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/mutex.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"
49 #include "mconsole_kern.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 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
80 bits = sizeof(data[0]) * 8;
83 return (data[n] & (1 << off)) != 0;
86 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
91 bits = sizeof(data[0]) * 8;
94 data[n] |= (1 << off);
96 /*End stuff from ubd_user.h*/
98 #define DRIVER_NAME "uml-blkdev"
100 static DEFINE_MUTEX(ubd_lock);
101 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
103 static int ubd_open(struct block_device *bdev, fmode_t mode);
104 static int ubd_release(struct gendisk *disk, fmode_t mode);
105 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
106 unsigned int cmd, unsigned long arg);
107 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
111 static const struct block_device_operations ubd_blops = {
112 .owner = THIS_MODULE,
114 .release = ubd_release,
116 .getgeo = ubd_getgeo,
119 /* Protected by ubd_lock */
120 static int fake_major = UBD_MAJOR;
121 static struct gendisk *ubd_gendisk[MAX_DEV];
122 static struct gendisk *fake_gendisk[MAX_DEV];
124 #ifdef CONFIG_BLK_DEV_UBD_SYNC
125 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
128 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
131 static struct openflags global_openflags = OPEN_FLAGS;
134 /* backing file name */
136 /* backing file fd */
138 unsigned long *bitmap;
139 unsigned long bitmap_len;
147 struct list_head restart;
148 /* name (and fd, below) of the file opened for writing, either the
149 * backing or the cow file. */
154 struct openflags boot_openflags;
155 struct openflags openflags;
159 struct platform_device pdev;
160 struct request_queue *queue;
162 struct scatterlist sg[MAX_SG];
163 struct request *request;
164 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(ubd_devs.lock), \
193 /* Protected by ubd_lock */
194 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
196 /* Only changed by fake_ide_setup which is a setup */
197 static int fake_ide = 0;
198 static struct proc_dir_entry *proc_ide_root = NULL;
199 static struct proc_dir_entry *proc_ide = NULL;
201 static void make_proc_ide(void)
203 proc_ide_root = proc_mkdir("ide", NULL);
204 proc_ide = proc_mkdir("ide0", proc_ide_root);
207 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
209 seq_puts(m, "disk\n");
213 static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
215 return single_open(file, fake_ide_media_proc_show, NULL);
218 static const struct file_operations fake_ide_media_proc_fops = {
219 .owner = THIS_MODULE,
220 .open = fake_ide_media_proc_open,
223 .release = single_release,
226 static void make_ide_entries(const char *dev_name)
228 struct proc_dir_entry *dir, *ent;
231 if(proc_ide_root == NULL) make_proc_ide();
233 dir = proc_mkdir(dev_name, proc_ide);
236 ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
238 snprintf(name, sizeof(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 != UBD_MAJOR) {
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"
432 " 'c' will cause the device to be treated as being shared between multiple\n"
433 " UMLs and file locking will be turned off - this is appropriate for a\n"
434 " cluster filesystem and inappropriate at almost all other times.\n\n"
437 static int udb_setup(char *str)
439 printk("udb%s specified on command line is almost certainly a ubd -> "
444 __setup("udb", udb_setup);
445 __uml_help(udb_setup,
447 " This option is here solely to catch ubd -> udb typos, which can be\n"
448 " to impossible to catch visually unless you specifically look for\n"
449 " them. The only result of any option starting with 'udb' is an error\n"
450 " in the boot output.\n\n"
453 static void do_ubd_request(struct request_queue * q);
455 /* Only changed by ubd_init, which is an initcall. */
456 static int thread_fd = -1;
457 static LIST_HEAD(restart);
459 /* XXX - move this inside ubd_intr. */
460 /* Called without dev->lock held, and only in interrupt context. */
461 static void ubd_handler(void)
463 struct io_thread_req *req;
465 struct list_head *list, *next_ele;
470 n = os_read_file(thread_fd, &req,
471 sizeof(struct io_thread_req *));
472 if(n != sizeof(req)){
475 printk(KERN_ERR "spurious interrupt in ubd_handler, "
480 blk_end_request(req->req, 0, req->length);
483 reactivate_fd(thread_fd, UBD_IRQ);
485 list_for_each_safe(list, next_ele, &restart){
486 ubd = container_of(list, struct ubd, restart);
487 list_del_init(&ubd->restart);
488 spin_lock_irqsave(&ubd->lock, flags);
489 do_ubd_request(ubd->queue);
490 spin_unlock_irqrestore(&ubd->lock, flags);
494 static irqreturn_t ubd_intr(int irq, void *dev)
500 /* Only changed by ubd_init, which is an initcall. */
501 static int io_pid = -1;
503 static void kill_io_thread(void)
506 os_kill_process(io_pid, 1);
509 __uml_exitcall(kill_io_thread);
511 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
521 unsigned long long size;
525 if (ubd_dev->file && ubd_dev->cow.file) {
526 file = ubd_dev->cow.file;
531 fd = os_open_file(ubd_dev->file, global_openflags, 0);
535 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
536 &mtime, &size, §or_size, &align, &bitmap_offset);
540 file = ubd_dev->file;
545 return os_file_size(file, size_out);
548 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
552 err = os_seek_file(fd, offset);
556 err = os_read_file(fd, buf, len);
563 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
565 unsigned long modtime;
566 unsigned long long actual;
569 err = os_file_modtime(file, &modtime);
571 printk(KERN_ERR "Failed to get modification time of backing "
572 "file \"%s\", err = %d\n", file, -err);
576 err = os_file_size(file, &actual);
578 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
579 "err = %d\n", file, -err);
583 if (actual != size) {
584 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
586 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
587 "vs backing file\n", (unsigned long long) size, actual);
590 if (modtime != mtime) {
591 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
592 "backing file\n", mtime, modtime);
598 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
600 struct uml_stat buf1, buf2;
603 if (from_cmdline == NULL)
605 if (!strcmp(from_cmdline, from_cow))
608 err = os_stat_file(from_cmdline, &buf1);
610 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
614 err = os_stat_file(from_cow, &buf2);
616 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
620 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
623 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
624 "\"%s\" specified in COW header of \"%s\"\n",
625 from_cmdline, from_cow, cow);
629 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
630 char **backing_file_out, int *bitmap_offset_out,
631 unsigned long *bitmap_len_out, int *data_offset_out,
635 unsigned long long size;
636 __u32 version, align;
638 int fd, err, sectorsize, asked_switch, mode = 0644;
640 fd = os_open_file(file, *openflags, mode);
642 if ((fd == -ENOENT) && (create_cow_out != NULL))
645 ((fd != -EROFS) && (fd != -EACCES)))
648 fd = os_open_file(file, *openflags, mode);
654 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
656 err = os_lock_file(fd, openflags->w);
658 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
664 /* Successful return case! */
665 if (backing_file_out == NULL)
668 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
669 &size, §orsize, &align, bitmap_offset_out);
670 if (err && (*backing_file_out != NULL)) {
671 printk(KERN_ERR "Failed to read COW header from COW file "
672 "\"%s\", errno = %d\n", file, -err);
678 asked_switch = path_requires_switch(*backing_file_out, backing_file,
681 /* Allow switching only if no mismatch. */
682 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
684 printk(KERN_ERR "Switching backing file to '%s'\n",
686 err = write_cow_header(file, fd, *backing_file_out,
687 sectorsize, align, &size);
689 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
693 *backing_file_out = backing_file;
694 err = backing_file_mismatch(*backing_file_out, size, mtime);
699 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
700 bitmap_len_out, data_offset_out);
708 static int create_cow_file(char *cow_file, char *backing_file,
709 struct openflags flags,
710 int sectorsize, int alignment, int *bitmap_offset_out,
711 unsigned long *bitmap_len_out, int *data_offset_out)
716 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
719 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
724 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
725 bitmap_offset_out, bitmap_len_out,
734 static void ubd_close_dev(struct ubd *ubd_dev)
736 os_close_file(ubd_dev->fd);
737 if(ubd_dev->cow.file == NULL)
740 os_close_file(ubd_dev->cow.fd);
741 vfree(ubd_dev->cow.bitmap);
742 ubd_dev->cow.bitmap = NULL;
745 static int ubd_open_dev(struct ubd *ubd_dev)
747 struct openflags flags;
749 int err, create_cow, *create_ptr;
752 ubd_dev->openflags = ubd_dev->boot_openflags;
754 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
755 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
757 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
758 back_ptr, &ubd_dev->cow.bitmap_offset,
759 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
762 if((fd == -ENOENT) && create_cow){
763 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
764 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
765 &ubd_dev->cow.bitmap_offset,
766 &ubd_dev->cow.bitmap_len,
767 &ubd_dev->cow.data_offset);
769 printk(KERN_INFO "Creating \"%s\" as COW file for "
770 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
775 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
781 if(ubd_dev->cow.file != NULL){
782 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
785 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
786 if(ubd_dev->cow.bitmap == NULL){
787 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
790 flush_tlb_kernel_vm();
792 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
793 ubd_dev->cow.bitmap_offset,
794 ubd_dev->cow.bitmap_len);
798 flags = ubd_dev->openflags;
800 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
801 NULL, NULL, NULL, NULL);
802 if(err < 0) goto error;
803 ubd_dev->cow.fd = err;
807 os_close_file(ubd_dev->fd);
811 static void ubd_device_release(struct device *dev)
813 struct ubd *ubd_dev = dev_get_drvdata(dev);
815 blk_cleanup_queue(ubd_dev->queue);
816 *ubd_dev = ((struct ubd) DEFAULT_UBD);
819 static int ubd_disk_register(int major, u64 size, int unit,
820 struct gendisk **disk_out)
822 struct gendisk *disk;
824 disk = alloc_disk(1 << UBD_SHIFT);
829 disk->first_minor = unit << UBD_SHIFT;
830 disk->fops = &ubd_blops;
831 set_capacity(disk, size / 512);
832 if (major == UBD_MAJOR)
833 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
835 sprintf(disk->disk_name, "ubd_fake%d", unit);
837 /* sysfs register (not for ide fake devices) */
838 if (major == UBD_MAJOR) {
839 ubd_devs[unit].pdev.id = unit;
840 ubd_devs[unit].pdev.name = DRIVER_NAME;
841 ubd_devs[unit].pdev.dev.release = ubd_device_release;
842 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
843 platform_device_register(&ubd_devs[unit].pdev);
844 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
847 disk->private_data = &ubd_devs[unit];
848 disk->queue = ubd_devs[unit].queue;
855 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
857 static int ubd_add(int n, char **error_out)
859 struct ubd *ubd_dev = &ubd_devs[n];
862 if(ubd_dev->file == NULL)
865 err = ubd_file_size(ubd_dev, &ubd_dev->size);
867 *error_out = "Couldn't determine size of device's file";
871 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
873 INIT_LIST_HEAD(&ubd_dev->restart);
874 sg_init_table(ubd_dev->sg, MAX_SG);
877 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
878 if (ubd_dev->queue == NULL) {
879 *error_out = "Failed to initialize device queue";
882 ubd_dev->queue->queuedata = ubd_dev;
884 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
885 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
887 *error_out = "Failed to register device";
891 if (fake_major != UBD_MAJOR)
892 ubd_disk_register(fake_major, ubd_dev->size, n,
896 * Perhaps this should also be under the "if (fake_major)" above
897 * using the fake_disk->disk_name
900 make_ide_entries(ubd_gendisk[n]->disk_name);
907 blk_cleanup_queue(ubd_dev->queue);
911 static int ubd_config(char *str, char **error_out)
915 /* This string is possibly broken up and stored, so it's only
916 * freed if ubd_setup_common fails, or if only general options
919 str = kstrdup(str, GFP_KERNEL);
921 *error_out = "Failed to allocate memory";
925 ret = ubd_setup_common(str, &n, error_out);
934 mutex_lock(&ubd_lock);
935 ret = ubd_add(n, error_out);
937 ubd_devs[n].file = NULL;
938 mutex_unlock(&ubd_lock);
948 static int ubd_get_config(char *name, char *str, int size, char **error_out)
953 n = parse_unit(&name);
954 if((n >= MAX_DEV) || (n < 0)){
955 *error_out = "ubd_get_config : device number out of range";
959 ubd_dev = &ubd_devs[n];
960 mutex_lock(&ubd_lock);
962 if(ubd_dev->file == NULL){
963 CONFIG_CHUNK(str, size, len, "", 1);
967 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
969 if(ubd_dev->cow.file != NULL){
970 CONFIG_CHUNK(str, size, len, ",", 0);
971 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
973 else CONFIG_CHUNK(str, size, len, "", 1);
976 mutex_unlock(&ubd_lock);
980 static int ubd_id(char **str, int *start_out, int *end_out)
986 *end_out = MAX_DEV - 1;
990 static int ubd_remove(int n, char **error_out)
992 struct gendisk *disk = ubd_gendisk[n];
996 mutex_lock(&ubd_lock);
998 ubd_dev = &ubd_devs[n];
1000 if(ubd_dev->file == NULL)
1003 /* you cannot remove a open disk */
1005 if(ubd_dev->count > 0)
1008 ubd_gendisk[n] = NULL;
1014 if(fake_gendisk[n] != NULL){
1015 del_gendisk(fake_gendisk[n]);
1016 put_disk(fake_gendisk[n]);
1017 fake_gendisk[n] = NULL;
1021 platform_device_unregister(&ubd_dev->pdev);
1023 mutex_unlock(&ubd_lock);
1027 /* All these are called by mconsole in process context and without
1028 * ubd-specific locks. The structure itself is const except for .list.
1030 static struct mc_device ubd_mc = {
1031 .list = LIST_HEAD_INIT(ubd_mc.list),
1033 .config = ubd_config,
1034 .get_config = ubd_get_config,
1036 .remove = ubd_remove,
1039 static int __init ubd_mc_init(void)
1041 mconsole_register_dev(&ubd_mc);
1045 __initcall(ubd_mc_init);
1047 static int __init ubd0_init(void)
1049 struct ubd *ubd_dev = &ubd_devs[0];
1051 mutex_lock(&ubd_lock);
1052 if(ubd_dev->file == NULL)
1053 ubd_dev->file = "root_fs";
1054 mutex_unlock(&ubd_lock);
1059 __initcall(ubd0_init);
1061 /* Used in ubd_init, which is an initcall */
1062 static struct platform_driver ubd_driver = {
1064 .name = DRIVER_NAME,
1068 static int __init ubd_init(void)
1073 if (register_blkdev(UBD_MAJOR, "ubd"))
1076 if (fake_major != UBD_MAJOR) {
1077 char name[sizeof("ubd_nnn\0")];
1079 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1080 if (register_blkdev(fake_major, "ubd"))
1083 platform_driver_register(&ubd_driver);
1084 mutex_lock(&ubd_lock);
1085 for (i = 0; i < MAX_DEV; i++){
1086 err = ubd_add(i, &error);
1088 printk(KERN_ERR "Failed to initialize ubd device %d :"
1091 mutex_unlock(&ubd_lock);
1095 late_initcall(ubd_init);
1097 static int __init ubd_driver_init(void){
1098 unsigned long stack;
1101 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1102 if(global_openflags.s){
1103 printk(KERN_INFO "ubd: Synchronous mode\n");
1104 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1105 * enough. So use anyway the io thread. */
1107 stack = alloc_stack(0, 0);
1108 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1112 "ubd : Failed to start I/O thread (errno = %d) - "
1113 "falling back to synchronous I/O\n", -io_pid);
1117 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1118 IRQF_DISABLED, "ubd", ubd_devs);
1120 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1124 device_initcall(ubd_driver_init);
1126 static int ubd_open(struct block_device *bdev, fmode_t mode)
1128 struct gendisk *disk = bdev->bd_disk;
1129 struct ubd *ubd_dev = disk->private_data;
1132 mutex_lock(&ubd_mutex);
1133 if(ubd_dev->count == 0){
1134 err = ubd_open_dev(ubd_dev);
1136 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1137 disk->disk_name, ubd_dev->file, -err);
1142 set_disk_ro(disk, !ubd_dev->openflags.w);
1144 /* This should no more be needed. And it didn't work anyway to exclude
1145 * read-write remounting of filesystems.*/
1146 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1147 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1151 mutex_unlock(&ubd_mutex);
1155 static int ubd_release(struct gendisk *disk, fmode_t mode)
1157 struct ubd *ubd_dev = disk->private_data;
1159 mutex_lock(&ubd_mutex);
1160 if(--ubd_dev->count == 0)
1161 ubd_close_dev(ubd_dev);
1162 mutex_unlock(&ubd_mutex);
1166 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1167 __u64 *cow_offset, unsigned long *bitmap,
1168 __u64 bitmap_offset, unsigned long *bitmap_words,
1171 __u64 sector = io_offset >> 9;
1172 int i, update_bitmap = 0;
1174 for(i = 0; i < length >> 9; i++){
1175 if(cow_mask != NULL)
1176 ubd_set_bit(i, (unsigned char *) cow_mask);
1177 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1181 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1187 *cow_offset = sector / (sizeof(unsigned long) * 8);
1189 /* This takes care of the case where we're exactly at the end of the
1190 * device, and *cow_offset + 1 is off the end. So, just back it up
1191 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1192 * for the original diagnosis.
1194 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1195 sizeof(unsigned long)) - 1))
1198 bitmap_words[0] = bitmap[*cow_offset];
1199 bitmap_words[1] = bitmap[*cow_offset + 1];
1201 *cow_offset *= sizeof(unsigned long);
1202 *cow_offset += bitmap_offset;
1205 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1206 __u64 bitmap_offset, __u64 bitmap_len)
1208 __u64 sector = req->offset >> 9;
1211 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1212 panic("Operation too long");
1214 if(req->op == UBD_READ) {
1215 for(i = 0; i < req->length >> 9; i++){
1216 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1217 ubd_set_bit(i, (unsigned char *)
1221 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1222 &req->cow_offset, bitmap, bitmap_offset,
1223 req->bitmap_words, bitmap_len);
1226 /* Called with dev->lock held */
1227 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1228 unsigned long long offset, int page_offset,
1229 int len, struct page *page)
1231 struct gendisk *disk = req->rq_disk;
1232 struct ubd *ubd_dev = disk->private_data;
1235 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1237 io_req->fds[1] = ubd_dev->fd;
1238 io_req->cow_offset = -1;
1239 io_req->offset = offset;
1240 io_req->length = len;
1242 io_req->sector_mask = 0;
1244 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1245 io_req->offsets[0] = 0;
1246 io_req->offsets[1] = ubd_dev->cow.data_offset;
1247 io_req->buffer = page_address(page) + page_offset;
1248 io_req->sectorsize = 1 << 9;
1250 if(ubd_dev->cow.file != NULL)
1251 cowify_req(io_req, ubd_dev->cow.bitmap,
1252 ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1256 /* Called with dev->lock held */
1257 static void do_ubd_request(struct request_queue *q)
1259 struct io_thread_req *io_req;
1260 struct request *req;
1264 struct ubd *dev = q->queuedata;
1265 if(dev->end_sg == 0){
1266 struct request *req = blk_fetch_request(q);
1271 dev->rq_pos = blk_rq_pos(req);
1273 dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1277 while(dev->start_sg < dev->end_sg){
1278 struct scatterlist *sg = &dev->sg[dev->start_sg];
1280 io_req = kmalloc(sizeof(struct io_thread_req),
1283 if(list_empty(&dev->restart))
1284 list_add(&dev->restart, &restart);
1287 prepare_request(req, io_req,
1288 (unsigned long long)dev->rq_pos << 9,
1289 sg->offset, sg->length, sg_page(sg));
1291 n = os_write_file(thread_fd, &io_req,
1292 sizeof(struct io_thread_req *));
1293 if(n != sizeof(struct io_thread_req *)){
1295 printk("write to io thread failed, "
1296 "errno = %d\n", -n);
1297 else if(list_empty(&dev->restart))
1298 list_add(&dev->restart, &restart);
1303 dev->rq_pos += sg->length >> 9;
1307 dev->request = NULL;
1311 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1313 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1317 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1321 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1322 unsigned int cmd, unsigned long arg)
1324 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1325 u16 ubd_id[ATA_ID_WORDS];
1328 struct cdrom_volctrl volume;
1329 case HDIO_GET_IDENTITY:
1330 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1331 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1332 ubd_id[ATA_ID_HEADS] = 128;
1333 ubd_id[ATA_ID_SECTORS] = 32;
1334 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1340 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1342 volume.channel0 = 255;
1343 volume.channel1 = 255;
1344 volume.channel2 = 255;
1345 volume.channel3 = 255;
1346 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1353 static int update_bitmap(struct io_thread_req *req)
1357 if(req->cow_offset == -1)
1360 n = os_seek_file(req->fds[1], req->cow_offset);
1362 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1366 n = os_write_file(req->fds[1], &req->bitmap_words,
1367 sizeof(req->bitmap_words));
1368 if(n != sizeof(req->bitmap_words)){
1369 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1377 static void do_io(struct io_thread_req *req)
1381 int n, nsectors, start, end, bit;
1385 nsectors = req->length / req->sectorsize;
1388 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1390 while((end < nsectors) &&
1391 (ubd_test_bit(end, (unsigned char *)
1392 &req->sector_mask) == bit))
1395 off = req->offset + req->offsets[bit] +
1396 start * req->sectorsize;
1397 len = (end - start) * req->sectorsize;
1398 buf = &req->buffer[start * req->sectorsize];
1400 err = os_seek_file(req->fds[bit], off);
1402 printk("do_io - lseek failed : err = %d\n", -err);
1406 if(req->op == UBD_READ){
1411 n = os_read_file(req->fds[bit], buf, len);
1413 printk("do_io - read failed, err = %d "
1414 "fd = %d\n", -n, req->fds[bit]);
1418 } while((n < len) && (n != 0));
1419 if (n < len) memset(&buf[n], 0, len - n);
1421 n = os_write_file(req->fds[bit], buf, len);
1423 printk("do_io - write failed err = %d "
1424 "fd = %d\n", -n, req->fds[bit]);
1431 } while(start < nsectors);
1433 req->error = update_bitmap(req);
1436 /* Changed in start_io_thread, which is serialized by being called only
1437 * from ubd_init, which is an initcall.
1441 /* Only changed by the io thread. XXX: currently unused. */
1442 static int io_count = 0;
1444 int io_thread(void *arg)
1446 struct io_thread_req *req;
1449 ignore_sigwinch_sig();
1451 n = os_read_file(kernel_fd, &req,
1452 sizeof(struct io_thread_req *));
1453 if(n != sizeof(struct io_thread_req *)){
1455 printk("io_thread - read failed, fd = %d, "
1456 "err = %d\n", kernel_fd, -n);
1458 printk("io_thread - short read, fd = %d, "
1459 "length = %d\n", kernel_fd, n);
1465 n = os_write_file(kernel_fd, &req,
1466 sizeof(struct io_thread_req *));
1467 if(n != sizeof(struct io_thread_req *))
1468 printk("io_thread - write failed, fd = %d, err = %d\n",