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/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 "linux/scatterlist.h"
38 #include "asm/segment.h"
39 #include "asm/uaccess.h"
41 #include "asm/types.h"
42 #include "asm/tlbflush.h"
44 #include "kern_util.h"
46 #include "mconsole_kern.h"
56 enum ubd_req { UBD_READ, UBD_WRITE };
58 struct io_thread_req {
62 unsigned long offsets[2];
63 unsigned long long offset;
67 unsigned long sector_mask;
68 unsigned long long cow_offset;
69 unsigned long bitmap_words[2];
73 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
78 bits = sizeof(data[0]) * 8;
81 return (data[n] & (1 << off)) != 0;
84 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
89 bits = sizeof(data[0]) * 8;
92 data[n] |= (1 << off);
94 /*End stuff from ubd_user.h*/
96 #define DRIVER_NAME "uml-blkdev"
98 static DEFINE_MUTEX(ubd_lock);
100 static int ubd_open(struct block_device *bdev, fmode_t mode);
101 static int ubd_release(struct gendisk *disk, fmode_t mode);
102 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
103 unsigned int cmd, unsigned long arg);
104 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
108 static struct block_device_operations ubd_blops = {
109 .owner = THIS_MODULE,
111 .release = ubd_release,
113 .getgeo = ubd_getgeo,
116 /* Protected by ubd_lock */
117 static int fake_major = UBD_MAJOR;
118 static struct gendisk *ubd_gendisk[MAX_DEV];
119 static struct gendisk *fake_gendisk[MAX_DEV];
121 #ifdef CONFIG_BLK_DEV_UBD_SYNC
122 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
125 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
128 static struct openflags global_openflags = OPEN_FLAGS;
131 /* backing file name */
133 /* backing file fd */
135 unsigned long *bitmap;
136 unsigned long bitmap_len;
144 struct list_head restart;
145 /* name (and fd, below) of the file opened for writing, either the
146 * backing or the cow file. */
151 struct openflags boot_openflags;
152 struct openflags openflags;
156 struct platform_device pdev;
157 struct request_queue *queue;
159 struct scatterlist sg[MAX_SG];
160 struct request *request;
161 int start_sg, end_sg;
164 #define DEFAULT_COW { \
168 .bitmap_offset = 0, \
172 #define DEFAULT_UBD { \
177 .boot_openflags = OPEN_FLAGS, \
178 .openflags = OPEN_FLAGS, \
181 .cow = DEFAULT_COW, \
182 .lock = SPIN_LOCK_UNLOCKED, \
188 /* Protected by ubd_lock */
189 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
191 /* Only changed by fake_ide_setup which is a setup */
192 static int fake_ide = 0;
193 static struct proc_dir_entry *proc_ide_root = NULL;
194 static struct proc_dir_entry *proc_ide = NULL;
196 static void make_proc_ide(void)
198 proc_ide_root = proc_mkdir("ide", NULL);
199 proc_ide = proc_mkdir("ide0", proc_ide_root);
202 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
203 int *eof, void *data)
207 strcpy(page, "disk\n");
208 len = strlen("disk\n");
212 if (len <= 0) return 0;
219 static void make_ide_entries(const char *dev_name)
221 struct proc_dir_entry *dir, *ent;
224 if(proc_ide_root == NULL) make_proc_ide();
226 dir = proc_mkdir(dev_name, proc_ide);
229 ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
232 ent->read_proc = proc_ide_read_media;
233 ent->write_proc = NULL;
234 snprintf(name, sizeof(name), "ide0/%s", dev_name);
235 proc_symlink(dev_name, proc_ide_root, name);
238 static int fake_ide_setup(char *str)
244 __setup("fake_ide", fake_ide_setup);
246 __uml_help(fake_ide_setup,
248 " Create ide0 entries that map onto ubd devices.\n\n"
251 static int parse_unit(char **ptr)
253 char *str = *ptr, *end;
257 n = simple_strtoul(str, &end, 0);
262 else if (('a' <= *str) && (*str <= 'z')) {
270 /* If *index_out == -1 at exit, the passed option was a general one;
271 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
272 * should not be freed on exit.
274 static int ubd_setup_common(char *str, int *index_out, char **error_out)
277 struct openflags flags = global_openflags;
281 if(index_out) *index_out = -1;
288 if(!strcmp(str, "sync")){
289 global_openflags = of_sync(global_openflags);
294 major = simple_strtoul(str, &end, 0);
295 if((*end != '\0') || (end == str)){
296 *error_out = "Didn't parse major number";
300 mutex_lock(&ubd_lock);
301 if (fake_major != UBD_MAJOR) {
302 *error_out = "Can't assign a fake major twice";
308 printk(KERN_INFO "Setting extra ubd major number to %d\n",
312 mutex_unlock(&ubd_lock);
316 n = parse_unit(&str);
318 *error_out = "Couldn't parse device number";
322 *error_out = "Device number out of range";
327 mutex_lock(&ubd_lock);
329 ubd_dev = &ubd_devs[n];
330 if(ubd_dev->file != NULL){
331 *error_out = "Device is already configured";
339 for (i = 0; i < sizeof("rscd="); i++) {
357 *error_out = "Expected '=' or flag letter "
365 *error_out = "Too many flags specified";
367 *error_out = "Missing '='";
371 backing_file = strchr(str, ',');
373 if (backing_file == NULL)
374 backing_file = strchr(str, ':');
376 if(backing_file != NULL){
378 *error_out = "Can't specify both 'd' and a cow file";
382 *backing_file = '\0';
388 ubd_dev->cow.file = backing_file;
389 ubd_dev->boot_openflags = flags;
391 mutex_unlock(&ubd_lock);
395 static int ubd_setup(char *str)
400 err = ubd_setup_common(str, NULL, &error);
402 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
407 __setup("ubd", ubd_setup);
408 __uml_help(ubd_setup,
409 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
410 " This is used to associate a device with a file in the underlying\n"
411 " filesystem. When specifying two filenames, the first one is the\n"
412 " COW name and the second is the backing file name. As separator you can\n"
413 " use either a ':' or a ',': the first one allows writing things like;\n"
414 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
415 " while with a ',' the shell would not expand the 2nd '~'.\n"
416 " When using only one filename, UML will detect whether to treat it like\n"
417 " a COW file or a backing file. To override this detection, add the 'd'\n"
419 " ubd0d=BackingFile\n"
420 " Usually, there is a filesystem in the file, but \n"
421 " that's not required. Swap devices containing swap files can be\n"
422 " specified like this. Also, a file which doesn't contain a\n"
423 " filesystem can have its contents read in the virtual \n"
424 " machine by running 'dd' on the device. <n> must be in the range\n"
425 " 0 to 7. Appending an 'r' to the number will cause that device\n"
426 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
427 " an 's' will cause data to be written to disk on the host immediately.\n"
428 " 'c' will cause the device to be treated as being shared between multiple\n"
429 " UMLs and file locking will be turned off - this is appropriate for a\n"
430 " cluster filesystem and inappropriate at almost all other times.\n\n"
433 static int udb_setup(char *str)
435 printk("udb%s specified on command line is almost certainly a ubd -> "
440 __setup("udb", udb_setup);
441 __uml_help(udb_setup,
443 " This option is here solely to catch ubd -> udb typos, which can be\n"
444 " to impossible to catch visually unless you specifically look for\n"
445 " them. The only result of any option starting with 'udb' is an error\n"
446 " in the boot output.\n\n"
449 static void do_ubd_request(struct request_queue * q);
451 /* Only changed by ubd_init, which is an initcall. */
452 static int thread_fd = -1;
454 static void ubd_end_request(struct request *req, int bytes, int error)
456 blk_end_request(req, error, bytes);
459 /* Callable only from interrupt context - otherwise you need to do
460 * spin_lock_irq()/spin_lock_irqsave() */
461 static inline void ubd_finish(struct request *req, int bytes)
464 ubd_end_request(req, 0, -EIO);
467 ubd_end_request(req, bytes, 0);
470 static LIST_HEAD(restart);
472 /* XXX - move this inside ubd_intr. */
473 /* Called without dev->lock held, and only in interrupt context. */
474 static void ubd_handler(void)
476 struct io_thread_req *req;
479 struct list_head *list, *next_ele;
484 n = os_read_file(thread_fd, &req,
485 sizeof(struct io_thread_req *));
486 if(n != sizeof(req)){
489 printk(KERN_ERR "spurious interrupt in ubd_handler, "
495 rq->nr_sectors -= req->length >> 9;
496 if(rq->nr_sectors == 0)
497 ubd_finish(rq, rq->hard_nr_sectors << 9);
500 reactivate_fd(thread_fd, UBD_IRQ);
502 list_for_each_safe(list, next_ele, &restart){
503 ubd = container_of(list, struct ubd, restart);
504 list_del_init(&ubd->restart);
505 spin_lock_irqsave(&ubd->lock, flags);
506 do_ubd_request(ubd->queue);
507 spin_unlock_irqrestore(&ubd->lock, flags);
511 static irqreturn_t ubd_intr(int irq, void *dev)
517 /* Only changed by ubd_init, which is an initcall. */
518 static int io_pid = -1;
520 static void kill_io_thread(void)
523 os_kill_process(io_pid, 1);
526 __uml_exitcall(kill_io_thread);
528 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
532 file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
533 return os_file_size(file, size_out);
536 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
540 err = os_seek_file(fd, offset);
544 err = os_read_file(fd, buf, len);
551 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
553 unsigned long modtime;
554 unsigned long long actual;
557 err = os_file_modtime(file, &modtime);
559 printk(KERN_ERR "Failed to get modification time of backing "
560 "file \"%s\", err = %d\n", file, -err);
564 err = os_file_size(file, &actual);
566 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
567 "err = %d\n", file, -err);
571 if (actual != size) {
572 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
574 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
575 "vs backing file\n", (unsigned long long) size, actual);
578 if (modtime != mtime) {
579 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
580 "backing file\n", mtime, modtime);
586 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
588 struct uml_stat buf1, buf2;
591 if (from_cmdline == NULL)
593 if (!strcmp(from_cmdline, from_cow))
596 err = os_stat_file(from_cmdline, &buf1);
598 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
602 err = os_stat_file(from_cow, &buf2);
604 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
608 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
611 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
612 "\"%s\" specified in COW header of \"%s\"\n",
613 from_cmdline, from_cow, cow);
617 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
618 char **backing_file_out, int *bitmap_offset_out,
619 unsigned long *bitmap_len_out, int *data_offset_out,
623 unsigned long long size;
624 __u32 version, align;
626 int fd, err, sectorsize, asked_switch, mode = 0644;
628 fd = os_open_file(file, *openflags, mode);
630 if ((fd == -ENOENT) && (create_cow_out != NULL))
633 ((fd != -EROFS) && (fd != -EACCES)))
636 fd = os_open_file(file, *openflags, mode);
642 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
644 err = os_lock_file(fd, openflags->w);
646 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
652 /* Successful return case! */
653 if (backing_file_out == NULL)
656 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
657 &size, §orsize, &align, bitmap_offset_out);
658 if (err && (*backing_file_out != NULL)) {
659 printk(KERN_ERR "Failed to read COW header from COW file "
660 "\"%s\", errno = %d\n", file, -err);
666 asked_switch = path_requires_switch(*backing_file_out, backing_file,
669 /* Allow switching only if no mismatch. */
670 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
672 printk(KERN_ERR "Switching backing file to '%s'\n",
674 err = write_cow_header(file, fd, *backing_file_out,
675 sectorsize, align, &size);
677 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
681 *backing_file_out = backing_file;
682 err = backing_file_mismatch(*backing_file_out, size, mtime);
687 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
688 bitmap_len_out, data_offset_out);
696 static int create_cow_file(char *cow_file, char *backing_file,
697 struct openflags flags,
698 int sectorsize, int alignment, int *bitmap_offset_out,
699 unsigned long *bitmap_len_out, int *data_offset_out)
704 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
707 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
712 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
713 bitmap_offset_out, bitmap_len_out,
722 static void ubd_close_dev(struct ubd *ubd_dev)
724 os_close_file(ubd_dev->fd);
725 if(ubd_dev->cow.file == NULL)
728 os_close_file(ubd_dev->cow.fd);
729 vfree(ubd_dev->cow.bitmap);
730 ubd_dev->cow.bitmap = NULL;
733 static int ubd_open_dev(struct ubd *ubd_dev)
735 struct openflags flags;
737 int err, create_cow, *create_ptr;
740 ubd_dev->openflags = ubd_dev->boot_openflags;
742 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
743 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
745 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
746 back_ptr, &ubd_dev->cow.bitmap_offset,
747 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
750 if((fd == -ENOENT) && create_cow){
751 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
752 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
753 &ubd_dev->cow.bitmap_offset,
754 &ubd_dev->cow.bitmap_len,
755 &ubd_dev->cow.data_offset);
757 printk(KERN_INFO "Creating \"%s\" as COW file for "
758 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
763 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
769 if(ubd_dev->cow.file != NULL){
770 blk_queue_max_sectors(ubd_dev->queue, 8 * sizeof(long));
773 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
774 if(ubd_dev->cow.bitmap == NULL){
775 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
778 flush_tlb_kernel_vm();
780 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
781 ubd_dev->cow.bitmap_offset,
782 ubd_dev->cow.bitmap_len);
786 flags = ubd_dev->openflags;
788 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
789 NULL, NULL, NULL, NULL);
790 if(err < 0) goto error;
791 ubd_dev->cow.fd = err;
795 os_close_file(ubd_dev->fd);
799 static void ubd_device_release(struct device *dev)
801 struct ubd *ubd_dev = dev->driver_data;
803 blk_cleanup_queue(ubd_dev->queue);
804 *ubd_dev = ((struct ubd) DEFAULT_UBD);
807 static int ubd_disk_register(int major, u64 size, int unit,
808 struct gendisk **disk_out)
810 struct gendisk *disk;
812 disk = alloc_disk(1 << UBD_SHIFT);
817 disk->first_minor = unit << UBD_SHIFT;
818 disk->fops = &ubd_blops;
819 set_capacity(disk, size / 512);
820 if (major == UBD_MAJOR)
821 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
823 sprintf(disk->disk_name, "ubd_fake%d", unit);
825 /* sysfs register (not for ide fake devices) */
826 if (major == UBD_MAJOR) {
827 ubd_devs[unit].pdev.id = unit;
828 ubd_devs[unit].pdev.name = DRIVER_NAME;
829 ubd_devs[unit].pdev.dev.release = ubd_device_release;
830 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
831 platform_device_register(&ubd_devs[unit].pdev);
832 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
835 disk->private_data = &ubd_devs[unit];
836 disk->queue = ubd_devs[unit].queue;
843 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
845 static int ubd_add(int n, char **error_out)
847 struct ubd *ubd_dev = &ubd_devs[n];
850 if(ubd_dev->file == NULL)
853 err = ubd_file_size(ubd_dev, &ubd_dev->size);
855 *error_out = "Couldn't determine size of device's file";
859 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
861 INIT_LIST_HEAD(&ubd_dev->restart);
862 sg_init_table(ubd_dev->sg, MAX_SG);
865 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
866 if (ubd_dev->queue == NULL) {
867 *error_out = "Failed to initialize device queue";
870 ubd_dev->queue->queuedata = ubd_dev;
872 blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG);
873 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
875 *error_out = "Failed to register device";
879 if (fake_major != UBD_MAJOR)
880 ubd_disk_register(fake_major, ubd_dev->size, n,
884 * Perhaps this should also be under the "if (fake_major)" above
885 * using the fake_disk->disk_name
888 make_ide_entries(ubd_gendisk[n]->disk_name);
895 blk_cleanup_queue(ubd_dev->queue);
899 static int ubd_config(char *str, char **error_out)
903 /* This string is possibly broken up and stored, so it's only
904 * freed if ubd_setup_common fails, or if only general options
907 str = kstrdup(str, GFP_KERNEL);
909 *error_out = "Failed to allocate memory";
913 ret = ubd_setup_common(str, &n, error_out);
922 mutex_lock(&ubd_lock);
923 ret = ubd_add(n, error_out);
925 ubd_devs[n].file = NULL;
926 mutex_unlock(&ubd_lock);
936 static int ubd_get_config(char *name, char *str, int size, char **error_out)
941 n = parse_unit(&name);
942 if((n >= MAX_DEV) || (n < 0)){
943 *error_out = "ubd_get_config : device number out of range";
947 ubd_dev = &ubd_devs[n];
948 mutex_lock(&ubd_lock);
950 if(ubd_dev->file == NULL){
951 CONFIG_CHUNK(str, size, len, "", 1);
955 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
957 if(ubd_dev->cow.file != NULL){
958 CONFIG_CHUNK(str, size, len, ",", 0);
959 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
961 else CONFIG_CHUNK(str, size, len, "", 1);
964 mutex_unlock(&ubd_lock);
968 static int ubd_id(char **str, int *start_out, int *end_out)
974 *end_out = MAX_DEV - 1;
978 static int ubd_remove(int n, char **error_out)
980 struct gendisk *disk = ubd_gendisk[n];
984 mutex_lock(&ubd_lock);
986 ubd_dev = &ubd_devs[n];
988 if(ubd_dev->file == NULL)
991 /* you cannot remove a open disk */
993 if(ubd_dev->count > 0)
996 ubd_gendisk[n] = NULL;
1002 if(fake_gendisk[n] != NULL){
1003 del_gendisk(fake_gendisk[n]);
1004 put_disk(fake_gendisk[n]);
1005 fake_gendisk[n] = NULL;
1009 platform_device_unregister(&ubd_dev->pdev);
1011 mutex_unlock(&ubd_lock);
1015 /* All these are called by mconsole in process context and without
1016 * ubd-specific locks. The structure itself is const except for .list.
1018 static struct mc_device ubd_mc = {
1019 .list = LIST_HEAD_INIT(ubd_mc.list),
1021 .config = ubd_config,
1022 .get_config = ubd_get_config,
1024 .remove = ubd_remove,
1027 static int __init ubd_mc_init(void)
1029 mconsole_register_dev(&ubd_mc);
1033 __initcall(ubd_mc_init);
1035 static int __init ubd0_init(void)
1037 struct ubd *ubd_dev = &ubd_devs[0];
1039 mutex_lock(&ubd_lock);
1040 if(ubd_dev->file == NULL)
1041 ubd_dev->file = "root_fs";
1042 mutex_unlock(&ubd_lock);
1047 __initcall(ubd0_init);
1049 /* Used in ubd_init, which is an initcall */
1050 static struct platform_driver ubd_driver = {
1052 .name = DRIVER_NAME,
1056 static int __init ubd_init(void)
1061 if (register_blkdev(UBD_MAJOR, "ubd"))
1064 if (fake_major != UBD_MAJOR) {
1065 char name[sizeof("ubd_nnn\0")];
1067 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1068 if (register_blkdev(fake_major, "ubd"))
1071 platform_driver_register(&ubd_driver);
1072 mutex_lock(&ubd_lock);
1073 for (i = 0; i < MAX_DEV; i++){
1074 err = ubd_add(i, &error);
1076 printk(KERN_ERR "Failed to initialize ubd device %d :"
1079 mutex_unlock(&ubd_lock);
1083 late_initcall(ubd_init);
1085 static int __init ubd_driver_init(void){
1086 unsigned long stack;
1089 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1090 if(global_openflags.s){
1091 printk(KERN_INFO "ubd: Synchronous mode\n");
1092 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1093 * enough. So use anyway the io thread. */
1095 stack = alloc_stack(0, 0);
1096 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1100 "ubd : Failed to start I/O thread (errno = %d) - "
1101 "falling back to synchronous I/O\n", -io_pid);
1105 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1106 IRQF_DISABLED, "ubd", ubd_devs);
1108 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1112 device_initcall(ubd_driver_init);
1114 static int ubd_open(struct block_device *bdev, fmode_t mode)
1116 struct gendisk *disk = bdev->bd_disk;
1117 struct ubd *ubd_dev = disk->private_data;
1120 if(ubd_dev->count == 0){
1121 err = ubd_open_dev(ubd_dev);
1123 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1124 disk->disk_name, ubd_dev->file, -err);
1129 set_disk_ro(disk, !ubd_dev->openflags.w);
1131 /* This should no more be needed. And it didn't work anyway to exclude
1132 * read-write remounting of filesystems.*/
1133 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1134 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1141 static int ubd_release(struct gendisk *disk, fmode_t mode)
1143 struct ubd *ubd_dev = disk->private_data;
1145 if(--ubd_dev->count == 0)
1146 ubd_close_dev(ubd_dev);
1150 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1151 __u64 *cow_offset, unsigned long *bitmap,
1152 __u64 bitmap_offset, unsigned long *bitmap_words,
1155 __u64 sector = io_offset >> 9;
1156 int i, update_bitmap = 0;
1158 for(i = 0; i < length >> 9; i++){
1159 if(cow_mask != NULL)
1160 ubd_set_bit(i, (unsigned char *) cow_mask);
1161 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1165 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1171 *cow_offset = sector / (sizeof(unsigned long) * 8);
1173 /* This takes care of the case where we're exactly at the end of the
1174 * device, and *cow_offset + 1 is off the end. So, just back it up
1175 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1176 * for the original diagnosis.
1178 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1179 sizeof(unsigned long)) - 1))
1182 bitmap_words[0] = bitmap[*cow_offset];
1183 bitmap_words[1] = bitmap[*cow_offset + 1];
1185 *cow_offset *= sizeof(unsigned long);
1186 *cow_offset += bitmap_offset;
1189 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1190 __u64 bitmap_offset, __u64 bitmap_len)
1192 __u64 sector = req->offset >> 9;
1195 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1196 panic("Operation too long");
1198 if(req->op == UBD_READ) {
1199 for(i = 0; i < req->length >> 9; i++){
1200 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1201 ubd_set_bit(i, (unsigned char *)
1205 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1206 &req->cow_offset, bitmap, bitmap_offset,
1207 req->bitmap_words, bitmap_len);
1210 /* Called with dev->lock held */
1211 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1212 unsigned long long offset, int page_offset,
1213 int len, struct page *page)
1215 struct gendisk *disk = req->rq_disk;
1216 struct ubd *ubd_dev = disk->private_data;
1219 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1221 io_req->fds[1] = ubd_dev->fd;
1222 io_req->cow_offset = -1;
1223 io_req->offset = offset;
1224 io_req->length = len;
1226 io_req->sector_mask = 0;
1228 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1229 io_req->offsets[0] = 0;
1230 io_req->offsets[1] = ubd_dev->cow.data_offset;
1231 io_req->buffer = page_address(page) + page_offset;
1232 io_req->sectorsize = 1 << 9;
1234 if(ubd_dev->cow.file != NULL)
1235 cowify_req(io_req, ubd_dev->cow.bitmap,
1236 ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1240 /* Called with dev->lock held */
1241 static void do_ubd_request(struct request_queue *q)
1243 struct io_thread_req *io_req;
1244 struct request *req;
1245 int n, last_sectors;
1248 struct ubd *dev = q->queuedata;
1249 if(dev->end_sg == 0){
1250 struct request *req = elv_next_request(q);
1255 blkdev_dequeue_request(req);
1257 dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1262 while(dev->start_sg < dev->end_sg){
1263 struct scatterlist *sg = &dev->sg[dev->start_sg];
1265 req->sector += last_sectors;
1266 io_req = kmalloc(sizeof(struct io_thread_req),
1269 if(list_empty(&dev->restart))
1270 list_add(&dev->restart, &restart);
1273 prepare_request(req, io_req,
1274 (unsigned long long) req->sector << 9,
1275 sg->offset, sg->length, sg_page(sg));
1277 last_sectors = sg->length >> 9;
1278 n = os_write_file(thread_fd, &io_req,
1279 sizeof(struct io_thread_req *));
1280 if(n != sizeof(struct io_thread_req *)){
1282 printk("write to io thread failed, "
1283 "errno = %d\n", -n);
1284 else if(list_empty(&dev->restart))
1285 list_add(&dev->restart, &restart);
1293 dev->request = NULL;
1297 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1299 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1303 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1307 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1308 unsigned int cmd, unsigned long arg)
1310 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1311 struct hd_driveid ubd_id = {
1318 struct cdrom_volctrl volume;
1319 case HDIO_GET_IDENTITY:
1320 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1321 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1327 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1329 volume.channel0 = 255;
1330 volume.channel1 = 255;
1331 volume.channel2 = 255;
1332 volume.channel3 = 255;
1333 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1340 static int update_bitmap(struct io_thread_req *req)
1344 if(req->cow_offset == -1)
1347 n = os_seek_file(req->fds[1], req->cow_offset);
1349 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1353 n = os_write_file(req->fds[1], &req->bitmap_words,
1354 sizeof(req->bitmap_words));
1355 if(n != sizeof(req->bitmap_words)){
1356 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1364 static void do_io(struct io_thread_req *req)
1368 int n, nsectors, start, end, bit;
1372 nsectors = req->length / req->sectorsize;
1375 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1377 while((end < nsectors) &&
1378 (ubd_test_bit(end, (unsigned char *)
1379 &req->sector_mask) == bit))
1382 off = req->offset + req->offsets[bit] +
1383 start * req->sectorsize;
1384 len = (end - start) * req->sectorsize;
1385 buf = &req->buffer[start * req->sectorsize];
1387 err = os_seek_file(req->fds[bit], off);
1389 printk("do_io - lseek failed : err = %d\n", -err);
1393 if(req->op == UBD_READ){
1398 n = os_read_file(req->fds[bit], buf, len);
1400 printk("do_io - read failed, err = %d "
1401 "fd = %d\n", -n, req->fds[bit]);
1405 } while((n < len) && (n != 0));
1406 if (n < len) memset(&buf[n], 0, len - n);
1408 n = os_write_file(req->fds[bit], buf, len);
1410 printk("do_io - write failed err = %d "
1411 "fd = %d\n", -n, req->fds[bit]);
1418 } while(start < nsectors);
1420 req->error = update_bitmap(req);
1423 /* Changed in start_io_thread, which is serialized by being called only
1424 * from ubd_init, which is an initcall.
1428 /* Only changed by the io thread. XXX: currently unused. */
1429 static int io_count = 0;
1431 int io_thread(void *arg)
1433 struct io_thread_req *req;
1436 ignore_sigwinch_sig();
1438 n = os_read_file(kernel_fd, &req,
1439 sizeof(struct io_thread_req *));
1440 if(n != sizeof(struct io_thread_req *)){
1442 printk("io_thread - read failed, fd = %d, "
1443 "err = %d\n", kernel_fd, -n);
1445 printk("io_thread - short read, fd = %d, "
1446 "length = %d\n", kernel_fd, n);
1452 n = os_write_file(kernel_fd, &req,
1453 sizeof(struct io_thread_req *));
1454 if(n != sizeof(struct io_thread_req *))
1455 printk("io_thread - write failed, fd = %d, err = %d\n",