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/module.h>
23 #include <linux/init.h>
24 #include <linux/blkdev.h>
25 #include <linux/ata.h>
26 #include <linux/hdreg.h>
27 #include <linux/cdrom.h>
28 #include <linux/proc_fs.h>
29 #include <linux/seq_file.h>
30 #include <linux/ctype.h>
31 #include <linux/slab.h>
32 #include <linux/vmalloc.h>
33 #include <linux/platform_device.h>
34 #include <linux/scatterlist.h>
35 #include <asm/tlbflush.h>
36 #include <kern_util.h>
37 #include "mconsole_kern.h"
44 enum ubd_req { UBD_READ, UBD_WRITE, UBD_FLUSH };
46 struct io_thread_req {
50 unsigned long offsets[2];
51 unsigned long long offset;
55 unsigned long sector_mask;
56 unsigned long long cow_offset;
57 unsigned long bitmap_words[2];
61 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
66 bits = sizeof(data[0]) * 8;
69 return (data[n] & (1 << off)) != 0;
72 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
77 bits = sizeof(data[0]) * 8;
80 data[n] |= (1 << off);
82 /*End stuff from ubd_user.h*/
84 #define DRIVER_NAME "uml-blkdev"
86 static DEFINE_MUTEX(ubd_lock);
87 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
89 static int ubd_open(struct block_device *bdev, fmode_t mode);
90 static void ubd_release(struct gendisk *disk, fmode_t mode);
91 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
92 unsigned int cmd, unsigned long arg);
93 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
97 static const struct block_device_operations ubd_blops = {
100 .release = ubd_release,
102 .getgeo = ubd_getgeo,
105 /* Protected by ubd_lock */
106 static int fake_major = UBD_MAJOR;
107 static struct gendisk *ubd_gendisk[MAX_DEV];
108 static struct gendisk *fake_gendisk[MAX_DEV];
110 #ifdef CONFIG_BLK_DEV_UBD_SYNC
111 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
114 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
117 static struct openflags global_openflags = OPEN_FLAGS;
120 /* backing file name */
122 /* backing file fd */
124 unsigned long *bitmap;
125 unsigned long bitmap_len;
133 struct list_head restart;
134 /* name (and fd, below) of the file opened for writing, either the
135 * backing or the cow file. */
140 struct openflags boot_openflags;
141 struct openflags openflags;
145 struct platform_device pdev;
146 struct request_queue *queue;
148 struct scatterlist sg[MAX_SG];
149 struct request *request;
150 int start_sg, end_sg;
154 #define DEFAULT_COW { \
158 .bitmap_offset = 0, \
162 #define DEFAULT_UBD { \
167 .boot_openflags = OPEN_FLAGS, \
168 .openflags = OPEN_FLAGS, \
171 .cow = DEFAULT_COW, \
172 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
179 /* Protected by ubd_lock */
180 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
182 /* Only changed by fake_ide_setup which is a setup */
183 static int fake_ide = 0;
184 static struct proc_dir_entry *proc_ide_root = NULL;
185 static struct proc_dir_entry *proc_ide = NULL;
187 static void make_proc_ide(void)
189 proc_ide_root = proc_mkdir("ide", NULL);
190 proc_ide = proc_mkdir("ide0", proc_ide_root);
193 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
195 seq_puts(m, "disk\n");
199 static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
201 return single_open(file, fake_ide_media_proc_show, NULL);
204 static const struct file_operations fake_ide_media_proc_fops = {
205 .owner = THIS_MODULE,
206 .open = fake_ide_media_proc_open,
209 .release = single_release,
212 static void make_ide_entries(const char *dev_name)
214 struct proc_dir_entry *dir, *ent;
217 if(proc_ide_root == NULL) make_proc_ide();
219 dir = proc_mkdir(dev_name, proc_ide);
222 ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
224 snprintf(name, sizeof(name), "ide0/%s", dev_name);
225 proc_symlink(dev_name, proc_ide_root, name);
228 static int fake_ide_setup(char *str)
234 __setup("fake_ide", fake_ide_setup);
236 __uml_help(fake_ide_setup,
238 " Create ide0 entries that map onto ubd devices.\n\n"
241 static int parse_unit(char **ptr)
243 char *str = *ptr, *end;
247 n = simple_strtoul(str, &end, 0);
252 else if (('a' <= *str) && (*str <= 'z')) {
260 /* If *index_out == -1 at exit, the passed option was a general one;
261 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
262 * should not be freed on exit.
264 static int ubd_setup_common(char *str, int *index_out, char **error_out)
267 struct openflags flags = global_openflags;
271 if(index_out) *index_out = -1;
278 if(!strcmp(str, "sync")){
279 global_openflags = of_sync(global_openflags);
284 major = simple_strtoul(str, &end, 0);
285 if((*end != '\0') || (end == str)){
286 *error_out = "Didn't parse major number";
290 mutex_lock(&ubd_lock);
291 if (fake_major != UBD_MAJOR) {
292 *error_out = "Can't assign a fake major twice";
298 printk(KERN_INFO "Setting extra ubd major number to %d\n",
302 mutex_unlock(&ubd_lock);
306 n = parse_unit(&str);
308 *error_out = "Couldn't parse device number";
312 *error_out = "Device number out of range";
317 mutex_lock(&ubd_lock);
319 ubd_dev = &ubd_devs[n];
320 if(ubd_dev->file != NULL){
321 *error_out = "Device is already configured";
329 for (i = 0; i < sizeof("rscd="); i++) {
347 *error_out = "Expected '=' or flag letter "
355 *error_out = "Too many flags specified";
357 *error_out = "Missing '='";
361 backing_file = strchr(str, ',');
363 if (backing_file == NULL)
364 backing_file = strchr(str, ':');
366 if(backing_file != NULL){
368 *error_out = "Can't specify both 'd' and a cow file";
372 *backing_file = '\0';
378 ubd_dev->cow.file = backing_file;
379 ubd_dev->boot_openflags = flags;
381 mutex_unlock(&ubd_lock);
385 static int ubd_setup(char *str)
390 err = ubd_setup_common(str, NULL, &error);
392 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
397 __setup("ubd", ubd_setup);
398 __uml_help(ubd_setup,
399 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
400 " This is used to associate a device with a file in the underlying\n"
401 " filesystem. When specifying two filenames, the first one is the\n"
402 " COW name and the second is the backing file name. As separator you can\n"
403 " use either a ':' or a ',': the first one allows writing things like;\n"
404 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
405 " while with a ',' the shell would not expand the 2nd '~'.\n"
406 " When using only one filename, UML will detect whether to treat it like\n"
407 " a COW file or a backing file. To override this detection, add the 'd'\n"
409 " ubd0d=BackingFile\n"
410 " Usually, there is a filesystem in the file, but \n"
411 " that's not required. Swap devices containing swap files can be\n"
412 " specified like this. Also, a file which doesn't contain a\n"
413 " filesystem can have its contents read in the virtual \n"
414 " machine by running 'dd' on the device. <n> must be in the range\n"
415 " 0 to 7. Appending an 'r' to the number will cause that device\n"
416 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
417 " an 's' will cause data to be written to disk on the host immediately.\n"
418 " 'c' will cause the device to be treated as being shared between multiple\n"
419 " UMLs and file locking will be turned off - this is appropriate for a\n"
420 " cluster filesystem and inappropriate at almost all other times.\n\n"
423 static int udb_setup(char *str)
425 printk("udb%s specified on command line is almost certainly a ubd -> "
430 __setup("udb", udb_setup);
431 __uml_help(udb_setup,
433 " This option is here solely to catch ubd -> udb typos, which can be\n"
434 " to impossible to catch visually unless you specifically look for\n"
435 " them. The only result of any option starting with 'udb' is an error\n"
436 " in the boot output.\n\n"
439 static void do_ubd_request(struct request_queue * q);
441 /* Only changed by ubd_init, which is an initcall. */
442 static int thread_fd = -1;
443 static LIST_HEAD(restart);
445 /* XXX - move this inside ubd_intr. */
446 /* Called without dev->lock held, and only in interrupt context. */
447 static void ubd_handler(void)
449 struct io_thread_req *req;
451 struct list_head *list, *next_ele;
456 n = os_read_file(thread_fd, &req,
457 sizeof(struct io_thread_req *));
458 if(n != sizeof(req)){
461 printk(KERN_ERR "spurious interrupt in ubd_handler, "
466 blk_end_request(req->req, 0, req->length);
469 reactivate_fd(thread_fd, UBD_IRQ);
471 list_for_each_safe(list, next_ele, &restart){
472 ubd = container_of(list, struct ubd, restart);
473 list_del_init(&ubd->restart);
474 spin_lock_irqsave(&ubd->lock, flags);
475 do_ubd_request(ubd->queue);
476 spin_unlock_irqrestore(&ubd->lock, flags);
480 static irqreturn_t ubd_intr(int irq, void *dev)
486 /* Only changed by ubd_init, which is an initcall. */
487 static int io_pid = -1;
489 static void kill_io_thread(void)
492 os_kill_process(io_pid, 1);
495 __uml_exitcall(kill_io_thread);
497 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
507 unsigned long long size;
511 if (ubd_dev->file && ubd_dev->cow.file) {
512 file = ubd_dev->cow.file;
517 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
521 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
522 &mtime, &size, §or_size, &align, &bitmap_offset);
526 file = ubd_dev->file;
531 return os_file_size(file, size_out);
534 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
538 err = os_pread_file(fd, buf, len, offset);
545 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
547 unsigned long modtime;
548 unsigned long long actual;
551 err = os_file_modtime(file, &modtime);
553 printk(KERN_ERR "Failed to get modification time of backing "
554 "file \"%s\", err = %d\n", file, -err);
558 err = os_file_size(file, &actual);
560 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
561 "err = %d\n", file, -err);
565 if (actual != size) {
566 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
568 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
569 "vs backing file\n", (unsigned long long) size, actual);
572 if (modtime != mtime) {
573 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
574 "backing file\n", mtime, modtime);
580 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
582 struct uml_stat buf1, buf2;
585 if (from_cmdline == NULL)
587 if (!strcmp(from_cmdline, from_cow))
590 err = os_stat_file(from_cmdline, &buf1);
592 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
596 err = os_stat_file(from_cow, &buf2);
598 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
602 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
605 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
606 "\"%s\" specified in COW header of \"%s\"\n",
607 from_cmdline, from_cow, cow);
611 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
612 char **backing_file_out, int *bitmap_offset_out,
613 unsigned long *bitmap_len_out, int *data_offset_out,
617 unsigned long long size;
618 __u32 version, align;
620 int fd, err, sectorsize, asked_switch, mode = 0644;
622 fd = os_open_file(file, *openflags, mode);
624 if ((fd == -ENOENT) && (create_cow_out != NULL))
627 ((fd != -EROFS) && (fd != -EACCES)))
630 fd = os_open_file(file, *openflags, mode);
636 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
638 err = os_lock_file(fd, openflags->w);
640 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
646 /* Successful return case! */
647 if (backing_file_out == NULL)
650 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
651 &size, §orsize, &align, bitmap_offset_out);
652 if (err && (*backing_file_out != NULL)) {
653 printk(KERN_ERR "Failed to read COW header from COW file "
654 "\"%s\", errno = %d\n", file, -err);
660 asked_switch = path_requires_switch(*backing_file_out, backing_file,
663 /* Allow switching only if no mismatch. */
664 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
666 printk(KERN_ERR "Switching backing file to '%s'\n",
668 err = write_cow_header(file, fd, *backing_file_out,
669 sectorsize, align, &size);
671 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
675 *backing_file_out = backing_file;
676 err = backing_file_mismatch(*backing_file_out, size, mtime);
681 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
682 bitmap_len_out, data_offset_out);
690 static int create_cow_file(char *cow_file, char *backing_file,
691 struct openflags flags,
692 int sectorsize, int alignment, int *bitmap_offset_out,
693 unsigned long *bitmap_len_out, int *data_offset_out)
698 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
701 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
706 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
707 bitmap_offset_out, bitmap_len_out,
716 static void ubd_close_dev(struct ubd *ubd_dev)
718 os_close_file(ubd_dev->fd);
719 if(ubd_dev->cow.file == NULL)
722 os_close_file(ubd_dev->cow.fd);
723 vfree(ubd_dev->cow.bitmap);
724 ubd_dev->cow.bitmap = NULL;
727 static int ubd_open_dev(struct ubd *ubd_dev)
729 struct openflags flags;
731 int err, create_cow, *create_ptr;
734 ubd_dev->openflags = ubd_dev->boot_openflags;
736 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
737 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
739 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
740 back_ptr, &ubd_dev->cow.bitmap_offset,
741 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
744 if((fd == -ENOENT) && create_cow){
745 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
746 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
747 &ubd_dev->cow.bitmap_offset,
748 &ubd_dev->cow.bitmap_len,
749 &ubd_dev->cow.data_offset);
751 printk(KERN_INFO "Creating \"%s\" as COW file for "
752 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
757 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
763 if(ubd_dev->cow.file != NULL){
764 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
767 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
768 if(ubd_dev->cow.bitmap == NULL){
769 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
772 flush_tlb_kernel_vm();
774 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
775 ubd_dev->cow.bitmap_offset,
776 ubd_dev->cow.bitmap_len);
780 flags = ubd_dev->openflags;
782 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
783 NULL, NULL, NULL, NULL);
784 if(err < 0) goto error;
785 ubd_dev->cow.fd = err;
789 os_close_file(ubd_dev->fd);
793 static void ubd_device_release(struct device *dev)
795 struct ubd *ubd_dev = dev_get_drvdata(dev);
797 blk_cleanup_queue(ubd_dev->queue);
798 *ubd_dev = ((struct ubd) DEFAULT_UBD);
801 static int ubd_disk_register(int major, u64 size, int unit,
802 struct gendisk **disk_out)
804 struct gendisk *disk;
806 disk = alloc_disk(1 << UBD_SHIFT);
811 disk->first_minor = unit << UBD_SHIFT;
812 disk->fops = &ubd_blops;
813 set_capacity(disk, size / 512);
814 if (major == UBD_MAJOR)
815 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
817 sprintf(disk->disk_name, "ubd_fake%d", unit);
819 /* sysfs register (not for ide fake devices) */
820 if (major == UBD_MAJOR) {
821 ubd_devs[unit].pdev.id = unit;
822 ubd_devs[unit].pdev.name = DRIVER_NAME;
823 ubd_devs[unit].pdev.dev.release = ubd_device_release;
824 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
825 platform_device_register(&ubd_devs[unit].pdev);
826 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
829 disk->private_data = &ubd_devs[unit];
830 disk->queue = ubd_devs[unit].queue;
837 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
839 static int ubd_add(int n, char **error_out)
841 struct ubd *ubd_dev = &ubd_devs[n];
844 if(ubd_dev->file == NULL)
847 err = ubd_file_size(ubd_dev, &ubd_dev->size);
849 *error_out = "Couldn't determine size of device's file";
853 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
855 INIT_LIST_HEAD(&ubd_dev->restart);
856 sg_init_table(ubd_dev->sg, MAX_SG);
859 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
860 if (ubd_dev->queue == NULL) {
861 *error_out = "Failed to initialize device queue";
864 ubd_dev->queue->queuedata = ubd_dev;
865 blk_queue_write_cache(ubd_dev->queue, true, false);
867 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
868 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
870 *error_out = "Failed to register device";
874 if (fake_major != UBD_MAJOR)
875 ubd_disk_register(fake_major, ubd_dev->size, n,
879 * Perhaps this should also be under the "if (fake_major)" above
880 * using the fake_disk->disk_name
883 make_ide_entries(ubd_gendisk[n]->disk_name);
890 blk_cleanup_queue(ubd_dev->queue);
894 static int ubd_config(char *str, char **error_out)
898 /* This string is possibly broken up and stored, so it's only
899 * freed if ubd_setup_common fails, or if only general options
902 str = kstrdup(str, GFP_KERNEL);
904 *error_out = "Failed to allocate memory";
908 ret = ubd_setup_common(str, &n, error_out);
917 mutex_lock(&ubd_lock);
918 ret = ubd_add(n, error_out);
920 ubd_devs[n].file = NULL;
921 mutex_unlock(&ubd_lock);
931 static int ubd_get_config(char *name, char *str, int size, char **error_out)
936 n = parse_unit(&name);
937 if((n >= MAX_DEV) || (n < 0)){
938 *error_out = "ubd_get_config : device number out of range";
942 ubd_dev = &ubd_devs[n];
943 mutex_lock(&ubd_lock);
945 if(ubd_dev->file == NULL){
946 CONFIG_CHUNK(str, size, len, "", 1);
950 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
952 if(ubd_dev->cow.file != NULL){
953 CONFIG_CHUNK(str, size, len, ",", 0);
954 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
956 else CONFIG_CHUNK(str, size, len, "", 1);
959 mutex_unlock(&ubd_lock);
963 static int ubd_id(char **str, int *start_out, int *end_out)
969 *end_out = MAX_DEV - 1;
973 static int ubd_remove(int n, char **error_out)
975 struct gendisk *disk = ubd_gendisk[n];
979 mutex_lock(&ubd_lock);
981 ubd_dev = &ubd_devs[n];
983 if(ubd_dev->file == NULL)
986 /* you cannot remove a open disk */
988 if(ubd_dev->count > 0)
991 ubd_gendisk[n] = NULL;
997 if(fake_gendisk[n] != NULL){
998 del_gendisk(fake_gendisk[n]);
999 put_disk(fake_gendisk[n]);
1000 fake_gendisk[n] = NULL;
1004 platform_device_unregister(&ubd_dev->pdev);
1006 mutex_unlock(&ubd_lock);
1010 /* All these are called by mconsole in process context and without
1011 * ubd-specific locks. The structure itself is const except for .list.
1013 static struct mc_device ubd_mc = {
1014 .list = LIST_HEAD_INIT(ubd_mc.list),
1016 .config = ubd_config,
1017 .get_config = ubd_get_config,
1019 .remove = ubd_remove,
1022 static int __init ubd_mc_init(void)
1024 mconsole_register_dev(&ubd_mc);
1028 __initcall(ubd_mc_init);
1030 static int __init ubd0_init(void)
1032 struct ubd *ubd_dev = &ubd_devs[0];
1034 mutex_lock(&ubd_lock);
1035 if(ubd_dev->file == NULL)
1036 ubd_dev->file = "root_fs";
1037 mutex_unlock(&ubd_lock);
1042 __initcall(ubd0_init);
1044 /* Used in ubd_init, which is an initcall */
1045 static struct platform_driver ubd_driver = {
1047 .name = DRIVER_NAME,
1051 static int __init ubd_init(void)
1056 if (register_blkdev(UBD_MAJOR, "ubd"))
1059 if (fake_major != UBD_MAJOR) {
1060 char name[sizeof("ubd_nnn\0")];
1062 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1063 if (register_blkdev(fake_major, "ubd"))
1066 platform_driver_register(&ubd_driver);
1067 mutex_lock(&ubd_lock);
1068 for (i = 0; i < MAX_DEV; i++){
1069 err = ubd_add(i, &error);
1071 printk(KERN_ERR "Failed to initialize ubd device %d :"
1074 mutex_unlock(&ubd_lock);
1078 late_initcall(ubd_init);
1080 static int __init ubd_driver_init(void){
1081 unsigned long stack;
1084 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1085 if(global_openflags.s){
1086 printk(KERN_INFO "ubd: Synchronous mode\n");
1087 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1088 * enough. So use anyway the io thread. */
1090 stack = alloc_stack(0, 0);
1091 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1095 "ubd : Failed to start I/O thread (errno = %d) - "
1096 "falling back to synchronous I/O\n", -io_pid);
1100 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1101 0, "ubd", ubd_devs);
1103 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1107 device_initcall(ubd_driver_init);
1109 static int ubd_open(struct block_device *bdev, fmode_t mode)
1111 struct gendisk *disk = bdev->bd_disk;
1112 struct ubd *ubd_dev = disk->private_data;
1115 mutex_lock(&ubd_mutex);
1116 if(ubd_dev->count == 0){
1117 err = ubd_open_dev(ubd_dev);
1119 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1120 disk->disk_name, ubd_dev->file, -err);
1125 set_disk_ro(disk, !ubd_dev->openflags.w);
1127 /* This should no more be needed. And it didn't work anyway to exclude
1128 * read-write remounting of filesystems.*/
1129 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1130 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1134 mutex_unlock(&ubd_mutex);
1138 static void ubd_release(struct gendisk *disk, fmode_t mode)
1140 struct ubd *ubd_dev = disk->private_data;
1142 mutex_lock(&ubd_mutex);
1143 if(--ubd_dev->count == 0)
1144 ubd_close_dev(ubd_dev);
1145 mutex_unlock(&ubd_mutex);
1148 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1149 __u64 *cow_offset, unsigned long *bitmap,
1150 __u64 bitmap_offset, unsigned long *bitmap_words,
1153 __u64 sector = io_offset >> 9;
1154 int i, update_bitmap = 0;
1156 for(i = 0; i < length >> 9; i++){
1157 if(cow_mask != NULL)
1158 ubd_set_bit(i, (unsigned char *) cow_mask);
1159 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1163 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1169 *cow_offset = sector / (sizeof(unsigned long) * 8);
1171 /* This takes care of the case where we're exactly at the end of the
1172 * device, and *cow_offset + 1 is off the end. So, just back it up
1173 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1174 * for the original diagnosis.
1176 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1177 sizeof(unsigned long)) - 1))
1180 bitmap_words[0] = bitmap[*cow_offset];
1181 bitmap_words[1] = bitmap[*cow_offset + 1];
1183 *cow_offset *= sizeof(unsigned long);
1184 *cow_offset += bitmap_offset;
1187 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1188 __u64 bitmap_offset, __u64 bitmap_len)
1190 __u64 sector = req->offset >> 9;
1193 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1194 panic("Operation too long");
1196 if(req->op == UBD_READ) {
1197 for(i = 0; i < req->length >> 9; i++){
1198 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1199 ubd_set_bit(i, (unsigned char *)
1203 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1204 &req->cow_offset, bitmap, bitmap_offset,
1205 req->bitmap_words, bitmap_len);
1208 /* Called with dev->lock held */
1209 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1210 unsigned long long offset, int page_offset,
1211 int len, struct page *page)
1213 struct gendisk *disk = req->rq_disk;
1214 struct ubd *ubd_dev = disk->private_data;
1217 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1219 io_req->fds[1] = ubd_dev->fd;
1220 io_req->cow_offset = -1;
1221 io_req->offset = offset;
1222 io_req->length = len;
1224 io_req->sector_mask = 0;
1226 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1227 io_req->offsets[0] = 0;
1228 io_req->offsets[1] = ubd_dev->cow.data_offset;
1229 io_req->buffer = page_address(page) + page_offset;
1230 io_req->sectorsize = 1 << 9;
1232 if(ubd_dev->cow.file != NULL)
1233 cowify_req(io_req, ubd_dev->cow.bitmap,
1234 ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1238 /* Called with dev->lock held */
1239 static void prepare_flush_request(struct request *req,
1240 struct io_thread_req *io_req)
1242 struct gendisk *disk = req->rq_disk;
1243 struct ubd *ubd_dev = disk->private_data;
1246 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1248 io_req->op = UBD_FLUSH;
1251 static bool submit_request(struct io_thread_req *io_req, struct ubd *dev)
1253 int n = os_write_file(thread_fd, &io_req,
1255 if (n != sizeof(io_req)) {
1257 printk("write to io thread failed, "
1258 "errno = %d\n", -n);
1259 else if (list_empty(&dev->restart))
1260 list_add(&dev->restart, &restart);
1268 /* Called with dev->lock held */
1269 static void do_ubd_request(struct request_queue *q)
1271 struct io_thread_req *io_req;
1272 struct request *req;
1275 struct ubd *dev = q->queuedata;
1276 if(dev->request == NULL){
1277 struct request *req = blk_fetch_request(q);
1282 dev->rq_pos = blk_rq_pos(req);
1284 dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1289 if (req_op(req) == REQ_OP_FLUSH) {
1290 io_req = kmalloc(sizeof(struct io_thread_req),
1292 if (io_req == NULL) {
1293 if (list_empty(&dev->restart))
1294 list_add(&dev->restart, &restart);
1297 prepare_flush_request(req, io_req);
1298 if (submit_request(io_req, dev) == false)
1302 while(dev->start_sg < dev->end_sg){
1303 struct scatterlist *sg = &dev->sg[dev->start_sg];
1305 io_req = kmalloc(sizeof(struct io_thread_req),
1308 if(list_empty(&dev->restart))
1309 list_add(&dev->restart, &restart);
1312 prepare_request(req, io_req,
1313 (unsigned long long)dev->rq_pos << 9,
1314 sg->offset, sg->length, sg_page(sg));
1316 if (submit_request(io_req, dev) == false)
1319 dev->rq_pos += sg->length >> 9;
1323 dev->request = NULL;
1327 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1329 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1333 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1337 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1338 unsigned int cmd, unsigned long arg)
1340 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1341 u16 ubd_id[ATA_ID_WORDS];
1344 struct cdrom_volctrl volume;
1345 case HDIO_GET_IDENTITY:
1346 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1347 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1348 ubd_id[ATA_ID_HEADS] = 128;
1349 ubd_id[ATA_ID_SECTORS] = 32;
1350 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1356 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1358 volume.channel0 = 255;
1359 volume.channel1 = 255;
1360 volume.channel2 = 255;
1361 volume.channel3 = 255;
1362 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1369 static int update_bitmap(struct io_thread_req *req)
1373 if(req->cow_offset == -1)
1376 n = os_pwrite_file(req->fds[1], &req->bitmap_words,
1377 sizeof(req->bitmap_words), req->cow_offset);
1378 if(n != sizeof(req->bitmap_words)){
1379 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1387 static void do_io(struct io_thread_req *req)
1391 int n, nsectors, start, end, bit;
1394 if (req->op == UBD_FLUSH) {
1395 /* fds[0] is always either the rw image or our cow file */
1396 n = os_sync_file(req->fds[0]);
1398 printk("do_io - sync failed err = %d "
1399 "fd = %d\n", -n, req->fds[0]);
1405 nsectors = req->length / req->sectorsize;
1408 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1410 while((end < nsectors) &&
1411 (ubd_test_bit(end, (unsigned char *)
1412 &req->sector_mask) == bit))
1415 off = req->offset + req->offsets[bit] +
1416 start * req->sectorsize;
1417 len = (end - start) * req->sectorsize;
1418 buf = &req->buffer[start * req->sectorsize];
1420 if(req->op == UBD_READ){
1425 n = os_pread_file(req->fds[bit], buf, len, off);
1427 printk("do_io - read failed, err = %d "
1428 "fd = %d\n", -n, req->fds[bit]);
1432 } while((n < len) && (n != 0));
1433 if (n < len) memset(&buf[n], 0, len - n);
1435 n = os_pwrite_file(req->fds[bit], buf, len, off);
1437 printk("do_io - write failed err = %d "
1438 "fd = %d\n", -n, req->fds[bit]);
1445 } while(start < nsectors);
1447 req->error = update_bitmap(req);
1450 /* Changed in start_io_thread, which is serialized by being called only
1451 * from ubd_init, which is an initcall.
1455 /* Only changed by the io thread. XXX: currently unused. */
1456 static int io_count = 0;
1458 int io_thread(void *arg)
1460 struct io_thread_req *req;
1463 os_fix_helper_signals();
1466 n = os_read_file(kernel_fd, &req,
1467 sizeof(struct io_thread_req *));
1468 if(n != sizeof(struct io_thread_req *)){
1470 printk("io_thread - read failed, fd = %d, "
1471 "err = %d\n", kernel_fd, -n);
1473 printk("io_thread - short read, fd = %d, "
1474 "length = %d\n", kernel_fd, n);
1480 n = os_write_file(kernel_fd, &req,
1481 sizeof(struct io_thread_req *));
1482 if(n != sizeof(struct io_thread_req *))
1483 printk("io_thread - write failed, fd = %d, err = %d\n",