2 * Copyright (C) 2015-2016 Anton Ivanov (aivanov@brocade.com)
3 * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
4 * Licensed under the GPL
7 /* 2001-09-28...2002-04-17
8 * Partition stuff by James_McMechan@hotmail.com
9 * old style ubd by setting UBD_SHIFT to 0
10 * 2002-09-27...2002-10-18 massive tinkering for 2.5
11 * partitions have changed in 2.5
12 * 2003-01-29 more tinkering for 2.5.59-1
13 * This should now address the sysfs problems and has
14 * the symlink for devfs to allow for booting with
15 * the common /dev/ubd/discX/... names rather than
16 * only /dev/ubdN/discN this version also has lots of
17 * clean ups preparing for ubd-many.
23 #include <linux/module.h>
24 #include <linux/init.h>
25 #include <linux/blkdev.h>
26 #include <linux/ata.h>
27 #include <linux/hdreg.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/slab.h>
33 #include <linux/vmalloc.h>
34 #include <linux/platform_device.h>
35 #include <linux/scatterlist.h>
36 #include <asm/tlbflush.h>
37 #include <kern_util.h>
38 #include "mconsole_kern.h"
45 enum ubd_req { UBD_READ, UBD_WRITE, UBD_FLUSH };
47 struct io_thread_req {
51 unsigned long offsets[2];
52 unsigned long long offset;
56 unsigned long sector_mask;
57 unsigned long long cow_offset;
58 unsigned long bitmap_words[2];
63 static struct io_thread_req * (*irq_req_buffer)[];
64 static struct io_thread_req *irq_remainder;
65 static int irq_remainder_size;
67 static struct io_thread_req * (*io_req_buffer)[];
68 static struct io_thread_req *io_remainder;
69 static int io_remainder_size;
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);
99 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
101 static int ubd_open(struct block_device *bdev, fmode_t mode);
102 static void ubd_release(struct gendisk *disk, fmode_t mode);
103 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
104 unsigned int cmd, unsigned long arg);
105 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
109 static const struct block_device_operations ubd_blops = {
110 .owner = THIS_MODULE,
112 .release = ubd_release,
114 .getgeo = ubd_getgeo,
117 /* Protected by ubd_lock */
118 static int fake_major = UBD_MAJOR;
119 static struct gendisk *ubd_gendisk[MAX_DEV];
120 static struct gendisk *fake_gendisk[MAX_DEV];
122 #ifdef CONFIG_BLK_DEV_UBD_SYNC
123 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
126 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
129 static struct openflags global_openflags = OPEN_FLAGS;
132 /* backing file name */
134 /* backing file fd */
136 unsigned long *bitmap;
137 unsigned long bitmap_len;
145 struct list_head restart;
146 /* name (and fd, below) of the file opened for writing, either the
147 * backing or the cow file. */
152 struct openflags boot_openflags;
153 struct openflags openflags;
157 struct platform_device pdev;
158 struct request_queue *queue;
160 struct scatterlist sg[MAX_SG];
161 struct request *request;
162 int start_sg, end_sg;
166 #define DEFAULT_COW { \
170 .bitmap_offset = 0, \
174 #define DEFAULT_UBD { \
179 .boot_openflags = OPEN_FLAGS, \
180 .openflags = OPEN_FLAGS, \
183 .cow = DEFAULT_COW, \
184 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
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 void make_ide_entries(const char *dev_name)
213 struct proc_dir_entry *dir, *ent;
216 if(proc_ide_root == NULL) make_proc_ide();
218 dir = proc_mkdir(dev_name, proc_ide);
221 ent = proc_create_single("media", S_IRUGO, dir,
222 fake_ide_media_proc_show);
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 /* Function to read several request pointers at a time
446 * handling fractional reads if (and as) needed
449 static int bulk_req_safe_read(
451 struct io_thread_req * (*request_buffer)[],
452 struct io_thread_req **remainder,
460 if (*remainder_size > 0) {
462 (char *) request_buffer,
463 (char *) remainder, *remainder_size
470 ((char *) request_buffer) + *remainder_size,
471 sizeof(struct io_thread_req *)*max_recs
476 if ((n % sizeof(struct io_thread_req *)) > 0) {
478 * Read somehow returned not a multiple of dword
479 * theoretically possible, but never observed in the
480 * wild, so read routine must be able to handle it
482 *remainder_size = n % sizeof(struct io_thread_req *);
483 WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
486 ((char *) request_buffer) +
487 (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
490 n = n - *remainder_size;
498 /* Called without dev->lock held, and only in interrupt context. */
499 static void ubd_handler(void)
502 struct list_head *list, *next_ele;
508 n = bulk_req_safe_read(
518 printk(KERN_ERR "spurious interrupt in ubd_handler, "
522 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
524 (*irq_req_buffer)[count]->req,
526 (*irq_req_buffer)[count]->length
528 kfree((*irq_req_buffer)[count]);
531 reactivate_fd(thread_fd, UBD_IRQ);
533 list_for_each_safe(list, next_ele, &restart){
534 ubd = container_of(list, struct ubd, restart);
535 list_del_init(&ubd->restart);
536 spin_lock_irqsave(&ubd->lock, flags);
537 do_ubd_request(ubd->queue);
538 spin_unlock_irqrestore(&ubd->lock, flags);
542 static irqreturn_t ubd_intr(int irq, void *dev)
548 /* Only changed by ubd_init, which is an initcall. */
549 static int io_pid = -1;
551 static void kill_io_thread(void)
554 os_kill_process(io_pid, 1);
557 __uml_exitcall(kill_io_thread);
559 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
569 unsigned long long size;
573 if (ubd_dev->file && ubd_dev->cow.file) {
574 file = ubd_dev->cow.file;
579 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
583 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
584 &mtime, &size, §or_size, &align, &bitmap_offset);
588 file = ubd_dev->file;
593 return os_file_size(file, size_out);
596 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
600 err = os_pread_file(fd, buf, len, offset);
607 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
609 unsigned long modtime;
610 unsigned long long actual;
613 err = os_file_modtime(file, &modtime);
615 printk(KERN_ERR "Failed to get modification time of backing "
616 "file \"%s\", err = %d\n", file, -err);
620 err = os_file_size(file, &actual);
622 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
623 "err = %d\n", file, -err);
627 if (actual != size) {
628 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
630 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
631 "vs backing file\n", (unsigned long long) size, actual);
634 if (modtime != mtime) {
635 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
636 "backing file\n", mtime, modtime);
642 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
644 struct uml_stat buf1, buf2;
647 if (from_cmdline == NULL)
649 if (!strcmp(from_cmdline, from_cow))
652 err = os_stat_file(from_cmdline, &buf1);
654 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
658 err = os_stat_file(from_cow, &buf2);
660 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
664 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
667 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
668 "\"%s\" specified in COW header of \"%s\"\n",
669 from_cmdline, from_cow, cow);
673 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
674 char **backing_file_out, int *bitmap_offset_out,
675 unsigned long *bitmap_len_out, int *data_offset_out,
679 unsigned long long size;
680 __u32 version, align;
682 int fd, err, sectorsize, asked_switch, mode = 0644;
684 fd = os_open_file(file, *openflags, mode);
686 if ((fd == -ENOENT) && (create_cow_out != NULL))
689 ((fd != -EROFS) && (fd != -EACCES)))
692 fd = os_open_file(file, *openflags, mode);
698 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
700 err = os_lock_file(fd, openflags->w);
702 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
708 /* Successful return case! */
709 if (backing_file_out == NULL)
712 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
713 &size, §orsize, &align, bitmap_offset_out);
714 if (err && (*backing_file_out != NULL)) {
715 printk(KERN_ERR "Failed to read COW header from COW file "
716 "\"%s\", errno = %d\n", file, -err);
722 asked_switch = path_requires_switch(*backing_file_out, backing_file,
725 /* Allow switching only if no mismatch. */
726 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
728 printk(KERN_ERR "Switching backing file to '%s'\n",
730 err = write_cow_header(file, fd, *backing_file_out,
731 sectorsize, align, &size);
733 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
737 *backing_file_out = backing_file;
738 err = backing_file_mismatch(*backing_file_out, size, mtime);
743 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
744 bitmap_len_out, data_offset_out);
752 static int create_cow_file(char *cow_file, char *backing_file,
753 struct openflags flags,
754 int sectorsize, int alignment, int *bitmap_offset_out,
755 unsigned long *bitmap_len_out, int *data_offset_out)
760 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
763 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
768 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
769 bitmap_offset_out, bitmap_len_out,
778 static void ubd_close_dev(struct ubd *ubd_dev)
780 os_close_file(ubd_dev->fd);
781 if(ubd_dev->cow.file == NULL)
784 os_close_file(ubd_dev->cow.fd);
785 vfree(ubd_dev->cow.bitmap);
786 ubd_dev->cow.bitmap = NULL;
789 static int ubd_open_dev(struct ubd *ubd_dev)
791 struct openflags flags;
793 int err, create_cow, *create_ptr;
796 ubd_dev->openflags = ubd_dev->boot_openflags;
798 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
799 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
801 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
802 back_ptr, &ubd_dev->cow.bitmap_offset,
803 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
806 if((fd == -ENOENT) && create_cow){
807 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
808 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
809 &ubd_dev->cow.bitmap_offset,
810 &ubd_dev->cow.bitmap_len,
811 &ubd_dev->cow.data_offset);
813 printk(KERN_INFO "Creating \"%s\" as COW file for "
814 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
819 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
825 if(ubd_dev->cow.file != NULL){
826 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
829 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
830 if(ubd_dev->cow.bitmap == NULL){
831 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
834 flush_tlb_kernel_vm();
836 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
837 ubd_dev->cow.bitmap_offset,
838 ubd_dev->cow.bitmap_len);
842 flags = ubd_dev->openflags;
844 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
845 NULL, NULL, NULL, NULL);
846 if(err < 0) goto error;
847 ubd_dev->cow.fd = err;
851 os_close_file(ubd_dev->fd);
855 static void ubd_device_release(struct device *dev)
857 struct ubd *ubd_dev = dev_get_drvdata(dev);
859 blk_cleanup_queue(ubd_dev->queue);
860 *ubd_dev = ((struct ubd) DEFAULT_UBD);
863 static int ubd_disk_register(int major, u64 size, int unit,
864 struct gendisk **disk_out)
866 struct device *parent = NULL;
867 struct gendisk *disk;
869 disk = alloc_disk(1 << UBD_SHIFT);
874 disk->first_minor = unit << UBD_SHIFT;
875 disk->fops = &ubd_blops;
876 set_capacity(disk, size / 512);
877 if (major == UBD_MAJOR)
878 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
880 sprintf(disk->disk_name, "ubd_fake%d", unit);
882 /* sysfs register (not for ide fake devices) */
883 if (major == UBD_MAJOR) {
884 ubd_devs[unit].pdev.id = unit;
885 ubd_devs[unit].pdev.name = DRIVER_NAME;
886 ubd_devs[unit].pdev.dev.release = ubd_device_release;
887 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
888 platform_device_register(&ubd_devs[unit].pdev);
889 parent = &ubd_devs[unit].pdev.dev;
892 disk->private_data = &ubd_devs[unit];
893 disk->queue = ubd_devs[unit].queue;
894 device_add_disk(parent, disk, NULL);
900 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
902 static int ubd_add(int n, char **error_out)
904 struct ubd *ubd_dev = &ubd_devs[n];
907 if(ubd_dev->file == NULL)
910 err = ubd_file_size(ubd_dev, &ubd_dev->size);
912 *error_out = "Couldn't determine size of device's file";
916 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
918 INIT_LIST_HEAD(&ubd_dev->restart);
919 sg_init_table(ubd_dev->sg, MAX_SG);
922 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
923 if (ubd_dev->queue == NULL) {
924 *error_out = "Failed to initialize device queue";
927 ubd_dev->queue->queuedata = ubd_dev;
928 blk_queue_write_cache(ubd_dev->queue, true, false);
930 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
931 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
933 *error_out = "Failed to register device";
937 if (fake_major != UBD_MAJOR)
938 ubd_disk_register(fake_major, ubd_dev->size, n,
942 * Perhaps this should also be under the "if (fake_major)" above
943 * using the fake_disk->disk_name
946 make_ide_entries(ubd_gendisk[n]->disk_name);
953 blk_cleanup_queue(ubd_dev->queue);
957 static int ubd_config(char *str, char **error_out)
961 /* This string is possibly broken up and stored, so it's only
962 * freed if ubd_setup_common fails, or if only general options
965 str = kstrdup(str, GFP_KERNEL);
967 *error_out = "Failed to allocate memory";
971 ret = ubd_setup_common(str, &n, error_out);
980 mutex_lock(&ubd_lock);
981 ret = ubd_add(n, error_out);
983 ubd_devs[n].file = NULL;
984 mutex_unlock(&ubd_lock);
994 static int ubd_get_config(char *name, char *str, int size, char **error_out)
999 n = parse_unit(&name);
1000 if((n >= MAX_DEV) || (n < 0)){
1001 *error_out = "ubd_get_config : device number out of range";
1005 ubd_dev = &ubd_devs[n];
1006 mutex_lock(&ubd_lock);
1008 if(ubd_dev->file == NULL){
1009 CONFIG_CHUNK(str, size, len, "", 1);
1013 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
1015 if(ubd_dev->cow.file != NULL){
1016 CONFIG_CHUNK(str, size, len, ",", 0);
1017 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
1019 else CONFIG_CHUNK(str, size, len, "", 1);
1022 mutex_unlock(&ubd_lock);
1026 static int ubd_id(char **str, int *start_out, int *end_out)
1030 n = parse_unit(str);
1032 *end_out = MAX_DEV - 1;
1036 static int ubd_remove(int n, char **error_out)
1038 struct gendisk *disk = ubd_gendisk[n];
1039 struct ubd *ubd_dev;
1042 mutex_lock(&ubd_lock);
1044 ubd_dev = &ubd_devs[n];
1046 if(ubd_dev->file == NULL)
1049 /* you cannot remove a open disk */
1051 if(ubd_dev->count > 0)
1054 ubd_gendisk[n] = NULL;
1060 if(fake_gendisk[n] != NULL){
1061 del_gendisk(fake_gendisk[n]);
1062 put_disk(fake_gendisk[n]);
1063 fake_gendisk[n] = NULL;
1067 platform_device_unregister(&ubd_dev->pdev);
1069 mutex_unlock(&ubd_lock);
1073 /* All these are called by mconsole in process context and without
1074 * ubd-specific locks. The structure itself is const except for .list.
1076 static struct mc_device ubd_mc = {
1077 .list = LIST_HEAD_INIT(ubd_mc.list),
1079 .config = ubd_config,
1080 .get_config = ubd_get_config,
1082 .remove = ubd_remove,
1085 static int __init ubd_mc_init(void)
1087 mconsole_register_dev(&ubd_mc);
1091 __initcall(ubd_mc_init);
1093 static int __init ubd0_init(void)
1095 struct ubd *ubd_dev = &ubd_devs[0];
1097 mutex_lock(&ubd_lock);
1098 if(ubd_dev->file == NULL)
1099 ubd_dev->file = "root_fs";
1100 mutex_unlock(&ubd_lock);
1105 __initcall(ubd0_init);
1107 /* Used in ubd_init, which is an initcall */
1108 static struct platform_driver ubd_driver = {
1110 .name = DRIVER_NAME,
1114 static int __init ubd_init(void)
1119 if (register_blkdev(UBD_MAJOR, "ubd"))
1122 if (fake_major != UBD_MAJOR) {
1123 char name[sizeof("ubd_nnn\0")];
1125 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1126 if (register_blkdev(fake_major, "ubd"))
1130 irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1131 sizeof(struct io_thread_req *),
1136 if (irq_req_buffer == NULL) {
1137 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1140 io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1141 sizeof(struct io_thread_req *),
1147 if (io_req_buffer == NULL) {
1148 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1151 platform_driver_register(&ubd_driver);
1152 mutex_lock(&ubd_lock);
1153 for (i = 0; i < MAX_DEV; i++){
1154 err = ubd_add(i, &error);
1156 printk(KERN_ERR "Failed to initialize ubd device %d :"
1159 mutex_unlock(&ubd_lock);
1163 late_initcall(ubd_init);
1165 static int __init ubd_driver_init(void){
1166 unsigned long stack;
1169 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1170 if(global_openflags.s){
1171 printk(KERN_INFO "ubd: Synchronous mode\n");
1172 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1173 * enough. So use anyway the io thread. */
1175 stack = alloc_stack(0, 0);
1176 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1180 "ubd : Failed to start I/O thread (errno = %d) - "
1181 "falling back to synchronous I/O\n", -io_pid);
1185 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1186 0, "ubd", ubd_devs);
1188 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1192 device_initcall(ubd_driver_init);
1194 static int ubd_open(struct block_device *bdev, fmode_t mode)
1196 struct gendisk *disk = bdev->bd_disk;
1197 struct ubd *ubd_dev = disk->private_data;
1200 mutex_lock(&ubd_mutex);
1201 if(ubd_dev->count == 0){
1202 err = ubd_open_dev(ubd_dev);
1204 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1205 disk->disk_name, ubd_dev->file, -err);
1210 set_disk_ro(disk, !ubd_dev->openflags.w);
1212 /* This should no more be needed. And it didn't work anyway to exclude
1213 * read-write remounting of filesystems.*/
1214 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1215 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1219 mutex_unlock(&ubd_mutex);
1223 static void ubd_release(struct gendisk *disk, fmode_t mode)
1225 struct ubd *ubd_dev = disk->private_data;
1227 mutex_lock(&ubd_mutex);
1228 if(--ubd_dev->count == 0)
1229 ubd_close_dev(ubd_dev);
1230 mutex_unlock(&ubd_mutex);
1233 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1234 __u64 *cow_offset, unsigned long *bitmap,
1235 __u64 bitmap_offset, unsigned long *bitmap_words,
1238 __u64 sector = io_offset >> 9;
1239 int i, update_bitmap = 0;
1241 for(i = 0; i < length >> 9; i++){
1242 if(cow_mask != NULL)
1243 ubd_set_bit(i, (unsigned char *) cow_mask);
1244 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1248 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1254 *cow_offset = sector / (sizeof(unsigned long) * 8);
1256 /* This takes care of the case where we're exactly at the end of the
1257 * device, and *cow_offset + 1 is off the end. So, just back it up
1258 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1259 * for the original diagnosis.
1261 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1262 sizeof(unsigned long)) - 1))
1265 bitmap_words[0] = bitmap[*cow_offset];
1266 bitmap_words[1] = bitmap[*cow_offset + 1];
1268 *cow_offset *= sizeof(unsigned long);
1269 *cow_offset += bitmap_offset;
1272 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1273 __u64 bitmap_offset, __u64 bitmap_len)
1275 __u64 sector = req->offset >> 9;
1278 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1279 panic("Operation too long");
1281 if(req->op == UBD_READ) {
1282 for(i = 0; i < req->length >> 9; i++){
1283 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1284 ubd_set_bit(i, (unsigned char *)
1288 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1289 &req->cow_offset, bitmap, bitmap_offset,
1290 req->bitmap_words, bitmap_len);
1293 /* Called with dev->lock held */
1294 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1295 unsigned long long offset, int page_offset,
1296 int len, struct page *page)
1298 struct gendisk *disk = req->rq_disk;
1299 struct ubd *ubd_dev = disk->private_data;
1302 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1304 io_req->fds[1] = ubd_dev->fd;
1305 io_req->cow_offset = -1;
1306 io_req->offset = offset;
1307 io_req->length = len;
1309 io_req->sector_mask = 0;
1311 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1312 io_req->offsets[0] = 0;
1313 io_req->offsets[1] = ubd_dev->cow.data_offset;
1314 io_req->buffer = page_address(page) + page_offset;
1315 io_req->sectorsize = 1 << 9;
1317 if(ubd_dev->cow.file != NULL)
1318 cowify_req(io_req, ubd_dev->cow.bitmap,
1319 ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1323 /* Called with dev->lock held */
1324 static void prepare_flush_request(struct request *req,
1325 struct io_thread_req *io_req)
1327 struct gendisk *disk = req->rq_disk;
1328 struct ubd *ubd_dev = disk->private_data;
1331 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1333 io_req->op = UBD_FLUSH;
1336 static bool submit_request(struct io_thread_req *io_req, struct ubd *dev)
1338 int n = os_write_file(thread_fd, &io_req,
1340 if (n != sizeof(io_req)) {
1342 printk("write to io thread failed, "
1343 "errno = %d\n", -n);
1344 else if (list_empty(&dev->restart))
1345 list_add(&dev->restart, &restart);
1353 /* Called with dev->lock held */
1354 static void do_ubd_request(struct request_queue *q)
1356 struct io_thread_req *io_req;
1357 struct request *req;
1360 struct ubd *dev = q->queuedata;
1361 if(dev->request == NULL){
1362 struct request *req = blk_fetch_request(q);
1367 dev->rq_pos = blk_rq_pos(req);
1369 dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1374 if (req_op(req) == REQ_OP_FLUSH) {
1375 io_req = kmalloc(sizeof(struct io_thread_req),
1377 if (io_req == NULL) {
1378 if (list_empty(&dev->restart))
1379 list_add(&dev->restart, &restart);
1382 prepare_flush_request(req, io_req);
1383 if (submit_request(io_req, dev) == false)
1387 while(dev->start_sg < dev->end_sg){
1388 struct scatterlist *sg = &dev->sg[dev->start_sg];
1390 io_req = kmalloc(sizeof(struct io_thread_req),
1393 if(list_empty(&dev->restart))
1394 list_add(&dev->restart, &restart);
1397 prepare_request(req, io_req,
1398 (unsigned long long)dev->rq_pos << 9,
1399 sg->offset, sg->length, sg_page(sg));
1401 if (submit_request(io_req, dev) == false)
1404 dev->rq_pos += sg->length >> 9;
1408 dev->request = NULL;
1412 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1414 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1418 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1422 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1423 unsigned int cmd, unsigned long arg)
1425 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1426 u16 ubd_id[ATA_ID_WORDS];
1429 struct cdrom_volctrl volume;
1430 case HDIO_GET_IDENTITY:
1431 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1432 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1433 ubd_id[ATA_ID_HEADS] = 128;
1434 ubd_id[ATA_ID_SECTORS] = 32;
1435 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1441 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1443 volume.channel0 = 255;
1444 volume.channel1 = 255;
1445 volume.channel2 = 255;
1446 volume.channel3 = 255;
1447 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1454 static int update_bitmap(struct io_thread_req *req)
1458 if(req->cow_offset == -1)
1461 n = os_pwrite_file(req->fds[1], &req->bitmap_words,
1462 sizeof(req->bitmap_words), req->cow_offset);
1463 if(n != sizeof(req->bitmap_words)){
1464 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1472 static void do_io(struct io_thread_req *req)
1476 int n, nsectors, start, end, bit;
1479 if (req->op == UBD_FLUSH) {
1480 /* fds[0] is always either the rw image or our cow file */
1481 n = os_sync_file(req->fds[0]);
1483 printk("do_io - sync failed err = %d "
1484 "fd = %d\n", -n, req->fds[0]);
1490 nsectors = req->length / req->sectorsize;
1493 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1495 while((end < nsectors) &&
1496 (ubd_test_bit(end, (unsigned char *)
1497 &req->sector_mask) == bit))
1500 off = req->offset + req->offsets[bit] +
1501 start * req->sectorsize;
1502 len = (end - start) * req->sectorsize;
1503 buf = &req->buffer[start * req->sectorsize];
1505 if(req->op == UBD_READ){
1510 n = os_pread_file(req->fds[bit], buf, len, off);
1512 printk("do_io - read failed, err = %d "
1513 "fd = %d\n", -n, req->fds[bit]);
1517 } while((n < len) && (n != 0));
1518 if (n < len) memset(&buf[n], 0, len - n);
1520 n = os_pwrite_file(req->fds[bit], buf, len, off);
1522 printk("do_io - write failed err = %d "
1523 "fd = %d\n", -n, req->fds[bit]);
1530 } while(start < nsectors);
1532 req->error = update_bitmap(req);
1535 /* Changed in start_io_thread, which is serialized by being called only
1536 * from ubd_init, which is an initcall.
1540 /* Only changed by the io thread. XXX: currently unused. */
1541 static int io_count = 0;
1543 int io_thread(void *arg)
1545 int n, count, written, res;
1547 os_fix_helper_signals();
1550 n = bulk_req_safe_read(
1562 printk("io_thread - read failed, fd = %d, "
1565 kernel_fd, -n, io_remainder_size);
1569 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1571 do_io((*io_req_buffer)[count]);
1577 res = os_write_file(kernel_fd,
1578 ((char *) io_req_buffer) + written,
1583 if (res != -EAGAIN) {
1584 printk("io_thread - write failed, fd = %d, "
1585 "err = %d\n", kernel_fd, -n);
1591 } while (written < n);