Revert "Fix very high interrupt rate for IRQ8 (rtc) unless pnpacpi=off"
[pv_ops_mirror.git] / arch / um / drivers / ubd_kern.c
blob7e6cdde62ead261e6140c30cc51d8517e0b65aac
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/kernel.h"
24 #include "linux/module.h"
25 #include "linux/blkdev.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/ctype.h"
31 #include "linux/capability.h"
32 #include "linux/mm.h"
33 #include "linux/vmalloc.h"
34 #include "linux/blkpg.h"
35 #include "linux/genhd.h"
36 #include "linux/spinlock.h"
37 #include "linux/platform_device.h"
38 #include "linux/scatterlist.h"
39 #include "asm/segment.h"
40 #include "asm/uaccess.h"
41 #include "asm/irq.h"
42 #include "asm/types.h"
43 #include "asm/tlbflush.h"
44 #include "mem_user.h"
45 #include "kern_util.h"
46 #include "kern.h"
47 #include "mconsole_kern.h"
48 #include "init.h"
49 #include "irq_user.h"
50 #include "irq_kern.h"
51 #include "ubd_user.h"
52 #include "os.h"
53 #include "mem.h"
54 #include "mem_kern.h"
55 #include "cow.h"
57 enum ubd_req { UBD_READ, UBD_WRITE };
59 struct io_thread_req {
60 struct request *req;
61 enum ubd_req op;
62 int fds[2];
63 unsigned long offsets[2];
64 unsigned long long offset;
65 unsigned long length;
66 char *buffer;
67 int sectorsize;
68 unsigned long sector_mask;
69 unsigned long long cow_offset;
70 unsigned long bitmap_words[2];
71 int error;
74 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
75 char **backing_file_out, int *bitmap_offset_out,
76 unsigned long *bitmap_len_out, int *data_offset_out,
77 int *create_cow_out);
78 extern int create_cow_file(char *cow_file, char *backing_file,
79 struct openflags flags, int sectorsize,
80 int alignment, int *bitmap_offset_out,
81 unsigned long *bitmap_len_out,
82 int *data_offset_out);
83 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
84 extern void do_io(struct io_thread_req *req);
86 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
88 __u64 n;
89 int bits, off;
91 bits = sizeof(data[0]) * 8;
92 n = bit / bits;
93 off = bit % bits;
94 return (data[n] & (1 << off)) != 0;
97 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
99 __u64 n;
100 int bits, off;
102 bits = sizeof(data[0]) * 8;
103 n = bit / bits;
104 off = bit % bits;
105 data[n] |= (1 << off);
107 /*End stuff from ubd_user.h*/
109 #define DRIVER_NAME "uml-blkdev"
111 static DEFINE_MUTEX(ubd_lock);
113 static int ubd_open(struct inode * inode, struct file * filp);
114 static int ubd_release(struct inode * inode, struct file * file);
115 static int ubd_ioctl(struct inode * inode, struct file * file,
116 unsigned int cmd, unsigned long arg);
117 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
119 #define MAX_DEV (16)
121 static struct block_device_operations ubd_blops = {
122 .owner = THIS_MODULE,
123 .open = ubd_open,
124 .release = ubd_release,
125 .ioctl = ubd_ioctl,
126 .getgeo = ubd_getgeo,
129 /* Protected by ubd_lock */
130 static int fake_major = MAJOR_NR;
131 static struct gendisk *ubd_gendisk[MAX_DEV];
132 static struct gendisk *fake_gendisk[MAX_DEV];
134 #ifdef CONFIG_BLK_DEV_UBD_SYNC
135 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
136 .cl = 1 })
137 #else
138 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
139 .cl = 1 })
140 #endif
141 static struct openflags global_openflags = OPEN_FLAGS;
143 struct cow {
144 /* backing file name */
145 char *file;
146 /* backing file fd */
147 int fd;
148 unsigned long *bitmap;
149 unsigned long bitmap_len;
150 int bitmap_offset;
151 int data_offset;
154 #define MAX_SG 64
156 struct ubd {
157 struct list_head restart;
158 /* name (and fd, below) of the file opened for writing, either the
159 * backing or the cow file. */
160 char *file;
161 int count;
162 int fd;
163 __u64 size;
164 struct openflags boot_openflags;
165 struct openflags openflags;
166 unsigned shared:1;
167 unsigned no_cow:1;
168 struct cow cow;
169 struct platform_device pdev;
170 struct request_queue *queue;
171 spinlock_t lock;
172 struct scatterlist sg[MAX_SG];
173 struct request *request;
174 int start_sg, end_sg;
177 #define DEFAULT_COW { \
178 .file = NULL, \
179 .fd = -1, \
180 .bitmap = NULL, \
181 .bitmap_offset = 0, \
182 .data_offset = 0, \
185 #define DEFAULT_UBD { \
186 .file = NULL, \
187 .count = 0, \
188 .fd = -1, \
189 .size = -1, \
190 .boot_openflags = OPEN_FLAGS, \
191 .openflags = OPEN_FLAGS, \
192 .no_cow = 0, \
193 .shared = 0, \
194 .cow = DEFAULT_COW, \
195 .lock = SPIN_LOCK_UNLOCKED, \
196 .request = NULL, \
197 .start_sg = 0, \
198 .end_sg = 0, \
201 /* Protected by ubd_lock */
202 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
204 /* Only changed by fake_ide_setup which is a setup */
205 static int fake_ide = 0;
206 static struct proc_dir_entry *proc_ide_root = NULL;
207 static struct proc_dir_entry *proc_ide = NULL;
209 static void make_proc_ide(void)
211 proc_ide_root = proc_mkdir("ide", NULL);
212 proc_ide = proc_mkdir("ide0", proc_ide_root);
215 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
216 int *eof, void *data)
218 int len;
220 strcpy(page, "disk\n");
221 len = strlen("disk\n");
222 len -= off;
223 if (len < count){
224 *eof = 1;
225 if (len <= 0) return 0;
227 else len = count;
228 *start = page + off;
229 return len;
232 static void make_ide_entries(char *dev_name)
234 struct proc_dir_entry *dir, *ent;
235 char name[64];
237 if(proc_ide_root == NULL) make_proc_ide();
239 dir = proc_mkdir(dev_name, proc_ide);
240 if(!dir) return;
242 ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
243 if(!ent) return;
244 ent->data = NULL;
245 ent->read_proc = proc_ide_read_media;
246 ent->write_proc = NULL;
247 sprintf(name,"ide0/%s", dev_name);
248 proc_symlink(dev_name, proc_ide_root, name);
251 static int fake_ide_setup(char *str)
253 fake_ide = 1;
254 return 1;
257 __setup("fake_ide", fake_ide_setup);
259 __uml_help(fake_ide_setup,
260 "fake_ide\n"
261 " Create ide0 entries that map onto ubd devices.\n\n"
264 static int parse_unit(char **ptr)
266 char *str = *ptr, *end;
267 int n = -1;
269 if(isdigit(*str)) {
270 n = simple_strtoul(str, &end, 0);
271 if(end == str)
272 return -1;
273 *ptr = end;
275 else if (('a' <= *str) && (*str <= 'z')) {
276 n = *str - 'a';
277 str++;
278 *ptr = str;
280 return n;
283 /* If *index_out == -1 at exit, the passed option was a general one;
284 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
285 * should not be freed on exit.
287 static int ubd_setup_common(char *str, int *index_out, char **error_out)
289 struct ubd *ubd_dev;
290 struct openflags flags = global_openflags;
291 char *backing_file;
292 int n, err = 0, i;
294 if(index_out) *index_out = -1;
295 n = *str;
296 if(n == '='){
297 char *end;
298 int major;
300 str++;
301 if(!strcmp(str, "sync")){
302 global_openflags = of_sync(global_openflags);
303 goto out1;
306 err = -EINVAL;
307 major = simple_strtoul(str, &end, 0);
308 if((*end != '\0') || (end == str)){
309 *error_out = "Didn't parse major number";
310 goto out1;
313 mutex_lock(&ubd_lock);
314 if(fake_major != MAJOR_NR){
315 *error_out = "Can't assign a fake major twice";
316 goto out1;
319 fake_major = major;
321 printk(KERN_INFO "Setting extra ubd major number to %d\n",
322 major);
323 err = 0;
324 out1:
325 mutex_unlock(&ubd_lock);
326 return err;
329 n = parse_unit(&str);
330 if(n < 0){
331 *error_out = "Couldn't parse device number";
332 return -EINVAL;
334 if(n >= MAX_DEV){
335 *error_out = "Device number out of range";
336 return 1;
339 err = -EBUSY;
340 mutex_lock(&ubd_lock);
342 ubd_dev = &ubd_devs[n];
343 if(ubd_dev->file != NULL){
344 *error_out = "Device is already configured";
345 goto out;
348 if (index_out)
349 *index_out = n;
351 err = -EINVAL;
352 for (i = 0; i < sizeof("rscd="); i++) {
353 switch (*str) {
354 case 'r':
355 flags.w = 0;
356 break;
357 case 's':
358 flags.s = 1;
359 break;
360 case 'd':
361 ubd_dev->no_cow = 1;
362 break;
363 case 'c':
364 ubd_dev->shared = 1;
365 break;
366 case '=':
367 str++;
368 goto break_loop;
369 default:
370 *error_out = "Expected '=' or flag letter "
371 "(r, s, c, or d)";
372 goto out;
374 str++;
377 if (*str == '=')
378 *error_out = "Too many flags specified";
379 else
380 *error_out = "Missing '='";
381 goto out;
383 break_loop:
384 backing_file = strchr(str, ',');
386 if (backing_file == NULL)
387 backing_file = strchr(str, ':');
389 if(backing_file != NULL){
390 if(ubd_dev->no_cow){
391 *error_out = "Can't specify both 'd' and a cow file";
392 goto out;
394 else {
395 *backing_file = '\0';
396 backing_file++;
399 err = 0;
400 ubd_dev->file = str;
401 ubd_dev->cow.file = backing_file;
402 ubd_dev->boot_openflags = flags;
403 out:
404 mutex_unlock(&ubd_lock);
405 return err;
408 static int ubd_setup(char *str)
410 char *error;
411 int err;
413 err = ubd_setup_common(str, NULL, &error);
414 if(err)
415 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
416 "%s\n", str, error);
417 return 1;
420 __setup("ubd", ubd_setup);
421 __uml_help(ubd_setup,
422 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
423 " This is used to associate a device with a file in the underlying\n"
424 " filesystem. When specifying two filenames, the first one is the\n"
425 " COW name and the second is the backing file name. As separator you can\n"
426 " use either a ':' or a ',': the first one allows writing things like;\n"
427 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
428 " while with a ',' the shell would not expand the 2nd '~'.\n"
429 " When using only one filename, UML will detect whether to treat it like\n"
430 " a COW file or a backing file. To override this detection, add the 'd'\n"
431 " flag:\n"
432 " ubd0d=BackingFile\n"
433 " Usually, there is a filesystem in the file, but \n"
434 " that's not required. Swap devices containing swap files can be\n"
435 " specified like this. Also, a file which doesn't contain a\n"
436 " filesystem can have its contents read in the virtual \n"
437 " machine by running 'dd' on the device. <n> must be in the range\n"
438 " 0 to 7. Appending an 'r' to the number will cause that device\n"
439 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
440 " an 's' will cause data to be written to disk on the host immediately.\n\n"
443 static int udb_setup(char *str)
445 printk("udb%s specified on command line is almost certainly a ubd -> "
446 "udb TYPO\n", str);
447 return 1;
450 __setup("udb", udb_setup);
451 __uml_help(udb_setup,
452 "udb\n"
453 " This option is here solely to catch ubd -> udb typos, which can be\n"
454 " to impossible to catch visually unless you specifically look for\n"
455 " them. The only result of any option starting with 'udb' is an error\n"
456 " in the boot output.\n\n"
459 static int fakehd_set = 0;
460 static int fakehd(char *str)
462 printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
463 fakehd_set = 1;
464 return 1;
467 __setup("fakehd", fakehd);
468 __uml_help(fakehd,
469 "fakehd\n"
470 " Change the ubd device name to \"hd\".\n\n"
473 static void do_ubd_request(struct request_queue * q);
475 /* Only changed by ubd_init, which is an initcall. */
476 int thread_fd = -1;
478 static void ubd_end_request(struct request *req, int bytes, int uptodate)
480 if (!end_that_request_first(req, uptodate, bytes >> 9)) {
481 struct ubd *dev = req->rq_disk->private_data;
482 unsigned long flags;
484 add_disk_randomness(req->rq_disk);
485 spin_lock_irqsave(&dev->lock, flags);
486 end_that_request_last(req, uptodate);
487 spin_unlock_irqrestore(&dev->lock, flags);
491 /* Callable only from interrupt context - otherwise you need to do
492 * spin_lock_irq()/spin_lock_irqsave() */
493 static inline void ubd_finish(struct request *req, int bytes)
495 if(bytes < 0){
496 ubd_end_request(req, 0, 0);
497 return;
499 ubd_end_request(req, bytes, 1);
502 static LIST_HEAD(restart);
504 /* XXX - move this inside ubd_intr. */
505 /* Called without dev->lock held, and only in interrupt context. */
506 static void ubd_handler(void)
508 struct io_thread_req *req;
509 struct request *rq;
510 struct ubd *ubd;
511 struct list_head *list, *next_ele;
512 unsigned long flags;
513 int n;
515 while(1){
516 n = os_read_file(thread_fd, &req,
517 sizeof(struct io_thread_req *));
518 if(n != sizeof(req)){
519 if(n == -EAGAIN)
520 break;
521 printk(KERN_ERR "spurious interrupt in ubd_handler, "
522 "err = %d\n", -n);
523 return;
526 rq = req->req;
527 rq->nr_sectors -= req->length >> 9;
528 if(rq->nr_sectors == 0)
529 ubd_finish(rq, rq->hard_nr_sectors << 9);
530 kfree(req);
532 reactivate_fd(thread_fd, UBD_IRQ);
534 list_for_each_safe(list, next_ele, &restart){
535 ubd = container_of(list, struct ubd, restart);
536 list_del_init(&ubd->restart);
537 spin_lock_irqsave(&ubd->lock, flags);
538 do_ubd_request(ubd->queue);
539 spin_unlock_irqrestore(&ubd->lock, flags);
543 static irqreturn_t ubd_intr(int irq, void *dev)
545 ubd_handler();
546 return IRQ_HANDLED;
549 /* Only changed by ubd_init, which is an initcall. */
550 static int io_pid = -1;
552 void kill_io_thread(void)
554 if(io_pid != -1)
555 os_kill_process(io_pid, 1);
558 __uml_exitcall(kill_io_thread);
560 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
562 char *file;
564 file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
565 return os_file_size(file, size_out);
568 static void ubd_close_dev(struct ubd *ubd_dev)
570 os_close_file(ubd_dev->fd);
571 if(ubd_dev->cow.file == NULL)
572 return;
574 os_close_file(ubd_dev->cow.fd);
575 vfree(ubd_dev->cow.bitmap);
576 ubd_dev->cow.bitmap = NULL;
579 static int ubd_open_dev(struct ubd *ubd_dev)
581 struct openflags flags;
582 char **back_ptr;
583 int err, create_cow, *create_ptr;
584 int fd;
586 ubd_dev->openflags = ubd_dev->boot_openflags;
587 create_cow = 0;
588 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
589 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
591 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
592 back_ptr, &ubd_dev->cow.bitmap_offset,
593 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
594 create_ptr);
596 if((fd == -ENOENT) && create_cow){
597 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
598 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
599 &ubd_dev->cow.bitmap_offset,
600 &ubd_dev->cow.bitmap_len,
601 &ubd_dev->cow.data_offset);
602 if(fd >= 0){
603 printk(KERN_INFO "Creating \"%s\" as COW file for "
604 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
608 if(fd < 0){
609 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
610 -fd);
611 return fd;
613 ubd_dev->fd = fd;
615 if(ubd_dev->cow.file != NULL){
616 blk_queue_max_sectors(ubd_dev->queue, 8 * sizeof(long));
618 err = -ENOMEM;
619 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
620 if(ubd_dev->cow.bitmap == NULL){
621 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
622 goto error;
624 flush_tlb_kernel_vm();
626 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
627 ubd_dev->cow.bitmap_offset,
628 ubd_dev->cow.bitmap_len);
629 if(err < 0)
630 goto error;
632 flags = ubd_dev->openflags;
633 flags.w = 0;
634 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
635 NULL, NULL, NULL, NULL);
636 if(err < 0) goto error;
637 ubd_dev->cow.fd = err;
639 return 0;
640 error:
641 os_close_file(ubd_dev->fd);
642 return err;
645 static void ubd_device_release(struct device *dev)
647 struct ubd *ubd_dev = dev->driver_data;
649 blk_cleanup_queue(ubd_dev->queue);
650 *ubd_dev = ((struct ubd) DEFAULT_UBD);
653 static int ubd_disk_register(int major, u64 size, int unit,
654 struct gendisk **disk_out)
656 struct gendisk *disk;
658 disk = alloc_disk(1 << UBD_SHIFT);
659 if(disk == NULL)
660 return -ENOMEM;
662 disk->major = major;
663 disk->first_minor = unit << UBD_SHIFT;
664 disk->fops = &ubd_blops;
665 set_capacity(disk, size / 512);
666 if(major == MAJOR_NR)
667 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
668 else
669 sprintf(disk->disk_name, "ubd_fake%d", unit);
671 /* sysfs register (not for ide fake devices) */
672 if (major == MAJOR_NR) {
673 ubd_devs[unit].pdev.id = unit;
674 ubd_devs[unit].pdev.name = DRIVER_NAME;
675 ubd_devs[unit].pdev.dev.release = ubd_device_release;
676 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
677 platform_device_register(&ubd_devs[unit].pdev);
678 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
681 disk->private_data = &ubd_devs[unit];
682 disk->queue = ubd_devs[unit].queue;
683 add_disk(disk);
685 *disk_out = disk;
686 return 0;
689 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
691 static int ubd_add(int n, char **error_out)
693 struct ubd *ubd_dev = &ubd_devs[n];
694 int err = 0;
696 if(ubd_dev->file == NULL)
697 goto out;
699 err = ubd_file_size(ubd_dev, &ubd_dev->size);
700 if(err < 0){
701 *error_out = "Couldn't determine size of device's file";
702 goto out;
705 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
707 INIT_LIST_HEAD(&ubd_dev->restart);
708 sg_init_table(ubd_dev->sg, MAX_SG);
710 err = -ENOMEM;
711 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
712 if (ubd_dev->queue == NULL) {
713 *error_out = "Failed to initialize device queue";
714 goto out;
716 ubd_dev->queue->queuedata = ubd_dev;
718 blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG);
719 err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
720 if(err){
721 *error_out = "Failed to register device";
722 goto out_cleanup;
725 if(fake_major != MAJOR_NR)
726 ubd_disk_register(fake_major, ubd_dev->size, n,
727 &fake_gendisk[n]);
729 /* perhaps this should also be under the "if (fake_major)" above */
730 /* using the fake_disk->disk_name and also the fakehd_set name */
731 if (fake_ide)
732 make_ide_entries(ubd_gendisk[n]->disk_name);
734 err = 0;
735 out:
736 return err;
738 out_cleanup:
739 blk_cleanup_queue(ubd_dev->queue);
740 goto out;
743 static int ubd_config(char *str, char **error_out)
745 int n, ret;
747 /* This string is possibly broken up and stored, so it's only
748 * freed if ubd_setup_common fails, or if only general options
749 * were set.
751 str = kstrdup(str, GFP_KERNEL);
752 if (str == NULL) {
753 *error_out = "Failed to allocate memory";
754 return -ENOMEM;
757 ret = ubd_setup_common(str, &n, error_out);
758 if (ret)
759 goto err_free;
761 if (n == -1) {
762 ret = 0;
763 goto err_free;
766 mutex_lock(&ubd_lock);
767 ret = ubd_add(n, error_out);
768 if (ret)
769 ubd_devs[n].file = NULL;
770 mutex_unlock(&ubd_lock);
772 out:
773 return ret;
775 err_free:
776 kfree(str);
777 goto out;
780 static int ubd_get_config(char *name, char *str, int size, char **error_out)
782 struct ubd *ubd_dev;
783 int n, len = 0;
785 n = parse_unit(&name);
786 if((n >= MAX_DEV) || (n < 0)){
787 *error_out = "ubd_get_config : device number out of range";
788 return -1;
791 ubd_dev = &ubd_devs[n];
792 mutex_lock(&ubd_lock);
794 if(ubd_dev->file == NULL){
795 CONFIG_CHUNK(str, size, len, "", 1);
796 goto out;
799 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
801 if(ubd_dev->cow.file != NULL){
802 CONFIG_CHUNK(str, size, len, ",", 0);
803 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
805 else CONFIG_CHUNK(str, size, len, "", 1);
807 out:
808 mutex_unlock(&ubd_lock);
809 return len;
812 static int ubd_id(char **str, int *start_out, int *end_out)
814 int n;
816 n = parse_unit(str);
817 *start_out = 0;
818 *end_out = MAX_DEV - 1;
819 return n;
822 static int ubd_remove(int n, char **error_out)
824 struct gendisk *disk = ubd_gendisk[n];
825 struct ubd *ubd_dev;
826 int err = -ENODEV;
828 mutex_lock(&ubd_lock);
830 ubd_dev = &ubd_devs[n];
832 if(ubd_dev->file == NULL)
833 goto out;
835 /* you cannot remove a open disk */
836 err = -EBUSY;
837 if(ubd_dev->count > 0)
838 goto out;
840 ubd_gendisk[n] = NULL;
841 if(disk != NULL){
842 del_gendisk(disk);
843 put_disk(disk);
846 if(fake_gendisk[n] != NULL){
847 del_gendisk(fake_gendisk[n]);
848 put_disk(fake_gendisk[n]);
849 fake_gendisk[n] = NULL;
852 err = 0;
853 platform_device_unregister(&ubd_dev->pdev);
854 out:
855 mutex_unlock(&ubd_lock);
856 return err;
859 /* All these are called by mconsole in process context and without
860 * ubd-specific locks. The structure itself is const except for .list.
862 static struct mc_device ubd_mc = {
863 .list = LIST_HEAD_INIT(ubd_mc.list),
864 .name = "ubd",
865 .config = ubd_config,
866 .get_config = ubd_get_config,
867 .id = ubd_id,
868 .remove = ubd_remove,
871 static int __init ubd_mc_init(void)
873 mconsole_register_dev(&ubd_mc);
874 return 0;
877 __initcall(ubd_mc_init);
879 static int __init ubd0_init(void)
881 struct ubd *ubd_dev = &ubd_devs[0];
883 mutex_lock(&ubd_lock);
884 if(ubd_dev->file == NULL)
885 ubd_dev->file = "root_fs";
886 mutex_unlock(&ubd_lock);
888 return 0;
891 __initcall(ubd0_init);
893 /* Used in ubd_init, which is an initcall */
894 static struct platform_driver ubd_driver = {
895 .driver = {
896 .name = DRIVER_NAME,
900 static int __init ubd_init(void)
902 char *error;
903 int i, err;
905 if (register_blkdev(MAJOR_NR, "ubd"))
906 return -1;
908 if (fake_major != MAJOR_NR) {
909 char name[sizeof("ubd_nnn\0")];
911 snprintf(name, sizeof(name), "ubd_%d", fake_major);
912 if (register_blkdev(fake_major, "ubd"))
913 return -1;
915 platform_driver_register(&ubd_driver);
916 mutex_lock(&ubd_lock);
917 for (i = 0; i < MAX_DEV; i++){
918 err = ubd_add(i, &error);
919 if(err)
920 printk(KERN_ERR "Failed to initialize ubd device %d :"
921 "%s\n", i, error);
923 mutex_unlock(&ubd_lock);
924 return 0;
927 late_initcall(ubd_init);
929 static int __init ubd_driver_init(void){
930 unsigned long stack;
931 int err;
933 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
934 if(global_openflags.s){
935 printk(KERN_INFO "ubd: Synchronous mode\n");
936 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
937 * enough. So use anyway the io thread. */
939 stack = alloc_stack(0, 0);
940 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
941 &thread_fd);
942 if(io_pid < 0){
943 printk(KERN_ERR
944 "ubd : Failed to start I/O thread (errno = %d) - "
945 "falling back to synchronous I/O\n", -io_pid);
946 io_pid = -1;
947 return 0;
949 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
950 IRQF_DISABLED, "ubd", ubd_devs);
951 if(err != 0)
952 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
953 return 0;
956 device_initcall(ubd_driver_init);
958 static int ubd_open(struct inode *inode, struct file *filp)
960 struct gendisk *disk = inode->i_bdev->bd_disk;
961 struct ubd *ubd_dev = disk->private_data;
962 int err = 0;
964 if(ubd_dev->count == 0){
965 err = ubd_open_dev(ubd_dev);
966 if(err){
967 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
968 disk->disk_name, ubd_dev->file, -err);
969 goto out;
972 ubd_dev->count++;
973 set_disk_ro(disk, !ubd_dev->openflags.w);
975 /* This should no more be needed. And it didn't work anyway to exclude
976 * read-write remounting of filesystems.*/
977 /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
978 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
979 err = -EROFS;
981 out:
982 return err;
985 static int ubd_release(struct inode * inode, struct file * file)
987 struct gendisk *disk = inode->i_bdev->bd_disk;
988 struct ubd *ubd_dev = disk->private_data;
990 if(--ubd_dev->count == 0)
991 ubd_close_dev(ubd_dev);
992 return 0;
995 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
996 __u64 *cow_offset, unsigned long *bitmap,
997 __u64 bitmap_offset, unsigned long *bitmap_words,
998 __u64 bitmap_len)
1000 __u64 sector = io_offset >> 9;
1001 int i, update_bitmap = 0;
1003 for(i = 0; i < length >> 9; i++){
1004 if(cow_mask != NULL)
1005 ubd_set_bit(i, (unsigned char *) cow_mask);
1006 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1007 continue;
1009 update_bitmap = 1;
1010 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1013 if(!update_bitmap)
1014 return;
1016 *cow_offset = sector / (sizeof(unsigned long) * 8);
1018 /* This takes care of the case where we're exactly at the end of the
1019 * device, and *cow_offset + 1 is off the end. So, just back it up
1020 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1021 * for the original diagnosis.
1023 if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
1024 sizeof(unsigned long) - 1))
1025 (*cow_offset)--;
1027 bitmap_words[0] = bitmap[*cow_offset];
1028 bitmap_words[1] = bitmap[*cow_offset + 1];
1030 *cow_offset *= sizeof(unsigned long);
1031 *cow_offset += bitmap_offset;
1034 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1035 __u64 bitmap_offset, __u64 bitmap_len)
1037 __u64 sector = req->offset >> 9;
1038 int i;
1040 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1041 panic("Operation too long");
1043 if(req->op == UBD_READ) {
1044 for(i = 0; i < req->length >> 9; i++){
1045 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1046 ubd_set_bit(i, (unsigned char *)
1047 &req->sector_mask);
1050 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1051 &req->cow_offset, bitmap, bitmap_offset,
1052 req->bitmap_words, bitmap_len);
1055 /* Called with dev->lock held */
1056 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1057 unsigned long long offset, int page_offset,
1058 int len, struct page *page)
1060 struct gendisk *disk = req->rq_disk;
1061 struct ubd *ubd_dev = disk->private_data;
1063 io_req->req = req;
1064 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1065 ubd_dev->fd;
1066 io_req->fds[1] = ubd_dev->fd;
1067 io_req->cow_offset = -1;
1068 io_req->offset = offset;
1069 io_req->length = len;
1070 io_req->error = 0;
1071 io_req->sector_mask = 0;
1073 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1074 io_req->offsets[0] = 0;
1075 io_req->offsets[1] = ubd_dev->cow.data_offset;
1076 io_req->buffer = page_address(page) + page_offset;
1077 io_req->sectorsize = 1 << 9;
1079 if(ubd_dev->cow.file != NULL)
1080 cowify_req(io_req, ubd_dev->cow.bitmap,
1081 ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1085 /* Called with dev->lock held */
1086 static void do_ubd_request(struct request_queue *q)
1088 struct io_thread_req *io_req;
1089 struct request *req;
1090 int n, last_sectors;
1092 while(1){
1093 struct ubd *dev = q->queuedata;
1094 if(dev->end_sg == 0){
1095 struct request *req = elv_next_request(q);
1096 if(req == NULL)
1097 return;
1099 dev->request = req;
1100 blkdev_dequeue_request(req);
1101 dev->start_sg = 0;
1102 dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1105 req = dev->request;
1106 last_sectors = 0;
1107 while(dev->start_sg < dev->end_sg){
1108 struct scatterlist *sg = &dev->sg[dev->start_sg];
1110 req->sector += last_sectors;
1111 io_req = kmalloc(sizeof(struct io_thread_req),
1112 GFP_ATOMIC);
1113 if(io_req == NULL){
1114 if(list_empty(&dev->restart))
1115 list_add(&dev->restart, &restart);
1116 return;
1118 prepare_request(req, io_req,
1119 (unsigned long long) req->sector << 9,
1120 sg->offset, sg->length, sg_page(sg));
1122 last_sectors = sg->length >> 9;
1123 n = os_write_file(thread_fd, &io_req,
1124 sizeof(struct io_thread_req *));
1125 if(n != sizeof(struct io_thread_req *)){
1126 if(n != -EAGAIN)
1127 printk("write to io thread failed, "
1128 "errno = %d\n", -n);
1129 else if(list_empty(&dev->restart))
1130 list_add(&dev->restart, &restart);
1131 return;
1134 dev->start_sg++;
1136 dev->end_sg = 0;
1137 dev->request = NULL;
1141 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1143 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1145 geo->heads = 128;
1146 geo->sectors = 32;
1147 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1148 return 0;
1151 static int ubd_ioctl(struct inode * inode, struct file * file,
1152 unsigned int cmd, unsigned long arg)
1154 struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1155 struct hd_driveid ubd_id = {
1156 .cyls = 0,
1157 .heads = 128,
1158 .sectors = 32,
1161 switch (cmd) {
1162 struct cdrom_volctrl volume;
1163 case HDIO_GET_IDENTITY:
1164 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1165 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1166 sizeof(ubd_id)))
1167 return -EFAULT;
1168 return 0;
1170 case CDROMVOLREAD:
1171 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1172 return -EFAULT;
1173 volume.channel0 = 255;
1174 volume.channel1 = 255;
1175 volume.channel2 = 255;
1176 volume.channel3 = 255;
1177 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1178 return -EFAULT;
1179 return 0;
1181 return -EINVAL;
1184 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1186 struct uml_stat buf1, buf2;
1187 int err;
1189 if(from_cmdline == NULL)
1190 return 0;
1191 if(!strcmp(from_cmdline, from_cow))
1192 return 0;
1194 err = os_stat_file(from_cmdline, &buf1);
1195 if(err < 0){
1196 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1197 return 0;
1199 err = os_stat_file(from_cow, &buf2);
1200 if(err < 0){
1201 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1202 return 1;
1204 if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1205 return 0;
1207 printk("Backing file mismatch - \"%s\" requested,\n"
1208 "\"%s\" specified in COW header of \"%s\"\n",
1209 from_cmdline, from_cow, cow);
1210 return 1;
1213 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1215 unsigned long modtime;
1216 unsigned long long actual;
1217 int err;
1219 err = os_file_modtime(file, &modtime);
1220 if(err < 0){
1221 printk("Failed to get modification time of backing file "
1222 "\"%s\", err = %d\n", file, -err);
1223 return err;
1226 err = os_file_size(file, &actual);
1227 if(err < 0){
1228 printk("Failed to get size of backing file \"%s\", "
1229 "err = %d\n", file, -err);
1230 return err;
1233 if(actual != size){
1234 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1235 * the typecast.*/
1236 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1237 "file\n", (unsigned long long) size, actual);
1238 return -EINVAL;
1240 if(modtime != mtime){
1241 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1242 "file\n", mtime, modtime);
1243 return -EINVAL;
1245 return 0;
1248 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1250 int err;
1252 err = os_seek_file(fd, offset);
1253 if(err < 0)
1254 return err;
1256 err = os_read_file(fd, buf, len);
1257 if(err < 0)
1258 return err;
1260 return 0;
1263 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1264 char **backing_file_out, int *bitmap_offset_out,
1265 unsigned long *bitmap_len_out, int *data_offset_out,
1266 int *create_cow_out)
1268 time_t mtime;
1269 unsigned long long size;
1270 __u32 version, align;
1271 char *backing_file;
1272 int fd, err, sectorsize, asked_switch, mode = 0644;
1274 fd = os_open_file(file, *openflags, mode);
1275 if (fd < 0) {
1276 if ((fd == -ENOENT) && (create_cow_out != NULL))
1277 *create_cow_out = 1;
1278 if (!openflags->w ||
1279 ((fd != -EROFS) && (fd != -EACCES)))
1280 return fd;
1281 openflags->w = 0;
1282 fd = os_open_file(file, *openflags, mode);
1283 if (fd < 0)
1284 return fd;
1287 if(shared)
1288 printk("Not locking \"%s\" on the host\n", file);
1289 else {
1290 err = os_lock_file(fd, openflags->w);
1291 if(err < 0){
1292 printk("Failed to lock '%s', err = %d\n", file, -err);
1293 goto out_close;
1297 /* Successful return case! */
1298 if(backing_file_out == NULL)
1299 return fd;
1301 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1302 &size, &sectorsize, &align, bitmap_offset_out);
1303 if(err && (*backing_file_out != NULL)){
1304 printk("Failed to read COW header from COW file \"%s\", "
1305 "errno = %d\n", file, -err);
1306 goto out_close;
1308 if(err)
1309 return fd;
1311 asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1313 /* Allow switching only if no mismatch. */
1314 if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1315 printk("Switching backing file to '%s'\n", *backing_file_out);
1316 err = write_cow_header(file, fd, *backing_file_out,
1317 sectorsize, align, &size);
1318 if (err) {
1319 printk("Switch failed, errno = %d\n", -err);
1320 goto out_close;
1322 } else {
1323 *backing_file_out = backing_file;
1324 err = backing_file_mismatch(*backing_file_out, size, mtime);
1325 if (err)
1326 goto out_close;
1329 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1330 bitmap_len_out, data_offset_out);
1332 return fd;
1333 out_close:
1334 os_close_file(fd);
1335 return err;
1338 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1339 int sectorsize, int alignment, int *bitmap_offset_out,
1340 unsigned long *bitmap_len_out, int *data_offset_out)
1342 int err, fd;
1344 flags.c = 1;
1345 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1346 if(fd < 0){
1347 err = fd;
1348 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1349 -err);
1350 goto out;
1353 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1354 bitmap_offset_out, bitmap_len_out,
1355 data_offset_out);
1356 if(!err)
1357 return fd;
1358 os_close_file(fd);
1359 out:
1360 return err;
1363 static int update_bitmap(struct io_thread_req *req)
1365 int n;
1367 if(req->cow_offset == -1)
1368 return 0;
1370 n = os_seek_file(req->fds[1], req->cow_offset);
1371 if(n < 0){
1372 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1373 return 1;
1376 n = os_write_file(req->fds[1], &req->bitmap_words,
1377 sizeof(req->bitmap_words));
1378 if(n != sizeof(req->bitmap_words)){
1379 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1380 req->fds[1]);
1381 return 1;
1384 return 0;
1387 void do_io(struct io_thread_req *req)
1389 char *buf;
1390 unsigned long len;
1391 int n, nsectors, start, end, bit;
1392 int err;
1393 __u64 off;
1395 nsectors = req->length / req->sectorsize;
1396 start = 0;
1397 do {
1398 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1399 end = start;
1400 while((end < nsectors) &&
1401 (ubd_test_bit(end, (unsigned char *)
1402 &req->sector_mask) == bit))
1403 end++;
1405 off = req->offset + req->offsets[bit] +
1406 start * req->sectorsize;
1407 len = (end - start) * req->sectorsize;
1408 buf = &req->buffer[start * req->sectorsize];
1410 err = os_seek_file(req->fds[bit], off);
1411 if(err < 0){
1412 printk("do_io - lseek failed : err = %d\n", -err);
1413 req->error = 1;
1414 return;
1416 if(req->op == UBD_READ){
1417 n = 0;
1418 do {
1419 buf = &buf[n];
1420 len -= n;
1421 n = os_read_file(req->fds[bit], buf, len);
1422 if (n < 0) {
1423 printk("do_io - read failed, err = %d "
1424 "fd = %d\n", -n, req->fds[bit]);
1425 req->error = 1;
1426 return;
1428 } while((n < len) && (n != 0));
1429 if (n < len) memset(&buf[n], 0, len - n);
1430 } else {
1431 n = os_write_file(req->fds[bit], buf, len);
1432 if(n != len){
1433 printk("do_io - write failed err = %d "
1434 "fd = %d\n", -n, req->fds[bit]);
1435 req->error = 1;
1436 return;
1440 start = end;
1441 } while(start < nsectors);
1443 req->error = update_bitmap(req);
1446 /* Changed in start_io_thread, which is serialized by being called only
1447 * from ubd_init, which is an initcall.
1449 int kernel_fd = -1;
1451 /* Only changed by the io thread. XXX: currently unused. */
1452 static int io_count = 0;
1454 int io_thread(void *arg)
1456 struct io_thread_req *req;
1457 int n;
1459 ignore_sigwinch_sig();
1460 while(1){
1461 n = os_read_file(kernel_fd, &req,
1462 sizeof(struct io_thread_req *));
1463 if(n != sizeof(struct io_thread_req *)){
1464 if(n < 0)
1465 printk("io_thread - read failed, fd = %d, "
1466 "err = %d\n", kernel_fd, -n);
1467 else {
1468 printk("io_thread - short read, fd = %d, "
1469 "length = %d\n", kernel_fd, n);
1471 continue;
1473 io_count++;
1474 do_io(req);
1475 n = os_write_file(kernel_fd, &req,
1476 sizeof(struct io_thread_req *));
1477 if(n != sizeof(struct io_thread_req *))
1478 printk("io_thread - write failed, fd = %d, err = %d\n",
1479 kernel_fd, -n);
1482 return 0;