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"
57 enum ubd_req { UBD_READ, UBD_WRITE };
59 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, int shared,
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_MUTEX(ubd_lock);
113 static int ubd_open(struct inode * inode, struct file * filp);
114 static int ubd_release(struct inode * inode, struct file * file);
115 static int ubd_ioctl(struct inode * inode, struct file * file,
116 unsigned int cmd, unsigned long arg);
117 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
121 static struct block_device_operations ubd_blops = {
122 .owner = THIS_MODULE,
124 .release = ubd_release,
126 .getgeo = ubd_getgeo,
129 /* Protected by ubd_lock */
130 static int fake_major = MAJOR_NR;
131 static struct gendisk *ubd_gendisk[MAX_DEV];
132 static struct gendisk *fake_gendisk[MAX_DEV];
134 #ifdef CONFIG_BLK_DEV_UBD_SYNC
135 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
138 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
141 static struct openflags global_openflags = OPEN_FLAGS;
144 /* backing file name */
146 /* backing file fd */
148 unsigned long *bitmap;
149 unsigned long bitmap_len;
157 struct list_head restart;
158 /* name (and fd, below) of the file opened for writing, either the
159 * backing or the cow file. */
164 struct openflags boot_openflags;
165 struct openflags openflags;
169 struct platform_device pdev;
170 struct request_queue *queue;
172 struct scatterlist sg[MAX_SG];
173 struct request *request;
174 int start_sg, end_sg;
177 #define DEFAULT_COW { \
181 .bitmap_offset = 0, \
185 #define DEFAULT_UBD { \
190 .boot_openflags = OPEN_FLAGS, \
191 .openflags = OPEN_FLAGS, \
194 .cow = DEFAULT_COW, \
195 .lock = SPIN_LOCK_UNLOCKED, \
201 /* Protected by ubd_lock */
202 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
204 /* Only changed by fake_ide_setup which is a setup */
205 static int fake_ide = 0;
206 static struct proc_dir_entry *proc_ide_root = NULL;
207 static struct proc_dir_entry *proc_ide = NULL;
209 static void make_proc_ide(void)
211 proc_ide_root = proc_mkdir("ide", NULL);
212 proc_ide = proc_mkdir("ide0", proc_ide_root);
215 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
216 int *eof, void *data)
220 strcpy(page, "disk\n");
221 len = strlen("disk\n");
225 if (len <= 0) return 0;
232 static void make_ide_entries(char *dev_name)
234 struct proc_dir_entry *dir, *ent;
237 if(proc_ide_root == NULL) make_proc_ide();
239 dir = proc_mkdir(dev_name, proc_ide);
242 ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
245 ent->read_proc = proc_ide_read_media;
246 ent->write_proc = NULL;
247 sprintf(name,"ide0/%s", dev_name);
248 proc_symlink(dev_name, proc_ide_root, name);
251 static int fake_ide_setup(char *str)
257 __setup("fake_ide", fake_ide_setup);
259 __uml_help(fake_ide_setup,
261 " Create ide0 entries that map onto ubd devices.\n\n"
264 static int parse_unit(char **ptr)
266 char *str = *ptr, *end;
270 n = simple_strtoul(str, &end, 0);
275 else if (('a' <= *str) && (*str <= 'z')) {
283 /* If *index_out == -1 at exit, the passed option was a general one;
284 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
285 * should not be freed on exit.
287 static int ubd_setup_common(char *str, int *index_out, char **error_out)
290 struct openflags flags = global_openflags;
294 if(index_out) *index_out = -1;
301 if(!strcmp(str, "sync")){
302 global_openflags = of_sync(global_openflags);
307 major = simple_strtoul(str, &end, 0);
308 if((*end != '\0') || (end == str)){
309 *error_out = "Didn't parse major number";
313 mutex_lock(&ubd_lock);
314 if(fake_major != MAJOR_NR){
315 *error_out = "Can't assign a fake major twice";
321 printk(KERN_INFO "Setting extra ubd major number to %d\n",
325 mutex_unlock(&ubd_lock);
329 n = parse_unit(&str);
331 *error_out = "Couldn't parse device number";
335 *error_out = "Device number out of range";
340 mutex_lock(&ubd_lock);
342 ubd_dev = &ubd_devs[n];
343 if(ubd_dev->file != NULL){
344 *error_out = "Device is already configured";
352 for (i = 0; i < sizeof("rscd="); i++) {
370 *error_out = "Expected '=' or flag letter "
378 *error_out = "Too many flags specified";
380 *error_out = "Missing '='";
384 backing_file = strchr(str, ',');
386 if (backing_file == NULL)
387 backing_file = strchr(str, ':');
389 if(backing_file != NULL){
391 *error_out = "Can't specify both 'd' and a cow file";
395 *backing_file = '\0';
401 ubd_dev->cow.file = backing_file;
402 ubd_dev->boot_openflags = flags;
404 mutex_unlock(&ubd_lock);
408 static int ubd_setup(char *str)
413 err = ubd_setup_common(str, NULL, &error);
415 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
420 __setup("ubd", ubd_setup);
421 __uml_help(ubd_setup,
422 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
423 " This is used to associate a device with a file in the underlying\n"
424 " filesystem. When specifying two filenames, the first one is the\n"
425 " COW name and the second is the backing file name. As separator you can\n"
426 " use either a ':' or a ',': the first one allows writing things like;\n"
427 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
428 " while with a ',' the shell would not expand the 2nd '~'.\n"
429 " When using only one filename, UML will detect whether to treat it like\n"
430 " a COW file or a backing file. To override this detection, add the 'd'\n"
432 " ubd0d=BackingFile\n"
433 " Usually, there is a filesystem in the file, but \n"
434 " that's not required. Swap devices containing swap files can be\n"
435 " specified like this. Also, a file which doesn't contain a\n"
436 " filesystem can have its contents read in the virtual \n"
437 " machine by running 'dd' on the device. <n> must be in the range\n"
438 " 0 to 7. Appending an 'r' to the number will cause that device\n"
439 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
440 " an 's' will cause data to be written to disk on the host immediately.\n\n"
443 static int udb_setup(char *str)
445 printk("udb%s specified on command line is almost certainly a ubd -> "
450 __setup("udb", udb_setup);
451 __uml_help(udb_setup,
453 " This option is here solely to catch ubd -> udb typos, which can be\n"
454 " to impossible to catch visually unless you specifically look for\n"
455 " them. The only result of any option starting with 'udb' is an error\n"
456 " in the boot output.\n\n"
459 static int fakehd_set = 0;
460 static int fakehd(char *str)
462 printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
467 __setup("fakehd", fakehd);
470 " Change the ubd device name to \"hd\".\n\n"
473 static void do_ubd_request(struct request_queue * q);
475 /* Only changed by ubd_init, which is an initcall. */
478 static void ubd_end_request(struct request *req, int bytes, int error)
480 blk_end_request(req, error, bytes);
483 /* Callable only from interrupt context - otherwise you need to do
484 * spin_lock_irq()/spin_lock_irqsave() */
485 static inline void ubd_finish(struct request *req, int bytes)
488 ubd_end_request(req, 0, -EIO);
491 ubd_end_request(req, bytes, 0);
494 static LIST_HEAD(restart);
496 /* XXX - move this inside ubd_intr. */
497 /* Called without dev->lock held, and only in interrupt context. */
498 static void ubd_handler(void)
500 struct io_thread_req *req;
503 struct list_head *list, *next_ele;
508 n = os_read_file(thread_fd, &req,
509 sizeof(struct io_thread_req *));
510 if(n != sizeof(req)){
513 printk(KERN_ERR "spurious interrupt in ubd_handler, "
519 rq->nr_sectors -= req->length >> 9;
520 if(rq->nr_sectors == 0)
521 ubd_finish(rq, rq->hard_nr_sectors << 9);
524 reactivate_fd(thread_fd, UBD_IRQ);
526 list_for_each_safe(list, next_ele, &restart){
527 ubd = container_of(list, struct ubd, restart);
528 list_del_init(&ubd->restart);
529 spin_lock_irqsave(&ubd->lock, flags);
530 do_ubd_request(ubd->queue);
531 spin_unlock_irqrestore(&ubd->lock, flags);
535 static irqreturn_t ubd_intr(int irq, void *dev)
541 /* Only changed by ubd_init, which is an initcall. */
542 static int io_pid = -1;
544 void kill_io_thread(void)
547 os_kill_process(io_pid, 1);
550 __uml_exitcall(kill_io_thread);
552 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
556 file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
557 return os_file_size(file, size_out);
560 static void ubd_close_dev(struct ubd *ubd_dev)
562 os_close_file(ubd_dev->fd);
563 if(ubd_dev->cow.file == NULL)
566 os_close_file(ubd_dev->cow.fd);
567 vfree(ubd_dev->cow.bitmap);
568 ubd_dev->cow.bitmap = NULL;
571 static int ubd_open_dev(struct ubd *ubd_dev)
573 struct openflags flags;
575 int err, create_cow, *create_ptr;
578 ubd_dev->openflags = ubd_dev->boot_openflags;
580 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
581 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
583 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
584 back_ptr, &ubd_dev->cow.bitmap_offset,
585 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
588 if((fd == -ENOENT) && create_cow){
589 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
590 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
591 &ubd_dev->cow.bitmap_offset,
592 &ubd_dev->cow.bitmap_len,
593 &ubd_dev->cow.data_offset);
595 printk(KERN_INFO "Creating \"%s\" as COW file for "
596 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
601 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
607 if(ubd_dev->cow.file != NULL){
608 blk_queue_max_sectors(ubd_dev->queue, 8 * sizeof(long));
611 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
612 if(ubd_dev->cow.bitmap == NULL){
613 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
616 flush_tlb_kernel_vm();
618 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
619 ubd_dev->cow.bitmap_offset,
620 ubd_dev->cow.bitmap_len);
624 flags = ubd_dev->openflags;
626 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
627 NULL, NULL, NULL, NULL);
628 if(err < 0) goto error;
629 ubd_dev->cow.fd = err;
633 os_close_file(ubd_dev->fd);
637 static void ubd_device_release(struct device *dev)
639 struct ubd *ubd_dev = dev->driver_data;
641 blk_cleanup_queue(ubd_dev->queue);
642 *ubd_dev = ((struct ubd) DEFAULT_UBD);
645 static int ubd_disk_register(int major, u64 size, int unit,
646 struct gendisk **disk_out)
648 struct gendisk *disk;
650 disk = alloc_disk(1 << UBD_SHIFT);
655 disk->first_minor = unit << UBD_SHIFT;
656 disk->fops = &ubd_blops;
657 set_capacity(disk, size / 512);
658 if(major == MAJOR_NR)
659 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
661 sprintf(disk->disk_name, "ubd_fake%d", unit);
663 /* sysfs register (not for ide fake devices) */
664 if (major == MAJOR_NR) {
665 ubd_devs[unit].pdev.id = unit;
666 ubd_devs[unit].pdev.name = DRIVER_NAME;
667 ubd_devs[unit].pdev.dev.release = ubd_device_release;
668 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
669 platform_device_register(&ubd_devs[unit].pdev);
670 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
673 disk->private_data = &ubd_devs[unit];
674 disk->queue = ubd_devs[unit].queue;
681 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
683 static int ubd_add(int n, char **error_out)
685 struct ubd *ubd_dev = &ubd_devs[n];
688 if(ubd_dev->file == NULL)
691 err = ubd_file_size(ubd_dev, &ubd_dev->size);
693 *error_out = "Couldn't determine size of device's file";
697 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
699 INIT_LIST_HEAD(&ubd_dev->restart);
700 sg_init_table(ubd_dev->sg, MAX_SG);
703 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
704 if (ubd_dev->queue == NULL) {
705 *error_out = "Failed to initialize device queue";
708 ubd_dev->queue->queuedata = ubd_dev;
710 blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG);
711 err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
713 *error_out = "Failed to register device";
717 if(fake_major != MAJOR_NR)
718 ubd_disk_register(fake_major, ubd_dev->size, n,
721 /* perhaps this should also be under the "if (fake_major)" above */
722 /* using the fake_disk->disk_name and also the fakehd_set name */
724 make_ide_entries(ubd_gendisk[n]->disk_name);
731 blk_cleanup_queue(ubd_dev->queue);
735 static int ubd_config(char *str, char **error_out)
739 /* This string is possibly broken up and stored, so it's only
740 * freed if ubd_setup_common fails, or if only general options
743 str = kstrdup(str, GFP_KERNEL);
745 *error_out = "Failed to allocate memory";
749 ret = ubd_setup_common(str, &n, error_out);
758 mutex_lock(&ubd_lock);
759 ret = ubd_add(n, error_out);
761 ubd_devs[n].file = NULL;
762 mutex_unlock(&ubd_lock);
772 static int ubd_get_config(char *name, char *str, int size, char **error_out)
777 n = parse_unit(&name);
778 if((n >= MAX_DEV) || (n < 0)){
779 *error_out = "ubd_get_config : device number out of range";
783 ubd_dev = &ubd_devs[n];
784 mutex_lock(&ubd_lock);
786 if(ubd_dev->file == NULL){
787 CONFIG_CHUNK(str, size, len, "", 1);
791 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
793 if(ubd_dev->cow.file != NULL){
794 CONFIG_CHUNK(str, size, len, ",", 0);
795 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
797 else CONFIG_CHUNK(str, size, len, "", 1);
800 mutex_unlock(&ubd_lock);
804 static int ubd_id(char **str, int *start_out, int *end_out)
810 *end_out = MAX_DEV - 1;
814 static int ubd_remove(int n, char **error_out)
816 struct gendisk *disk = ubd_gendisk[n];
820 mutex_lock(&ubd_lock);
822 ubd_dev = &ubd_devs[n];
824 if(ubd_dev->file == NULL)
827 /* you cannot remove a open disk */
829 if(ubd_dev->count > 0)
832 ubd_gendisk[n] = NULL;
838 if(fake_gendisk[n] != NULL){
839 del_gendisk(fake_gendisk[n]);
840 put_disk(fake_gendisk[n]);
841 fake_gendisk[n] = NULL;
845 platform_device_unregister(&ubd_dev->pdev);
847 mutex_unlock(&ubd_lock);
851 /* All these are called by mconsole in process context and without
852 * ubd-specific locks. The structure itself is const except for .list.
854 static struct mc_device ubd_mc = {
855 .list = LIST_HEAD_INIT(ubd_mc.list),
857 .config = ubd_config,
858 .get_config = ubd_get_config,
860 .remove = ubd_remove,
863 static int __init ubd_mc_init(void)
865 mconsole_register_dev(&ubd_mc);
869 __initcall(ubd_mc_init);
871 static int __init ubd0_init(void)
873 struct ubd *ubd_dev = &ubd_devs[0];
875 mutex_lock(&ubd_lock);
876 if(ubd_dev->file == NULL)
877 ubd_dev->file = "root_fs";
878 mutex_unlock(&ubd_lock);
883 __initcall(ubd0_init);
885 /* Used in ubd_init, which is an initcall */
886 static struct platform_driver ubd_driver = {
892 static int __init ubd_init(void)
897 if (register_blkdev(MAJOR_NR, "ubd"))
900 if (fake_major != MAJOR_NR) {
901 char name[sizeof("ubd_nnn\0")];
903 snprintf(name, sizeof(name), "ubd_%d", fake_major);
904 if (register_blkdev(fake_major, "ubd"))
907 platform_driver_register(&ubd_driver);
908 mutex_lock(&ubd_lock);
909 for (i = 0; i < MAX_DEV; i++){
910 err = ubd_add(i, &error);
912 printk(KERN_ERR "Failed to initialize ubd device %d :"
915 mutex_unlock(&ubd_lock);
919 late_initcall(ubd_init);
921 static int __init ubd_driver_init(void){
925 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
926 if(global_openflags.s){
927 printk(KERN_INFO "ubd: Synchronous mode\n");
928 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
929 * enough. So use anyway the io thread. */
931 stack = alloc_stack(0, 0);
932 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
936 "ubd : Failed to start I/O thread (errno = %d) - "
937 "falling back to synchronous I/O\n", -io_pid);
941 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
942 IRQF_DISABLED, "ubd", ubd_devs);
944 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
948 device_initcall(ubd_driver_init);
950 static int ubd_open(struct inode *inode, struct file *filp)
952 struct gendisk *disk = inode->i_bdev->bd_disk;
953 struct ubd *ubd_dev = disk->private_data;
956 if(ubd_dev->count == 0){
957 err = ubd_open_dev(ubd_dev);
959 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
960 disk->disk_name, ubd_dev->file, -err);
965 set_disk_ro(disk, !ubd_dev->openflags.w);
967 /* This should no more be needed. And it didn't work anyway to exclude
968 * read-write remounting of filesystems.*/
969 /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
970 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
977 static int ubd_release(struct inode * inode, struct file * file)
979 struct gendisk *disk = inode->i_bdev->bd_disk;
980 struct ubd *ubd_dev = disk->private_data;
982 if(--ubd_dev->count == 0)
983 ubd_close_dev(ubd_dev);
987 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
988 __u64 *cow_offset, unsigned long *bitmap,
989 __u64 bitmap_offset, unsigned long *bitmap_words,
992 __u64 sector = io_offset >> 9;
993 int i, update_bitmap = 0;
995 for(i = 0; i < length >> 9; i++){
997 ubd_set_bit(i, (unsigned char *) cow_mask);
998 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1002 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1008 *cow_offset = sector / (sizeof(unsigned long) * 8);
1010 /* This takes care of the case where we're exactly at the end of the
1011 * device, and *cow_offset + 1 is off the end. So, just back it up
1012 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1013 * for the original diagnosis.
1015 if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
1016 sizeof(unsigned long) - 1))
1019 bitmap_words[0] = bitmap[*cow_offset];
1020 bitmap_words[1] = bitmap[*cow_offset + 1];
1022 *cow_offset *= sizeof(unsigned long);
1023 *cow_offset += bitmap_offset;
1026 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1027 __u64 bitmap_offset, __u64 bitmap_len)
1029 __u64 sector = req->offset >> 9;
1032 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1033 panic("Operation too long");
1035 if(req->op == UBD_READ) {
1036 for(i = 0; i < req->length >> 9; i++){
1037 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1038 ubd_set_bit(i, (unsigned char *)
1042 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1043 &req->cow_offset, bitmap, bitmap_offset,
1044 req->bitmap_words, bitmap_len);
1047 /* Called with dev->lock held */
1048 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1049 unsigned long long offset, int page_offset,
1050 int len, struct page *page)
1052 struct gendisk *disk = req->rq_disk;
1053 struct ubd *ubd_dev = disk->private_data;
1056 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1058 io_req->fds[1] = ubd_dev->fd;
1059 io_req->cow_offset = -1;
1060 io_req->offset = offset;
1061 io_req->length = len;
1063 io_req->sector_mask = 0;
1065 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1066 io_req->offsets[0] = 0;
1067 io_req->offsets[1] = ubd_dev->cow.data_offset;
1068 io_req->buffer = page_address(page) + page_offset;
1069 io_req->sectorsize = 1 << 9;
1071 if(ubd_dev->cow.file != NULL)
1072 cowify_req(io_req, ubd_dev->cow.bitmap,
1073 ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1077 /* Called with dev->lock held */
1078 static void do_ubd_request(struct request_queue *q)
1080 struct io_thread_req *io_req;
1081 struct request *req;
1082 int n, last_sectors;
1085 struct ubd *dev = q->queuedata;
1086 if(dev->end_sg == 0){
1087 struct request *req = elv_next_request(q);
1092 blkdev_dequeue_request(req);
1094 dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1099 while(dev->start_sg < dev->end_sg){
1100 struct scatterlist *sg = &dev->sg[dev->start_sg];
1102 req->sector += last_sectors;
1103 io_req = kmalloc(sizeof(struct io_thread_req),
1106 if(list_empty(&dev->restart))
1107 list_add(&dev->restart, &restart);
1110 prepare_request(req, io_req,
1111 (unsigned long long) req->sector << 9,
1112 sg->offset, sg->length, sg_page(sg));
1114 last_sectors = sg->length >> 9;
1115 n = os_write_file(thread_fd, &io_req,
1116 sizeof(struct io_thread_req *));
1117 if(n != sizeof(struct io_thread_req *)){
1119 printk("write to io thread failed, "
1120 "errno = %d\n", -n);
1121 else if(list_empty(&dev->restart))
1122 list_add(&dev->restart, &restart);
1130 dev->request = NULL;
1134 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1136 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1140 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1144 static int ubd_ioctl(struct inode * inode, struct file * file,
1145 unsigned int cmd, unsigned long arg)
1147 struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1148 struct hd_driveid ubd_id = {
1155 struct cdrom_volctrl volume;
1156 case HDIO_GET_IDENTITY:
1157 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1158 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1164 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1166 volume.channel0 = 255;
1167 volume.channel1 = 255;
1168 volume.channel2 = 255;
1169 volume.channel3 = 255;
1170 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1177 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1179 struct uml_stat buf1, buf2;
1182 if(from_cmdline == NULL)
1184 if(!strcmp(from_cmdline, from_cow))
1187 err = os_stat_file(from_cmdline, &buf1);
1189 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1192 err = os_stat_file(from_cow, &buf2);
1194 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1197 if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1200 printk("Backing file mismatch - \"%s\" requested,\n"
1201 "\"%s\" specified in COW header of \"%s\"\n",
1202 from_cmdline, from_cow, cow);
1206 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1208 unsigned long modtime;
1209 unsigned long long actual;
1212 err = os_file_modtime(file, &modtime);
1214 printk("Failed to get modification time of backing file "
1215 "\"%s\", err = %d\n", file, -err);
1219 err = os_file_size(file, &actual);
1221 printk("Failed to get size of backing file \"%s\", "
1222 "err = %d\n", file, -err);
1227 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1229 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1230 "file\n", (unsigned long long) size, actual);
1233 if(modtime != mtime){
1234 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1235 "file\n", mtime, modtime);
1241 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1245 err = os_seek_file(fd, offset);
1249 err = os_read_file(fd, buf, len);
1256 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1257 char **backing_file_out, int *bitmap_offset_out,
1258 unsigned long *bitmap_len_out, int *data_offset_out,
1259 int *create_cow_out)
1262 unsigned long long size;
1263 __u32 version, align;
1265 int fd, err, sectorsize, asked_switch, mode = 0644;
1267 fd = os_open_file(file, *openflags, mode);
1269 if ((fd == -ENOENT) && (create_cow_out != NULL))
1270 *create_cow_out = 1;
1271 if (!openflags->w ||
1272 ((fd != -EROFS) && (fd != -EACCES)))
1275 fd = os_open_file(file, *openflags, mode);
1281 printk("Not locking \"%s\" on the host\n", file);
1283 err = os_lock_file(fd, openflags->w);
1285 printk("Failed to lock '%s', err = %d\n", file, -err);
1290 /* Successful return case! */
1291 if(backing_file_out == NULL)
1294 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1295 &size, §orsize, &align, bitmap_offset_out);
1296 if(err && (*backing_file_out != NULL)){
1297 printk("Failed to read COW header from COW file \"%s\", "
1298 "errno = %d\n", file, -err);
1304 asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1306 /* Allow switching only if no mismatch. */
1307 if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1308 printk("Switching backing file to '%s'\n", *backing_file_out);
1309 err = write_cow_header(file, fd, *backing_file_out,
1310 sectorsize, align, &size);
1312 printk("Switch failed, errno = %d\n", -err);
1316 *backing_file_out = backing_file;
1317 err = backing_file_mismatch(*backing_file_out, size, mtime);
1322 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1323 bitmap_len_out, data_offset_out);
1331 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1332 int sectorsize, int alignment, int *bitmap_offset_out,
1333 unsigned long *bitmap_len_out, int *data_offset_out)
1338 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1341 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1346 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1347 bitmap_offset_out, bitmap_len_out,
1356 static int update_bitmap(struct io_thread_req *req)
1360 if(req->cow_offset == -1)
1363 n = os_seek_file(req->fds[1], req->cow_offset);
1365 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1369 n = os_write_file(req->fds[1], &req->bitmap_words,
1370 sizeof(req->bitmap_words));
1371 if(n != sizeof(req->bitmap_words)){
1372 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1380 void do_io(struct io_thread_req *req)
1384 int n, nsectors, start, end, bit;
1388 nsectors = req->length / req->sectorsize;
1391 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1393 while((end < nsectors) &&
1394 (ubd_test_bit(end, (unsigned char *)
1395 &req->sector_mask) == bit))
1398 off = req->offset + req->offsets[bit] +
1399 start * req->sectorsize;
1400 len = (end - start) * req->sectorsize;
1401 buf = &req->buffer[start * req->sectorsize];
1403 err = os_seek_file(req->fds[bit], off);
1405 printk("do_io - lseek failed : err = %d\n", -err);
1409 if(req->op == UBD_READ){
1414 n = os_read_file(req->fds[bit], buf, len);
1416 printk("do_io - read failed, err = %d "
1417 "fd = %d\n", -n, req->fds[bit]);
1421 } while((n < len) && (n != 0));
1422 if (n < len) memset(&buf[n], 0, len - n);
1424 n = os_write_file(req->fds[bit], buf, len);
1426 printk("do_io - write failed err = %d "
1427 "fd = %d\n", -n, req->fds[bit]);
1434 } while(start < nsectors);
1436 req->error = update_bitmap(req);
1439 /* Changed in start_io_thread, which is serialized by being called only
1440 * from ubd_init, which is an initcall.
1444 /* Only changed by the io thread. XXX: currently unused. */
1445 static int io_count = 0;
1447 int io_thread(void *arg)
1449 struct io_thread_req *req;
1452 ignore_sigwinch_sig();
1454 n = os_read_file(kernel_fd, &req,
1455 sizeof(struct io_thread_req *));
1456 if(n != sizeof(struct io_thread_req *)){
1458 printk("io_thread - read failed, fd = %d, "
1459 "err = %d\n", kernel_fd, -n);
1461 printk("io_thread - short read, fd = %d, "
1462 "length = %d\n", kernel_fd, n);
1468 n = os_write_file(kernel_fd, &req,
1469 sizeof(struct io_thread_req *));
1470 if(n != sizeof(struct io_thread_req *))
1471 printk("io_thread - write failed, fd = %d, err = %d\n",