uml: start fixing os_read_file and os_write_file
[linux-2.6/kvm.git] / arch / um / drivers / ubd_kern.c
blob6d163c9e28857773bc0c0fc05c36e262a04fbb36
1 /*
2 * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3 * Licensed under the GPL
4 */
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.
17 * James McMechan
20 #define MAJOR_NR UBD_MAJOR
21 #define UBD_SHIFT 4
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"
31 #include "linux/mm.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 "asm/segment.h"
38 #include "asm/uaccess.h"
39 #include "asm/irq.h"
40 #include "asm/types.h"
41 #include "asm/tlbflush.h"
42 #include "mem_user.h"
43 #include "kern_util.h"
44 #include "kern.h"
45 #include "mconsole_kern.h"
46 #include "init.h"
47 #include "irq_user.h"
48 #include "irq_kern.h"
49 #include "ubd_user.h"
50 #include "os.h"
51 #include "mem.h"
52 #include "mem_kern.h"
53 #include "cow.h"
55 enum ubd_req { UBD_READ, UBD_WRITE };
57 struct io_thread_req {
58 struct request *req;
59 enum ubd_req op;
60 int fds[2];
61 unsigned long offsets[2];
62 unsigned long long offset;
63 unsigned long length;
64 char *buffer;
65 int sectorsize;
66 unsigned long sector_mask;
67 unsigned long long cow_offset;
68 unsigned long bitmap_words[2];
69 int error;
72 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
73 char **backing_file_out, int *bitmap_offset_out,
74 unsigned long *bitmap_len_out, int *data_offset_out,
75 int *create_cow_out);
76 extern int create_cow_file(char *cow_file, char *backing_file,
77 struct openflags flags, int sectorsize,
78 int alignment, int *bitmap_offset_out,
79 unsigned long *bitmap_len_out,
80 int *data_offset_out);
81 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
82 extern void do_io(struct io_thread_req *req);
84 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
86 __u64 n;
87 int bits, off;
89 bits = sizeof(data[0]) * 8;
90 n = bit / bits;
91 off = bit % bits;
92 return((data[n] & (1 << off)) != 0);
95 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
97 __u64 n;
98 int bits, off;
100 bits = sizeof(data[0]) * 8;
101 n = bit / bits;
102 off = bit % bits;
103 data[n] |= (1 << off);
105 /*End stuff from ubd_user.h*/
107 #define DRIVER_NAME "uml-blkdev"
109 static DEFINE_MUTEX(ubd_lock);
111 static int ubd_open(struct inode * inode, struct file * filp);
112 static int ubd_release(struct inode * inode, struct file * file);
113 static int ubd_ioctl(struct inode * inode, struct file * file,
114 unsigned int cmd, unsigned long arg);
115 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
117 #define MAX_DEV (16)
119 static struct block_device_operations ubd_blops = {
120 .owner = THIS_MODULE,
121 .open = ubd_open,
122 .release = ubd_release,
123 .ioctl = ubd_ioctl,
124 .getgeo = ubd_getgeo,
127 /* Protected by ubd_lock */
128 static int fake_major = MAJOR_NR;
129 static struct gendisk *ubd_gendisk[MAX_DEV];
130 static struct gendisk *fake_gendisk[MAX_DEV];
132 #ifdef CONFIG_BLK_DEV_UBD_SYNC
133 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
134 .cl = 1 })
135 #else
136 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
137 .cl = 1 })
138 #endif
139 static struct openflags global_openflags = OPEN_FLAGS;
141 struct cow {
142 /* backing file name */
143 char *file;
144 /* backing file fd */
145 int fd;
146 unsigned long *bitmap;
147 unsigned long bitmap_len;
148 int bitmap_offset;
149 int data_offset;
152 struct ubd {
153 /* name (and fd, below) of the file opened for writing, either the
154 * backing or the cow file. */
155 char *file;
156 int count;
157 int fd;
158 __u64 size;
159 struct openflags boot_openflags;
160 struct openflags openflags;
161 unsigned shared:1;
162 unsigned no_cow:1;
163 struct cow cow;
164 struct platform_device pdev;
165 struct request_queue *queue;
166 spinlock_t lock;
167 int active;
170 #define DEFAULT_COW { \
171 .file = NULL, \
172 .fd = -1, \
173 .bitmap = NULL, \
174 .bitmap_offset = 0, \
175 .data_offset = 0, \
178 #define DEFAULT_UBD { \
179 .file = NULL, \
180 .count = 0, \
181 .fd = -1, \
182 .size = -1, \
183 .boot_openflags = OPEN_FLAGS, \
184 .openflags = OPEN_FLAGS, \
185 .no_cow = 0, \
186 .shared = 0, \
187 .cow = DEFAULT_COW, \
188 .lock = SPIN_LOCK_UNLOCKED, \
189 .active = 0, \
192 /* Protected by ubd_lock */
193 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
195 /* Only changed by fake_ide_setup which is a setup */
196 static int fake_ide = 0;
197 static struct proc_dir_entry *proc_ide_root = NULL;
198 static struct proc_dir_entry *proc_ide = NULL;
200 static void make_proc_ide(void)
202 proc_ide_root = proc_mkdir("ide", NULL);
203 proc_ide = proc_mkdir("ide0", proc_ide_root);
206 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
207 int *eof, void *data)
209 int len;
211 strcpy(page, "disk\n");
212 len = strlen("disk\n");
213 len -= off;
214 if (len < count){
215 *eof = 1;
216 if (len <= 0) return 0;
218 else len = count;
219 *start = page + off;
220 return len;
223 static void make_ide_entries(char *dev_name)
225 struct proc_dir_entry *dir, *ent;
226 char name[64];
228 if(proc_ide_root == NULL) make_proc_ide();
230 dir = proc_mkdir(dev_name, proc_ide);
231 if(!dir) return;
233 ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
234 if(!ent) return;
235 ent->data = NULL;
236 ent->read_proc = proc_ide_read_media;
237 ent->write_proc = NULL;
238 sprintf(name,"ide0/%s", dev_name);
239 proc_symlink(dev_name, proc_ide_root, name);
242 static int fake_ide_setup(char *str)
244 fake_ide = 1;
245 return(1);
248 __setup("fake_ide", fake_ide_setup);
250 __uml_help(fake_ide_setup,
251 "fake_ide\n"
252 " Create ide0 entries that map onto ubd devices.\n\n"
255 static int parse_unit(char **ptr)
257 char *str = *ptr, *end;
258 int n = -1;
260 if(isdigit(*str)) {
261 n = simple_strtoul(str, &end, 0);
262 if(end == str)
263 return(-1);
264 *ptr = end;
266 else if (('a' <= *str) && (*str <= 'z')) {
267 n = *str - 'a';
268 str++;
269 *ptr = str;
271 return(n);
274 /* If *index_out == -1 at exit, the passed option was a general one;
275 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
276 * should not be freed on exit.
278 static int ubd_setup_common(char *str, int *index_out, char **error_out)
280 struct ubd *ubd_dev;
281 struct openflags flags = global_openflags;
282 char *backing_file;
283 int n, err = 0, i;
285 if(index_out) *index_out = -1;
286 n = *str;
287 if(n == '='){
288 char *end;
289 int major;
291 str++;
292 if(!strcmp(str, "sync")){
293 global_openflags = of_sync(global_openflags);
294 goto out1;
297 err = -EINVAL;
298 major = simple_strtoul(str, &end, 0);
299 if((*end != '\0') || (end == str)){
300 *error_out = "Didn't parse major number";
301 goto out1;
304 mutex_lock(&ubd_lock);
305 if(fake_major != MAJOR_NR){
306 *error_out = "Can't assign a fake major twice";
307 goto out1;
310 fake_major = major;
312 printk(KERN_INFO "Setting extra ubd major number to %d\n",
313 major);
314 err = 0;
315 out1:
316 mutex_unlock(&ubd_lock);
317 return err;
320 n = parse_unit(&str);
321 if(n < 0){
322 *error_out = "Couldn't parse device number";
323 return -EINVAL;
325 if(n >= MAX_DEV){
326 *error_out = "Device number out of range";
327 return 1;
330 err = -EBUSY;
331 mutex_lock(&ubd_lock);
333 ubd_dev = &ubd_devs[n];
334 if(ubd_dev->file != NULL){
335 *error_out = "Device is already configured";
336 goto out;
339 if (index_out)
340 *index_out = n;
342 err = -EINVAL;
343 for (i = 0; i < sizeof("rscd="); i++) {
344 switch (*str) {
345 case 'r':
346 flags.w = 0;
347 break;
348 case 's':
349 flags.s = 1;
350 break;
351 case 'd':
352 ubd_dev->no_cow = 1;
353 break;
354 case 'c':
355 ubd_dev->shared = 1;
356 break;
357 case '=':
358 str++;
359 goto break_loop;
360 default:
361 *error_out = "Expected '=' or flag letter "
362 "(r, s, c, or d)";
363 goto out;
365 str++;
368 if (*str == '=')
369 *error_out = "Too many flags specified";
370 else
371 *error_out = "Missing '='";
372 goto out;
374 break_loop:
375 backing_file = strchr(str, ',');
377 if (backing_file == NULL)
378 backing_file = strchr(str, ':');
380 if(backing_file != NULL){
381 if(ubd_dev->no_cow){
382 *error_out = "Can't specify both 'd' and a cow file";
383 goto out;
385 else {
386 *backing_file = '\0';
387 backing_file++;
390 err = 0;
391 ubd_dev->file = str;
392 ubd_dev->cow.file = backing_file;
393 ubd_dev->boot_openflags = flags;
394 out:
395 mutex_unlock(&ubd_lock);
396 return err;
399 static int ubd_setup(char *str)
401 char *error;
402 int err;
404 err = ubd_setup_common(str, NULL, &error);
405 if(err)
406 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
407 "%s\n", str, error);
408 return 1;
411 __setup("ubd", ubd_setup);
412 __uml_help(ubd_setup,
413 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
414 " This is used to associate a device with a file in the underlying\n"
415 " filesystem. When specifying two filenames, the first one is the\n"
416 " COW name and the second is the backing file name. As separator you can\n"
417 " use either a ':' or a ',': the first one allows writing things like;\n"
418 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
419 " while with a ',' the shell would not expand the 2nd '~'.\n"
420 " When using only one filename, UML will detect whether to treat it like\n"
421 " a COW file or a backing file. To override this detection, add the 'd'\n"
422 " flag:\n"
423 " ubd0d=BackingFile\n"
424 " Usually, there is a filesystem in the file, but \n"
425 " that's not required. Swap devices containing swap files can be\n"
426 " specified like this. Also, a file which doesn't contain a\n"
427 " filesystem can have its contents read in the virtual \n"
428 " machine by running 'dd' on the device. <n> must be in the range\n"
429 " 0 to 7. Appending an 'r' to the number will cause that device\n"
430 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
431 " an 's' will cause data to be written to disk on the host immediately.\n\n"
434 static int udb_setup(char *str)
436 printk("udb%s specified on command line is almost certainly a ubd -> "
437 "udb TYPO\n", str);
438 return(1);
441 __setup("udb", udb_setup);
442 __uml_help(udb_setup,
443 "udb\n"
444 " This option is here solely to catch ubd -> udb typos, which can be\n"
445 " to impossible to catch visually unless you specifically look for\n"
446 " them. The only result of any option starting with 'udb' is an error\n"
447 " in the boot output.\n\n"
450 static int fakehd_set = 0;
451 static int fakehd(char *str)
453 printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
454 fakehd_set = 1;
455 return 1;
458 __setup("fakehd", fakehd);
459 __uml_help(fakehd,
460 "fakehd\n"
461 " Change the ubd device name to \"hd\".\n\n"
464 static void do_ubd_request(request_queue_t * q);
466 /* Only changed by ubd_init, which is an initcall. */
467 int thread_fd = -1;
469 /* call ubd_finish if you need to serialize */
470 static void __ubd_finish(struct request *req, int error)
472 int nsect;
474 if(error){
475 end_request(req, 0);
476 return;
478 nsect = req->current_nr_sectors;
479 req->sector += nsect;
480 req->buffer += nsect << 9;
481 req->errors = 0;
482 req->nr_sectors -= nsect;
483 req->current_nr_sectors = 0;
484 end_request(req, 1);
487 /* Callable only from interrupt context - otherwise you need to do
488 * spin_lock_irq()/spin_lock_irqsave() */
489 static inline void ubd_finish(struct request *req, int error)
491 struct ubd *dev = req->rq_disk->private_data;
493 spin_lock(&dev->lock);
494 __ubd_finish(req, error);
495 spin_unlock(&dev->lock);
498 /* XXX - move this inside ubd_intr. */
499 /* Called without dev->lock held, and only in interrupt context. */
500 static void ubd_handler(void)
502 struct io_thread_req req;
503 struct request *rq;
504 struct ubd *dev;
505 int n;
507 n = os_read_file_k(thread_fd, &req, sizeof(req));
508 if(n != sizeof(req)){
509 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
510 "err = %d\n", os_getpid(), -n);
511 return;
514 rq = req.req;
515 dev = rq->rq_disk->private_data;
516 dev->active = 0;
518 ubd_finish(rq, req.error);
519 reactivate_fd(thread_fd, UBD_IRQ);
520 spin_lock(&dev->lock);
521 do_ubd_request(dev->queue);
522 spin_unlock(&dev->lock);
525 static irqreturn_t ubd_intr(int irq, void *dev)
527 ubd_handler();
528 return(IRQ_HANDLED);
531 /* Only changed by ubd_init, which is an initcall. */
532 static int io_pid = -1;
534 void kill_io_thread(void)
536 if(io_pid != -1)
537 os_kill_process(io_pid, 1);
540 __uml_exitcall(kill_io_thread);
542 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
544 char *file;
546 file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
547 return(os_file_size(file, size_out));
550 static void ubd_close_dev(struct ubd *ubd_dev)
552 os_close_file(ubd_dev->fd);
553 if(ubd_dev->cow.file == NULL)
554 return;
556 os_close_file(ubd_dev->cow.fd);
557 vfree(ubd_dev->cow.bitmap);
558 ubd_dev->cow.bitmap = NULL;
561 static int ubd_open_dev(struct ubd *ubd_dev)
563 struct openflags flags;
564 char **back_ptr;
565 int err, create_cow, *create_ptr;
566 int fd;
568 ubd_dev->openflags = ubd_dev->boot_openflags;
569 create_cow = 0;
570 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
571 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
573 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
574 back_ptr, &ubd_dev->cow.bitmap_offset,
575 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
576 create_ptr);
578 if((fd == -ENOENT) && create_cow){
579 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
580 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
581 &ubd_dev->cow.bitmap_offset,
582 &ubd_dev->cow.bitmap_len,
583 &ubd_dev->cow.data_offset);
584 if(fd >= 0){
585 printk(KERN_INFO "Creating \"%s\" as COW file for "
586 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
590 if(fd < 0){
591 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
592 -fd);
593 return fd;
595 ubd_dev->fd = fd;
597 if(ubd_dev->cow.file != NULL){
598 err = -ENOMEM;
599 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
600 if(ubd_dev->cow.bitmap == NULL){
601 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
602 goto error;
604 flush_tlb_kernel_vm();
606 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
607 ubd_dev->cow.bitmap_offset,
608 ubd_dev->cow.bitmap_len);
609 if(err < 0)
610 goto error;
612 flags = ubd_dev->openflags;
613 flags.w = 0;
614 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
615 NULL, NULL, NULL, NULL);
616 if(err < 0) goto error;
617 ubd_dev->cow.fd = err;
619 return(0);
620 error:
621 os_close_file(ubd_dev->fd);
622 return(err);
625 static void ubd_device_release(struct device *dev)
627 struct ubd *ubd_dev = dev->driver_data;
629 blk_cleanup_queue(ubd_dev->queue);
630 *ubd_dev = ((struct ubd) DEFAULT_UBD);
633 static int ubd_disk_register(int major, u64 size, int unit,
634 struct gendisk **disk_out)
636 struct gendisk *disk;
638 disk = alloc_disk(1 << UBD_SHIFT);
639 if(disk == NULL)
640 return(-ENOMEM);
642 disk->major = major;
643 disk->first_minor = unit << UBD_SHIFT;
644 disk->fops = &ubd_blops;
645 set_capacity(disk, size / 512);
646 if(major == MAJOR_NR)
647 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
648 else
649 sprintf(disk->disk_name, "ubd_fake%d", unit);
651 /* sysfs register (not for ide fake devices) */
652 if (major == MAJOR_NR) {
653 ubd_devs[unit].pdev.id = unit;
654 ubd_devs[unit].pdev.name = DRIVER_NAME;
655 ubd_devs[unit].pdev.dev.release = ubd_device_release;
656 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
657 platform_device_register(&ubd_devs[unit].pdev);
658 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
661 disk->private_data = &ubd_devs[unit];
662 disk->queue = ubd_devs[unit].queue;
663 add_disk(disk);
665 *disk_out = disk;
666 return 0;
669 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
671 static int ubd_add(int n, char **error_out)
673 struct ubd *ubd_dev = &ubd_devs[n];
674 int err = 0;
676 if(ubd_dev->file == NULL)
677 goto out;
679 err = ubd_file_size(ubd_dev, &ubd_dev->size);
680 if(err < 0){
681 *error_out = "Couldn't determine size of device's file";
682 goto out;
685 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
687 err = -ENOMEM;
688 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
689 if (ubd_dev->queue == NULL) {
690 *error_out = "Failed to initialize device queue";
691 goto out;
693 ubd_dev->queue->queuedata = ubd_dev;
695 err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
696 if(err){
697 *error_out = "Failed to register device";
698 goto out_cleanup;
701 if(fake_major != MAJOR_NR)
702 ubd_disk_register(fake_major, ubd_dev->size, n,
703 &fake_gendisk[n]);
705 /* perhaps this should also be under the "if (fake_major)" above */
706 /* using the fake_disk->disk_name and also the fakehd_set name */
707 if (fake_ide)
708 make_ide_entries(ubd_gendisk[n]->disk_name);
710 err = 0;
711 out:
712 return err;
714 out_cleanup:
715 blk_cleanup_queue(ubd_dev->queue);
716 goto out;
719 static int ubd_config(char *str, char **error_out)
721 int n, ret;
723 /* This string is possibly broken up and stored, so it's only
724 * freed if ubd_setup_common fails, or if only general options
725 * were set.
727 str = kstrdup(str, GFP_KERNEL);
728 if (str == NULL) {
729 *error_out = "Failed to allocate memory";
730 return -ENOMEM;
733 ret = ubd_setup_common(str, &n, error_out);
734 if (ret)
735 goto err_free;
737 if (n == -1) {
738 ret = 0;
739 goto err_free;
742 mutex_lock(&ubd_lock);
743 ret = ubd_add(n, error_out);
744 if (ret)
745 ubd_devs[n].file = NULL;
746 mutex_unlock(&ubd_lock);
748 out:
749 return ret;
751 err_free:
752 kfree(str);
753 goto out;
756 static int ubd_get_config(char *name, char *str, int size, char **error_out)
758 struct ubd *ubd_dev;
759 int n, len = 0;
761 n = parse_unit(&name);
762 if((n >= MAX_DEV) || (n < 0)){
763 *error_out = "ubd_get_config : device number out of range";
764 return(-1);
767 ubd_dev = &ubd_devs[n];
768 mutex_lock(&ubd_lock);
770 if(ubd_dev->file == NULL){
771 CONFIG_CHUNK(str, size, len, "", 1);
772 goto out;
775 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
777 if(ubd_dev->cow.file != NULL){
778 CONFIG_CHUNK(str, size, len, ",", 0);
779 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
781 else CONFIG_CHUNK(str, size, len, "", 1);
783 out:
784 mutex_unlock(&ubd_lock);
785 return(len);
788 static int ubd_id(char **str, int *start_out, int *end_out)
790 int n;
792 n = parse_unit(str);
793 *start_out = 0;
794 *end_out = MAX_DEV - 1;
795 return n;
798 static int ubd_remove(int n, char **error_out)
800 struct gendisk *disk = ubd_gendisk[n];
801 struct ubd *ubd_dev;
802 int err = -ENODEV;
804 mutex_lock(&ubd_lock);
806 ubd_dev = &ubd_devs[n];
808 if(ubd_dev->file == NULL)
809 goto out;
811 /* you cannot remove a open disk */
812 err = -EBUSY;
813 if(ubd_dev->count > 0)
814 goto out;
816 ubd_gendisk[n] = NULL;
817 if(disk != NULL){
818 del_gendisk(disk);
819 put_disk(disk);
822 if(fake_gendisk[n] != NULL){
823 del_gendisk(fake_gendisk[n]);
824 put_disk(fake_gendisk[n]);
825 fake_gendisk[n] = NULL;
828 err = 0;
829 platform_device_unregister(&ubd_dev->pdev);
830 out:
831 mutex_unlock(&ubd_lock);
832 return err;
835 /* All these are called by mconsole in process context and without
836 * ubd-specific locks. The structure itself is const except for .list.
838 static struct mc_device ubd_mc = {
839 .list = LIST_HEAD_INIT(ubd_mc.list),
840 .name = "ubd",
841 .config = ubd_config,
842 .get_config = ubd_get_config,
843 .id = ubd_id,
844 .remove = ubd_remove,
847 static int __init ubd_mc_init(void)
849 mconsole_register_dev(&ubd_mc);
850 return 0;
853 __initcall(ubd_mc_init);
855 static int __init ubd0_init(void)
857 struct ubd *ubd_dev = &ubd_devs[0];
859 mutex_lock(&ubd_lock);
860 if(ubd_dev->file == NULL)
861 ubd_dev->file = "root_fs";
862 mutex_unlock(&ubd_lock);
864 return(0);
867 __initcall(ubd0_init);
869 /* Used in ubd_init, which is an initcall */
870 static struct platform_driver ubd_driver = {
871 .driver = {
872 .name = DRIVER_NAME,
876 static int __init ubd_init(void)
878 char *error;
879 int i, err;
881 if (register_blkdev(MAJOR_NR, "ubd"))
882 return -1;
884 if (fake_major != MAJOR_NR) {
885 char name[sizeof("ubd_nnn\0")];
887 snprintf(name, sizeof(name), "ubd_%d", fake_major);
888 if (register_blkdev(fake_major, "ubd"))
889 return -1;
891 platform_driver_register(&ubd_driver);
892 mutex_lock(&ubd_lock);
893 for (i = 0; i < MAX_DEV; i++){
894 err = ubd_add(i, &error);
895 if(err)
896 printk(KERN_ERR "Failed to initialize ubd device %d :"
897 "%s\n", i, error);
899 mutex_unlock(&ubd_lock);
900 return 0;
903 late_initcall(ubd_init);
905 static int __init ubd_driver_init(void){
906 unsigned long stack;
907 int err;
909 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
910 if(global_openflags.s){
911 printk(KERN_INFO "ubd: Synchronous mode\n");
912 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
913 * enough. So use anyway the io thread. */
915 stack = alloc_stack(0, 0);
916 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
917 &thread_fd);
918 if(io_pid < 0){
919 printk(KERN_ERR
920 "ubd : Failed to start I/O thread (errno = %d) - "
921 "falling back to synchronous I/O\n", -io_pid);
922 io_pid = -1;
923 return(0);
925 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
926 IRQF_DISABLED, "ubd", ubd_devs);
927 if(err != 0)
928 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
929 return 0;
932 device_initcall(ubd_driver_init);
934 static int ubd_open(struct inode *inode, struct file *filp)
936 struct gendisk *disk = inode->i_bdev->bd_disk;
937 struct ubd *ubd_dev = disk->private_data;
938 int err = 0;
940 if(ubd_dev->count == 0){
941 err = ubd_open_dev(ubd_dev);
942 if(err){
943 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
944 disk->disk_name, ubd_dev->file, -err);
945 goto out;
948 ubd_dev->count++;
949 set_disk_ro(disk, !ubd_dev->openflags.w);
951 /* This should no more be needed. And it didn't work anyway to exclude
952 * read-write remounting of filesystems.*/
953 /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
954 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
955 err = -EROFS;
957 out:
958 return(err);
961 static int ubd_release(struct inode * inode, struct file * file)
963 struct gendisk *disk = inode->i_bdev->bd_disk;
964 struct ubd *ubd_dev = disk->private_data;
966 if(--ubd_dev->count == 0)
967 ubd_close_dev(ubd_dev);
968 return(0);
971 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
972 __u64 *cow_offset, unsigned long *bitmap,
973 __u64 bitmap_offset, unsigned long *bitmap_words,
974 __u64 bitmap_len)
976 __u64 sector = io_offset >> 9;
977 int i, update_bitmap = 0;
979 for(i = 0; i < length >> 9; i++){
980 if(cow_mask != NULL)
981 ubd_set_bit(i, (unsigned char *) cow_mask);
982 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
983 continue;
985 update_bitmap = 1;
986 ubd_set_bit(sector + i, (unsigned char *) bitmap);
989 if(!update_bitmap)
990 return;
992 *cow_offset = sector / (sizeof(unsigned long) * 8);
994 /* This takes care of the case where we're exactly at the end of the
995 * device, and *cow_offset + 1 is off the end. So, just back it up
996 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
997 * for the original diagnosis.
999 if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
1000 sizeof(unsigned long) - 1))
1001 (*cow_offset)--;
1003 bitmap_words[0] = bitmap[*cow_offset];
1004 bitmap_words[1] = bitmap[*cow_offset + 1];
1006 *cow_offset *= sizeof(unsigned long);
1007 *cow_offset += bitmap_offset;
1010 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1011 __u64 bitmap_offset, __u64 bitmap_len)
1013 __u64 sector = req->offset >> 9;
1014 int i;
1016 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1017 panic("Operation too long");
1019 if(req->op == UBD_READ) {
1020 for(i = 0; i < req->length >> 9; i++){
1021 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1022 ubd_set_bit(i, (unsigned char *)
1023 &req->sector_mask);
1026 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1027 &req->cow_offset, bitmap, bitmap_offset,
1028 req->bitmap_words, bitmap_len);
1031 /* Called with dev->lock held */
1032 static int prepare_request(struct request *req, struct io_thread_req *io_req)
1034 struct gendisk *disk = req->rq_disk;
1035 struct ubd *ubd_dev = disk->private_data;
1036 __u64 offset;
1037 int len;
1039 /* This should be impossible now */
1040 if((rq_data_dir(req) == WRITE) && !ubd_dev->openflags.w){
1041 printk("Write attempted on readonly ubd device %s\n",
1042 disk->disk_name);
1043 end_request(req, 0);
1044 return(1);
1047 offset = ((__u64) req->sector) << 9;
1048 len = req->current_nr_sectors << 9;
1050 io_req->req = req;
1051 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd : ubd_dev->fd;
1052 io_req->fds[1] = ubd_dev->fd;
1053 io_req->cow_offset = -1;
1054 io_req->offset = offset;
1055 io_req->length = len;
1056 io_req->error = 0;
1057 io_req->sector_mask = 0;
1059 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1060 io_req->offsets[0] = 0;
1061 io_req->offsets[1] = ubd_dev->cow.data_offset;
1062 io_req->buffer = req->buffer;
1063 io_req->sectorsize = 1 << 9;
1065 if(ubd_dev->cow.file != NULL)
1066 cowify_req(io_req, ubd_dev->cow.bitmap, ubd_dev->cow.bitmap_offset,
1067 ubd_dev->cow.bitmap_len);
1069 return(0);
1072 /* Called with dev->lock held */
1073 static void do_ubd_request(request_queue_t *q)
1075 struct io_thread_req io_req;
1076 struct request *req;
1077 int err, n;
1079 if(thread_fd == -1){
1080 while((req = elv_next_request(q)) != NULL){
1081 err = prepare_request(req, &io_req);
1082 if(!err){
1083 do_io(&io_req);
1084 __ubd_finish(req, io_req.error);
1088 else {
1089 struct ubd *dev = q->queuedata;
1090 if(dev->active || (req = elv_next_request(q)) == NULL)
1091 return;
1092 err = prepare_request(req, &io_req);
1093 if(!err){
1094 dev->active = 1;
1095 n = os_write_file_k(thread_fd, &io_req, sizeof(io_req));
1096 if(n != sizeof(io_req))
1097 printk("write to io thread failed, "
1098 "errno = %d\n", -n);
1103 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1105 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1107 geo->heads = 128;
1108 geo->sectors = 32;
1109 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1110 return 0;
1113 static int ubd_ioctl(struct inode * inode, struct file * file,
1114 unsigned int cmd, unsigned long arg)
1116 struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1117 struct hd_driveid ubd_id = {
1118 .cyls = 0,
1119 .heads = 128,
1120 .sectors = 32,
1123 switch (cmd) {
1124 struct cdrom_volctrl volume;
1125 case HDIO_GET_IDENTITY:
1126 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1127 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1128 sizeof(ubd_id)))
1129 return(-EFAULT);
1130 return(0);
1132 case CDROMVOLREAD:
1133 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1134 return(-EFAULT);
1135 volume.channel0 = 255;
1136 volume.channel1 = 255;
1137 volume.channel2 = 255;
1138 volume.channel3 = 255;
1139 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1140 return(-EFAULT);
1141 return(0);
1143 return(-EINVAL);
1146 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1148 struct uml_stat buf1, buf2;
1149 int err;
1151 if(from_cmdline == NULL)
1152 return 0;
1153 if(!strcmp(from_cmdline, from_cow))
1154 return 0;
1156 err = os_stat_file(from_cmdline, &buf1);
1157 if(err < 0){
1158 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1159 return 0;
1161 err = os_stat_file(from_cow, &buf2);
1162 if(err < 0){
1163 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1164 return 1;
1166 if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1167 return 0;
1169 printk("Backing file mismatch - \"%s\" requested,\n"
1170 "\"%s\" specified in COW header of \"%s\"\n",
1171 from_cmdline, from_cow, cow);
1172 return 1;
1175 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1177 unsigned long modtime;
1178 unsigned long long actual;
1179 int err;
1181 err = os_file_modtime(file, &modtime);
1182 if(err < 0){
1183 printk("Failed to get modification time of backing file "
1184 "\"%s\", err = %d\n", file, -err);
1185 return(err);
1188 err = os_file_size(file, &actual);
1189 if(err < 0){
1190 printk("Failed to get size of backing file \"%s\", "
1191 "err = %d\n", file, -err);
1192 return(err);
1195 if(actual != size){
1196 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1197 * the typecast.*/
1198 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1199 "file\n", (unsigned long long) size, actual);
1200 return(-EINVAL);
1202 if(modtime != mtime){
1203 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1204 "file\n", mtime, modtime);
1205 return(-EINVAL);
1207 return(0);
1210 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1212 int err;
1214 err = os_seek_file(fd, offset);
1215 if(err < 0)
1216 return(err);
1218 err = os_read_file(fd, buf, len);
1219 if(err < 0)
1220 return(err);
1222 return(0);
1225 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1226 char **backing_file_out, int *bitmap_offset_out,
1227 unsigned long *bitmap_len_out, int *data_offset_out,
1228 int *create_cow_out)
1230 time_t mtime;
1231 unsigned long long size;
1232 __u32 version, align;
1233 char *backing_file;
1234 int fd, err, sectorsize, asked_switch, mode = 0644;
1236 fd = os_open_file(file, *openflags, mode);
1237 if (fd < 0) {
1238 if ((fd == -ENOENT) && (create_cow_out != NULL))
1239 *create_cow_out = 1;
1240 if (!openflags->w ||
1241 ((fd != -EROFS) && (fd != -EACCES)))
1242 return fd;
1243 openflags->w = 0;
1244 fd = os_open_file(file, *openflags, mode);
1245 if (fd < 0)
1246 return fd;
1249 if(shared)
1250 printk("Not locking \"%s\" on the host\n", file);
1251 else {
1252 err = os_lock_file(fd, openflags->w);
1253 if(err < 0){
1254 printk("Failed to lock '%s', err = %d\n", file, -err);
1255 goto out_close;
1259 /* Successful return case! */
1260 if(backing_file_out == NULL)
1261 return(fd);
1263 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1264 &size, &sectorsize, &align, bitmap_offset_out);
1265 if(err && (*backing_file_out != NULL)){
1266 printk("Failed to read COW header from COW file \"%s\", "
1267 "errno = %d\n", file, -err);
1268 goto out_close;
1270 if(err)
1271 return(fd);
1273 asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1275 /* Allow switching only if no mismatch. */
1276 if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1277 printk("Switching backing file to '%s'\n", *backing_file_out);
1278 err = write_cow_header(file, fd, *backing_file_out,
1279 sectorsize, align, &size);
1280 if (err) {
1281 printk("Switch failed, errno = %d\n", -err);
1282 goto out_close;
1284 } else {
1285 *backing_file_out = backing_file;
1286 err = backing_file_mismatch(*backing_file_out, size, mtime);
1287 if (err)
1288 goto out_close;
1291 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1292 bitmap_len_out, data_offset_out);
1294 return fd;
1295 out_close:
1296 os_close_file(fd);
1297 return err;
1300 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1301 int sectorsize, int alignment, int *bitmap_offset_out,
1302 unsigned long *bitmap_len_out, int *data_offset_out)
1304 int err, fd;
1306 flags.c = 1;
1307 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1308 if(fd < 0){
1309 err = fd;
1310 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1311 -err);
1312 goto out;
1315 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1316 bitmap_offset_out, bitmap_len_out,
1317 data_offset_out);
1318 if(!err)
1319 return(fd);
1320 os_close_file(fd);
1321 out:
1322 return(err);
1325 static int update_bitmap(struct io_thread_req *req)
1327 int n;
1329 if(req->cow_offset == -1)
1330 return(0);
1332 n = os_seek_file(req->fds[1], req->cow_offset);
1333 if(n < 0){
1334 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1335 return(1);
1338 n = os_write_file_k(req->fds[1], &req->bitmap_words,
1339 sizeof(req->bitmap_words));
1340 if(n != sizeof(req->bitmap_words)){
1341 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1342 req->fds[1]);
1343 return(1);
1346 return(0);
1349 void do_io(struct io_thread_req *req)
1351 char *buf;
1352 unsigned long len;
1353 int n, nsectors, start, end, bit;
1354 int err;
1355 __u64 off;
1357 nsectors = req->length / req->sectorsize;
1358 start = 0;
1359 do {
1360 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1361 end = start;
1362 while((end < nsectors) &&
1363 (ubd_test_bit(end, (unsigned char *)
1364 &req->sector_mask) == bit))
1365 end++;
1367 off = req->offset + req->offsets[bit] +
1368 start * req->sectorsize;
1369 len = (end - start) * req->sectorsize;
1370 buf = &req->buffer[start * req->sectorsize];
1372 err = os_seek_file(req->fds[bit], off);
1373 if(err < 0){
1374 printk("do_io - lseek failed : err = %d\n", -err);
1375 req->error = 1;
1376 return;
1378 if(req->op == UBD_READ){
1379 n = 0;
1380 do {
1381 buf = &buf[n];
1382 len -= n;
1383 n = os_read_file_k(req->fds[bit], buf, len);
1384 if (n < 0) {
1385 printk("do_io - read failed, err = %d "
1386 "fd = %d\n", -n, req->fds[bit]);
1387 req->error = 1;
1388 return;
1390 } while((n < len) && (n != 0));
1391 if (n < len) memset(&buf[n], 0, len - n);
1392 } else {
1393 n = os_write_file_k(req->fds[bit], buf, len);
1394 if(n != len){
1395 printk("do_io - write failed err = %d "
1396 "fd = %d\n", -n, req->fds[bit]);
1397 req->error = 1;
1398 return;
1402 start = end;
1403 } while(start < nsectors);
1405 req->error = update_bitmap(req);
1408 /* Changed in start_io_thread, which is serialized by being called only
1409 * from ubd_init, which is an initcall.
1411 int kernel_fd = -1;
1413 /* Only changed by the io thread. XXX: currently unused. */
1414 static int io_count = 0;
1416 int io_thread(void *arg)
1418 struct io_thread_req req;
1419 int n;
1421 ignore_sigwinch_sig();
1422 while(1){
1423 n = os_read_file_k(kernel_fd, &req, sizeof(req));
1424 if(n != sizeof(req)){
1425 if(n < 0)
1426 printk("io_thread - read failed, fd = %d, "
1427 "err = %d\n", kernel_fd, -n);
1428 else {
1429 printk("io_thread - short read, fd = %d, "
1430 "length = %d\n", kernel_fd, n);
1432 continue;
1434 io_count++;
1435 do_io(&req);
1436 n = os_write_file_k(kernel_fd, &req, sizeof(req));
1437 if(n != sizeof(req))
1438 printk("io_thread - write failed, fd = %d, err = %d\n",
1439 kernel_fd, -n);
1442 return 0;