
))
1438 printk("io_thread - write failed, fd = %d, err = %d\n",