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/blkpg.h"
37 #include "linux/genhd.h"
38 #include "linux/spinlock.h"
39 #include "linux/platform_device.h"
40 #include "linux/scatterlist.h"
41 #include "asm/segment.h"
42 #include "asm/uaccess.h"
44 #include "asm/types.h"
45 #include "asm/tlbflush.h"
47 #include "kern_util.h"
49 #include "mconsole_kern.h"
59 enum ubd_req { UBD_READ, UBD_WRITE };
61 struct io_thread_req {
65 unsigned long offsets[2];
66 unsigned long long offset;
70 unsigned long sector_mask;
71 unsigned long long cow_offset;
72 unsigned long bitmap_words[2];
76 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
81 bits = sizeof(data[0]) * 8;
84 return (data[n] & (1 << off)) != 0;
87 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
92 bits = sizeof(data[0]) * 8;
95 data[n] |= (1 << off);
97 /*End stuff from ubd_user.h*/
99 #define DRIVER_NAME "uml-blkdev"
101 static DEFINE_MUTEX(ubd_lock);
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;
167 #define DEFAULT_COW { \
171 .bitmap_offset = 0, \
175 #define DEFAULT_UBD { \
180 .boot_openflags = OPEN_FLAGS, \
181 .openflags = OPEN_FLAGS, \
184 .cow = DEFAULT_COW, \
185 .lock = SPIN_LOCK_UNLOCKED, \
191 /* Protected by ubd_lock */
192 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
194 /* Only changed by fake_ide_setup which is a setup */
195 static int fake_ide = 0;
196 static struct proc_dir_entry *proc_ide_root = NULL;
197 static struct proc_dir_entry *proc_ide = NULL;
199 static void make_proc_ide(void)
201 proc_ide_root = proc_mkdir("ide", NULL);
202 proc_ide = proc_mkdir("ide0", proc_ide_root);
205 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
207 seq_puts(m, "disk\n");
211 static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
213 return single_open(file, fake_ide_media_proc_show, NULL);
216 static const struct file_operations fake_ide_media_proc_fops = {
217 .owner = THIS_MODULE,
218 .open = fake_ide_media_proc_open,
221 .release = single_release,
224 static void make_ide_entries(const char *dev_name)
226 struct proc_dir_entry *dir, *ent;
229 if(proc_ide_root == NULL) make_proc_ide();
231 dir = proc_mkdir(dev_name, proc_ide);
234 ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
236 snprintf(name, sizeof(name), "ide0/%s", dev_name);
237 proc_symlink(dev_name, proc_ide_root, name);
240 static int fake_ide_setup(char *str)
246 __setup("fake_ide", fake_ide_setup);
248 __uml_help(fake_ide_setup,
250 " Create ide0 entries that map onto ubd devices.\n\n"
253 static int parse_unit(char **ptr)
255 char *str = *ptr, *end;
259 n = simple_strtoul(str, &end, 0);
264 else if (('a' <= *str) && (*str <= 'z')) {
272 /* If *index_out == -1 at exit, the passed option was a general one;
273 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
274 * should not be freed on exit.
276 static int ubd_setup_common(char *str, int *index_out, char **error_out)
279 struct openflags flags = global_openflags;
283 if(index_out) *index_out = -1;
290 if(!strcmp(str, "sync")){
291 global_openflags = of_sync(global_openflags);
296 major = simple_strtoul(str, &end, 0);
297 if((*end != '\0') || (end == str)){
298 *error_out = "Didn't parse major number";
302 mutex_lock(&ubd_lock);
303 if (fake_major != UBD_MAJOR) {
304 *error_out = "Can't assign a fake major twice";
310 printk(KERN_INFO "Setting extra ubd major number to %d\n",
314 mutex_unlock(&ubd_lock);
318 n = parse_unit(&str);
320 *error_out = "Couldn't parse device number";
324 *error_out = "Device number out of range";
329 mutex_lock(&ubd_lock);
331 ubd_dev = &ubd_devs[n];
332 if(ubd_dev->file != NULL){
333 *error_out = "Device is already configured";
341 for (i = 0; i < sizeof("rscd="); i++) {
359 *error_out = "Expected '=' or flag letter "
367 *error_out = "Too many flags specified";
369 *error_out = "Missing '='";
373 backing_file = strchr(str, ',');
375 if (backing_file == NULL)
376 backing_file = strchr(str, ':');
378 if(backing_file != NULL){
380 *error_out = "Can't specify both 'd' and a cow file";
384 *backing_file = '\0';
390 ubd_dev->cow.file = backing_file;
391 ubd_dev->boot_openflags = flags;
393 mutex_unlock(&ubd_lock);
397 static int ubd_setup(char *str)
402 err = ubd_setup_common(str, NULL, &error);
404 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
409 __setup("ubd", ubd_setup);
410 __uml_help(ubd_setup,
411 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
412 " This is used to associate a device with a file in the underlying\n"
413 " filesystem. When specifying two filenames, the first one is the\n"
414 " COW name and the second is the backing file name. As separator you can\n"
415 " use either a ':' or a ',': the first one allows writing things like;\n"
416 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
417 " while with a ',' the shell would not expand the 2nd '~'.\n"
418 " When using only one filename, UML will detect whether to treat it like\n"
419 " a COW file or a backing file. To override this detection, add the 'd'\n"
421 " ubd0d=BackingFile\n"
422 " Usually, there is a filesystem in the file, but \n"
423 " that's not required. Swap devices containing swap files can be\n"
424 " specified like this. Also, a file which doesn't contain a\n"
425 " filesystem can have its contents read in the virtual \n"
426 " machine by running 'dd' on the device. <n> must be in the range\n"
427 " 0 to 7. Appending an 'r' to the number will cause that device\n"
428 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
429 " an 's' will cause data to be written to disk on the host immediately.\n"
430 " 'c' will cause the device to be treated as being shared between multiple\n"
431 " UMLs and file locking will be turned off - this is appropriate for a\n"
432 " cluster filesystem and inappropriate at almost all other times.\n\n"
435 static int udb_setup(char *str)
437 printk("udb%s specified on command line is almost certainly a ubd -> "
442 __setup("udb", udb_setup);
443 __uml_help(udb_setup,
445 " This option is here solely to catch ubd -> udb typos, which can be\n"
446 " to impossible to catch visually unless you specifically look for\n"
447 " them. The only result of any option starting with 'udb' is an error\n"
448 " in the boot output.\n\n"
451 static void do_ubd_request(struct request_queue * q);
453 /* Only changed by ubd_init, which is an initcall. */
454 static int thread_fd = -1;
455 static LIST_HEAD(restart);
457 /* XXX - move this inside ubd_intr. */
458 /* Called without dev->lock held, and only in interrupt context. */
459 static void ubd_handler(void)
461 struct io_thread_req *req;
463 struct list_head *list, *next_ele;
468 n = os_read_file(thread_fd, &req,
469 sizeof(struct io_thread_req *));
470 if(n != sizeof(req)){
473 printk(KERN_ERR "spurious interrupt in ubd_handler, "
478 blk_end_request(req->req, 0, req->length);
481 reactivate_fd(thread_fd, UBD_IRQ);
483 list_for_each_safe(list, next_ele, &restart){
484 ubd = container_of(list, struct ubd, restart);
485 list_del_init(&ubd->restart);
486 spin_lock_irqsave(&ubd->lock, flags);
487 do_ubd_request(ubd->queue);
488 spin_unlock_irqrestore(&ubd->lock, flags);
492 static irqreturn_t ubd_intr(int irq, void *dev)
498 /* Only changed by ubd_init, which is an initcall. */
499 static int io_pid = -1;
501 static void kill_io_thread(void)
504 os_kill_process(io_pid, 1);
507 __uml_exitcall(kill_io_thread);
509 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
513 file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
514 return os_file_size(file, size_out);
517 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
521 err = os_seek_file(fd, offset);
525 err = os_read_file(fd, buf, len);
532 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
534 unsigned long modtime;
535 unsigned long long actual;
538 err = os_file_modtime(file, &modtime);
540 printk(KERN_ERR "Failed to get modification time of backing "
541 "file \"%s\", err = %d\n", file, -err);
545 err = os_file_size(file, &actual);
547 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
548 "err = %d\n", file, -err);
552 if (actual != size) {
553 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
555 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
556 "vs backing file\n", (unsigned long long) size, actual);
559 if (modtime != mtime) {
560 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
561 "backing file\n", mtime, modtime);
567 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
569 struct uml_stat buf1, buf2;
572 if (from_cmdline == NULL)
574 if (!strcmp(from_cmdline, from_cow))
577 err = os_stat_file(from_cmdline, &buf1);
579 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
583 err = os_stat_file(from_cow, &buf2);
585 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
589 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
592 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
593 "\"%s\" specified in COW header of \"%s\"\n",
594 from_cmdline, from_cow, cow);
598 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
599 char **backing_file_out, int *bitmap_offset_out,
600 unsigned long *bitmap_len_out, int *data_offset_out,
604 unsigned long long size;
605 __u32 version, align;
607 int fd, err, sectorsize, asked_switch, mode = 0644;
609 fd = os_open_file(file, *openflags, mode);
611 if ((fd == -ENOENT) && (create_cow_out != NULL))
614 ((fd != -EROFS) && (fd != -EACCES)))
617 fd = os_open_file(file, *openflags, mode);
623 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
625 err = os_lock_file(fd, openflags->w);
627 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
633 /* Successful return case! */
634 if (backing_file_out == NULL)
637 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
638 &size, §orsize, &align, bitmap_offset_out);
639 if (err && (*backing_file_out != NULL)) {
640 printk(KERN_ERR "Failed to read COW header from COW file "
641 "\"%s\", errno = %d\n", file, -err);
647 asked_switch = path_requires_switch(*backing_file_out, backing_file,
650 /* Allow switching only if no mismatch. */
651 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
653 printk(KERN_ERR "Switching backing file to '%s'\n",
655 err = write_cow_header(file, fd, *backing_file_out,
656 sectorsize, align, &size);
658 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
662 *backing_file_out = backing_file;
663 err = backing_file_mismatch(*backing_file_out, size, mtime);
668 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
669 bitmap_len_out, data_offset_out);
677 static int create_cow_file(char *cow_file, char *backing_file,
678 struct openflags flags,
679 int sectorsize, int alignment, int *bitmap_offset_out,
680 unsigned long *bitmap_len_out, int *data_offset_out)
685 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
688 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
693 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
694 bitmap_offset_out, bitmap_len_out,
703 static void ubd_close_dev(struct ubd *ubd_dev)
705 os_close_file(ubd_dev->fd);
706 if(ubd_dev->cow.file == NULL)
709 os_close_file(ubd_dev->cow.fd);
710 vfree(ubd_dev->cow.bitmap);
711 ubd_dev->cow.bitmap = NULL;
714 static int ubd_open_dev(struct ubd *ubd_dev)
716 struct openflags flags;
718 int err, create_cow, *create_ptr;
721 ubd_dev->openflags = ubd_dev->boot_openflags;
723 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
724 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
726 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
727 back_ptr, &ubd_dev->cow.bitmap_offset,
728 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
731 if((fd == -ENOENT) && create_cow){
732 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
733 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
734 &ubd_dev->cow.bitmap_offset,
735 &ubd_dev->cow.bitmap_len,
736 &ubd_dev->cow.data_offset);
738 printk(KERN_INFO "Creating \"%s\" as COW file for "
739 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
744 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
750 if(ubd_dev->cow.file != NULL){
751 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
754 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
755 if(ubd_dev->cow.bitmap == NULL){
756 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
759 flush_tlb_kernel_vm();
761 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
762 ubd_dev->cow.bitmap_offset,
763 ubd_dev->cow.bitmap_len);
767 flags = ubd_dev->openflags;
769 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
770 NULL, NULL, NULL, NULL);
771 if(err < 0) goto error;
772 ubd_dev->cow.fd = err;
776 os_close_file(ubd_dev->fd);
780 static void ubd_device_release(struct device *dev)
782 struct ubd *ubd_dev = dev_get_drvdata(dev);
784 blk_cleanup_queue(ubd_dev->queue);
785 *ubd_dev = ((struct ubd) DEFAULT_UBD);
788 static int ubd_disk_register(int major, u64 size, int unit,
789 struct gendisk **disk_out)
791 struct gendisk *disk;
793 disk = alloc_disk(1 << UBD_SHIFT);
798 disk->first_minor = unit << UBD_SHIFT;
799 disk->fops = &ubd_blops;
800 set_capacity(disk, size / 512);
801 if (major == UBD_MAJOR)
802 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
804 sprintf(disk->disk_name, "ubd_fake%d", unit);
806 /* sysfs register (not for ide fake devices) */
807 if (major == UBD_MAJOR) {
808 ubd_devs[unit].pdev.id = unit;
809 ubd_devs[unit].pdev.name = DRIVER_NAME;
810 ubd_devs[unit].pdev.dev.release = ubd_device_release;
811 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
812 platform_device_register(&ubd_devs[unit].pdev);
813 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
816 disk->private_data = &ubd_devs[unit];
817 disk->queue = ubd_devs[unit].queue;
824 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
826 static int ubd_add(int n, char **error_out)
828 struct ubd *ubd_dev = &ubd_devs[n];
831 if(ubd_dev->file == NULL)
834 err = ubd_file_size(ubd_dev, &ubd_dev->size);
836 *error_out = "Couldn't determine size of device's file";
840 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
842 INIT_LIST_HEAD(&ubd_dev->restart);
843 sg_init_table(ubd_dev->sg, MAX_SG);
846 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
847 if (ubd_dev->queue == NULL) {
848 *error_out = "Failed to initialize device queue";
851 ubd_dev->queue->queuedata = ubd_dev;
853 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
854 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
856 *error_out = "Failed to register device";
860 if (fake_major != UBD_MAJOR)
861 ubd_disk_register(fake_major, ubd_dev->size, n,
865 * Perhaps this should also be under the "if (fake_major)" above
866 * using the fake_disk->disk_name
869 make_ide_entries(ubd_gendisk[n]->disk_name);
876 blk_cleanup_queue(ubd_dev->queue);
880 static int ubd_config(char *str, char **error_out)
884 /* This string is possibly broken up and stored, so it's only
885 * freed if ubd_setup_common fails, or if only general options
888 str = kstrdup(str, GFP_KERNEL);
890 *error_out = "Failed to allocate memory";
894 ret = ubd_setup_common(str, &n, error_out);
903 mutex_lock(&ubd_lock);
904 ret = ubd_add(n, error_out);
906 ubd_devs[n].file = NULL;
907 mutex_unlock(&ubd_lock);
917 static int ubd_get_config(char *name, char *str, int size, char **error_out)
922 n = parse_unit(&name);
923 if((n >= MAX_DEV) || (n < 0)){
924 *error_out = "ubd_get_config : device number out of range";
928 ubd_dev = &ubd_devs[n];
929 mutex_lock(&ubd_lock);
931 if(ubd_dev->file == NULL){
932 CONFIG_CHUNK(str, size, len, "", 1);
936 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
938 if(ubd_dev->cow.file != NULL){
939 CONFIG_CHUNK(str, size, len, ",", 0);
940 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
942 else CONFIG_CHUNK(str, size, len, "", 1);
945 mutex_unlock(&ubd_lock);
949 static int ubd_id(char **str, int *start_out, int *end_out)
955 *end_out = MAX_DEV - 1;
959 static int ubd_remove(int n, char **error_out)
961 struct gendisk *disk = ubd_gendisk[n];
965 mutex_lock(&ubd_lock);
967 ubd_dev = &ubd_devs[n];
969 if(ubd_dev->file == NULL)
972 /* you cannot remove a open disk */
974 if(ubd_dev->count > 0)
977 ubd_gendisk[n] = NULL;
983 if(fake_gendisk[n] != NULL){
984 del_gendisk(fake_gendisk[n]);
985 put_disk(fake_gendisk[n]);
986 fake_gendisk[n] = NULL;
990 platform_device_unregister(&ubd_dev->pdev);
992 mutex_unlock(&ubd_lock);
996 /* All these are called by mconsole in process context and without
997 * ubd-specific locks. The structure itself is const except for .list.
999 static struct mc_device ubd_mc = {
1000 .list = LIST_HEAD_INIT(ubd_mc.list),
1002 .config = ubd_config,
1003 .get_config = ubd_get_config,
1005 .remove = ubd_remove,
1008 static int __init ubd_mc_init(void)
1010 mconsole_register_dev(&ubd_mc);
1014 __initcall(ubd_mc_init);
1016 static int __init ubd0_init(void)
1018 struct ubd *ubd_dev = &ubd_devs[0];
1020 mutex_lock(&ubd_lock);
1021 if(ubd_dev->file == NULL)
1022 ubd_dev->file = "root_fs";
1023 mutex_unlock(&ubd_lock);
1028 __initcall(ubd0_init);
1030 /* Used in ubd_init, which is an initcall */
1031 static struct platform_driver ubd_driver = {
1033 .name = DRIVER_NAME,
1037 static int __init ubd_init(void)
1042 if (register_blkdev(UBD_MAJOR, "ubd"))
1045 if (fake_major != UBD_MAJOR) {
1046 char name[sizeof("ubd_nnn\0")];
1048 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1049 if (register_blkdev(fake_major, "ubd"))
1052 platform_driver_register(&ubd_driver);
1053 mutex_lock(&ubd_lock);
1054 for (i = 0; i < MAX_DEV; i++){
1055 err = ubd_add(i, &error);
1057 printk(KERN_ERR "Failed to initialize ubd device %d :"
1060 mutex_unlock(&ubd_lock);
1064 late_initcall(ubd_init);
1066 static int __init ubd_driver_init(void){
1067 unsigned long stack;
1070 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1071 if(global_openflags.s){
1072 printk(KERN_INFO "ubd: Synchronous mode\n");
1073 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1074 * enough. So use anyway the io thread. */
1076 stack = alloc_stack(0, 0);
1077 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1081 "ubd : Failed to start I/O thread (errno = %d) - "
1082 "falling back to synchronous I/O\n", -io_pid);
1086 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1087 IRQF_DISABLED, "ubd", ubd_devs);
1089 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1093 device_initcall(ubd_driver_init);
1095 static int ubd_open(struct block_device *bdev, fmode_t mode)
1097 struct gendisk *disk = bdev->bd_disk;
1098 struct ubd *ubd_dev = disk->private_data;
1101 if(ubd_dev->count == 0){
1102 err = ubd_open_dev(ubd_dev);
1104 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1105 disk->disk_name, ubd_dev->file, -err);
1110 set_disk_ro(disk, !ubd_dev->openflags.w);
1112 /* This should no more be needed. And it didn't work anyway to exclude
1113 * read-write remounting of filesystems.*/
1114 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1115 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1122 static int ubd_release(struct gendisk *disk, fmode_t mode)
1124 struct ubd *ubd_dev = disk->private_data;
1126 if(--ubd_dev->count == 0)
1127 ubd_close_dev(ubd_dev);
1131 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1132 __u64 *cow_offset, unsigned long *bitmap,
1133 __u64 bitmap_offset, unsigned long *bitmap_words,
1136 __u64 sector = io_offset >> 9;
1137 int i, update_bitmap = 0;
1139 for(i = 0; i < length >> 9; i++){
1140 if(cow_mask != NULL)
1141 ubd_set_bit(i, (unsigned char *) cow_mask);
1142 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1146 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1152 *cow_offset = sector / (sizeof(unsigned long) * 8);
1154 /* This takes care of the case where we're exactly at the end of the
1155 * device, and *cow_offset + 1 is off the end. So, just back it up
1156 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1157 * for the original diagnosis.
1159 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1160 sizeof(unsigned long)) - 1))
1163 bitmap_words[0] = bitmap[*cow_offset];
1164 bitmap_words[1] = bitmap[*cow_offset + 1];
1166 *cow_offset *= sizeof(unsigned long);
1167 *cow_offset += bitmap_offset;
1170 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1171 __u64 bitmap_offset, __u64 bitmap_len)
1173 __u64 sector = req->offset >> 9;
1176 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1177 panic("Operation too long");
1179 if(req->op == UBD_READ) {
1180 for(i = 0; i < req->length >> 9; i++){
1181 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1182 ubd_set_bit(i, (unsigned char *)
1186 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1187 &req->cow_offset, bitmap, bitmap_offset,
1188 req->bitmap_words, bitmap_len);
1191 /* Called with dev->lock held */
1192 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1193 unsigned long long offset, int page_offset,
1194 int len, struct page *page)
1196 struct gendisk *disk = req->rq_disk;
1197 struct ubd *ubd_dev = disk->private_data;
1200 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1202 io_req->fds[1] = ubd_dev->fd;
1203 io_req->cow_offset = -1;
1204 io_req->offset = offset;
1205 io_req->length = len;
1207 io_req->sector_mask = 0;
1209 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1210 io_req->offsets[0] = 0;
1211 io_req->offsets[1] = ubd_dev->cow.data_offset;
1212 io_req->buffer = page_address(page) + page_offset;
1213 io_req->sectorsize = 1 << 9;
1215 if(ubd_dev->cow.file != NULL)
1216 cowify_req(io_req, ubd_dev->cow.bitmap,
1217 ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1221 /* Called with dev->lock held */
1222 static void do_ubd_request(struct request_queue *q)
1224 struct io_thread_req *io_req;
1225 struct request *req;
1230 struct ubd *dev = q->queuedata;
1231 if(dev->end_sg == 0){
1232 struct request *req = blk_fetch_request(q);
1238 dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1242 sector = blk_rq_pos(req);
1243 while(dev->start_sg < dev->end_sg){
1244 struct scatterlist *sg = &dev->sg[dev->start_sg];
1246 io_req = kmalloc(sizeof(struct io_thread_req),
1249 if(list_empty(&dev->restart))
1250 list_add(&dev->restart, &restart);
1253 prepare_request(req, io_req,
1254 (unsigned long long)sector << 9,
1255 sg->offset, sg->length, sg_page(sg));
1257 sector += sg->length >> 9;
1258 n = os_write_file(thread_fd, &io_req,
1259 sizeof(struct io_thread_req *));
1260 if(n != sizeof(struct io_thread_req *)){
1262 printk("write to io thread failed, "
1263 "errno = %d\n", -n);
1264 else if(list_empty(&dev->restart))
1265 list_add(&dev->restart, &restart);
1273 dev->request = NULL;
1277 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1279 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1283 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1287 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1288 unsigned int cmd, unsigned long arg)
1290 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1291 u16 ubd_id[ATA_ID_WORDS];
1294 struct cdrom_volctrl volume;
1295 case HDIO_GET_IDENTITY:
1296 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1297 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1298 ubd_id[ATA_ID_HEADS] = 128;
1299 ubd_id[ATA_ID_SECTORS] = 32;
1300 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1306 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1308 volume.channel0 = 255;
1309 volume.channel1 = 255;
1310 volume.channel2 = 255;
1311 volume.channel3 = 255;
1312 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1319 static int update_bitmap(struct io_thread_req *req)
1323 if(req->cow_offset == -1)
1326 n = os_seek_file(req->fds[1], req->cow_offset);
1328 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1332 n = os_write_file(req->fds[1], &req->bitmap_words,
1333 sizeof(req->bitmap_words));
1334 if(n != sizeof(req->bitmap_words)){
1335 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1343 static void do_io(struct io_thread_req *req)
1347 int n, nsectors, start, end, bit;
1351 nsectors = req->length / req->sectorsize;
1354 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1356 while((end < nsectors) &&
1357 (ubd_test_bit(end, (unsigned char *)
1358 &req->sector_mask) == bit))
1361 off = req->offset + req->offsets[bit] +
1362 start * req->sectorsize;
1363 len = (end - start) * req->sectorsize;
1364 buf = &req->buffer[start * req->sectorsize];
1366 err = os_seek_file(req->fds[bit], off);
1368 printk("do_io - lseek failed : err = %d\n", -err);
1372 if(req->op == UBD_READ){
1377 n = os_read_file(req->fds[bit], buf, len);
1379 printk("do_io - read failed, err = %d "
1380 "fd = %d\n", -n, req->fds[bit]);
1384 } while((n < len) && (n != 0));
1385 if (n < len) memset(&buf[n], 0, len - n);
1387 n = os_write_file(req->fds[bit], buf, len);
1389 printk("do_io - write failed err = %d "
1390 "fd = %d\n", -n, req->fds[bit]);
1397 } while(start < nsectors);
1399 req->error = update_bitmap(req);
1402 /* Changed in start_io_thread, which is serialized by being called only
1403 * from ubd_init, which is an initcall.
1407 /* Only changed by the io thread. XXX: currently unused. */
1408 static int io_count = 0;
1410 int io_thread(void *arg)
1412 struct io_thread_req *req;
1415 ignore_sigwinch_sig();
1417 n = os_read_file(kernel_fd, &req,
1418 sizeof(struct io_thread_req *));
1419 if(n != sizeof(struct io_thread_req *)){
1421 printk("io_thread - read failed, fd = %d, "
1422 "err = %d\n", kernel_fd, -n);
1424 printk("io_thread - short read, fd = %d, "
1425 "length = %d\n", kernel_fd, n);
1431 n = os_write_file(kernel_fd, &req,
1432 sizeof(struct io_thread_req *));
1433 if(n != sizeof(struct io_thread_req *))
1434 printk("io_thread - write failed, fd = %d, err = %d\n",