]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - arch/um/drivers/ubd_kern.c
Pull icc-cleanup into release branch
[mirror_ubuntu-zesty-kernel.git] / arch / um / drivers / ubd_kern.c
1 /*
2 * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3 * Licensed under the GPL
4 */
5
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
18 */
19
20 #define MAJOR_NR UBD_MAJOR
21 #define UBD_SHIFT 4
22
23 #include "linux/config.h"
24 #include "linux/module.h"
25 #include "linux/blkdev.h"
26 #include "linux/hdreg.h"
27 #include "linux/init.h"
28 #include "linux/devfs_fs_kernel.h"
29 #include "linux/cdrom.h"
30 #include "linux/proc_fs.h"
31 #include "linux/ctype.h"
32 #include "linux/capability.h"
33 #include "linux/mm.h"
34 #include "linux/vmalloc.h"
35 #include "linux/blkpg.h"
36 #include "linux/genhd.h"
37 #include "linux/spinlock.h"
38 #include "linux/platform_device.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 "user_util.h"
45 #include "mem_user.h"
46 #include "kern_util.h"
47 #include "kern.h"
48 #include "mconsole_kern.h"
49 #include "init.h"
50 #include "irq_user.h"
51 #include "irq_kern.h"
52 #include "ubd_user.h"
53 #include "os.h"
54 #include "mem.h"
55 #include "mem_kern.h"
56 #include "cow.h"
57
58 enum ubd_req { UBD_READ, UBD_WRITE };
59
60 struct io_thread_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;
72 };
73
74 extern int open_ubd_file(char *file, struct openflags *openflags,
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);
85
86 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
87 {
88 __u64 n;
89 int bits, off;
90
91 bits = sizeof(data[0]) * 8;
92 n = bit / bits;
93 off = bit % bits;
94 return((data[n] & (1 << off)) != 0);
95 }
96
97 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
98 {
99 __u64 n;
100 int bits, off;
101
102 bits = sizeof(data[0]) * 8;
103 n = bit / bits;
104 off = bit % bits;
105 data[n] |= (1 << off);
106 }
107 /*End stuff from ubd_user.h*/
108
109 #define DRIVER_NAME "uml-blkdev"
110
111 static DEFINE_SPINLOCK(ubd_io_lock);
112 static DEFINE_SPINLOCK(ubd_lock);
113
114 static void (*do_ubd)(void);
115
116 static int ubd_open(struct inode * inode, struct file * filp);
117 static int ubd_release(struct inode * inode, struct file * file);
118 static int ubd_ioctl(struct inode * inode, struct file * file,
119 unsigned int cmd, unsigned long arg);
120 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
121
122 #define MAX_DEV (8)
123
124 static struct block_device_operations ubd_blops = {
125 .owner = THIS_MODULE,
126 .open = ubd_open,
127 .release = ubd_release,
128 .ioctl = ubd_ioctl,
129 .getgeo = ubd_getgeo,
130 };
131
132 /* Protected by the queue_lock */
133 static request_queue_t *ubd_queue;
134
135 /* Protected by ubd_lock */
136 static int fake_major = MAJOR_NR;
137
138 static struct gendisk *ubd_gendisk[MAX_DEV];
139 static struct gendisk *fake_gendisk[MAX_DEV];
140
141 #ifdef CONFIG_BLK_DEV_UBD_SYNC
142 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
143 .cl = 1 })
144 #else
145 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
146 .cl = 1 })
147 #endif
148
149 /* Not protected - changed only in ubd_setup_common and then only to
150 * to enable O_SYNC.
151 */
152 static struct openflags global_openflags = OPEN_FLAGS;
153
154 struct cow {
155 /* This is the backing file, actually */
156 char *file;
157 int fd;
158 unsigned long *bitmap;
159 unsigned long bitmap_len;
160 int bitmap_offset;
161 int data_offset;
162 };
163
164 struct ubd {
165 char *file;
166 int count;
167 int fd;
168 __u64 size;
169 struct openflags boot_openflags;
170 struct openflags openflags;
171 int no_cow;
172 struct cow cow;
173 struct platform_device pdev;
174 };
175
176 #define DEFAULT_COW { \
177 .file = NULL, \
178 .fd = -1, \
179 .bitmap = NULL, \
180 .bitmap_offset = 0, \
181 .data_offset = 0, \
182 }
183
184 #define DEFAULT_UBD { \
185 .file = NULL, \
186 .count = 0, \
187 .fd = -1, \
188 .size = -1, \
189 .boot_openflags = OPEN_FLAGS, \
190 .openflags = OPEN_FLAGS, \
191 .no_cow = 0, \
192 .cow = DEFAULT_COW, \
193 }
194
195 struct ubd ubd_dev[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
196
197 static int ubd0_init(void)
198 {
199 struct ubd *dev = &ubd_dev[0];
200
201 if(dev->file == NULL)
202 dev->file = "root_fs";
203 return(0);
204 }
205
206 __initcall(ubd0_init);
207
208 /* Only changed by fake_ide_setup which is a setup */
209 static int fake_ide = 0;
210 static struct proc_dir_entry *proc_ide_root = NULL;
211 static struct proc_dir_entry *proc_ide = NULL;
212
213 static void make_proc_ide(void)
214 {
215 proc_ide_root = proc_mkdir("ide", NULL);
216 proc_ide = proc_mkdir("ide0", proc_ide_root);
217 }
218
219 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
220 int *eof, void *data)
221 {
222 int len;
223
224 strcpy(page, "disk\n");
225 len = strlen("disk\n");
226 len -= off;
227 if (len < count){
228 *eof = 1;
229 if (len <= 0) return 0;
230 }
231 else len = count;
232 *start = page + off;
233 return len;
234 }
235
236 static void make_ide_entries(char *dev_name)
237 {
238 struct proc_dir_entry *dir, *ent;
239 char name[64];
240
241 if(proc_ide_root == NULL) make_proc_ide();
242
243 dir = proc_mkdir(dev_name, proc_ide);
244 if(!dir) return;
245
246 ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
247 if(!ent) return;
248 ent->nlink = 1;
249 ent->data = NULL;
250 ent->read_proc = proc_ide_read_media;
251 ent->write_proc = NULL;
252 sprintf(name,"ide0/%s", dev_name);
253 proc_symlink(dev_name, proc_ide_root, name);
254 }
255
256 static int fake_ide_setup(char *str)
257 {
258 fake_ide = 1;
259 return(1);
260 }
261
262 __setup("fake_ide", fake_ide_setup);
263
264 __uml_help(fake_ide_setup,
265 "fake_ide\n"
266 " Create ide0 entries that map onto ubd devices.\n\n"
267 );
268
269 static int parse_unit(char **ptr)
270 {
271 char *str = *ptr, *end;
272 int n = -1;
273
274 if(isdigit(*str)) {
275 n = simple_strtoul(str, &end, 0);
276 if(end == str)
277 return(-1);
278 *ptr = end;
279 }
280 else if (('a' <= *str) && (*str <= 'h')) {
281 n = *str - 'a';
282 str++;
283 *ptr = str;
284 }
285 return(n);
286 }
287
288 static int ubd_setup_common(char *str, int *index_out)
289 {
290 struct ubd *dev;
291 struct openflags flags = global_openflags;
292 char *backing_file;
293 int n, err, i;
294
295 if(index_out) *index_out = -1;
296 n = *str;
297 if(n == '='){
298 char *end;
299 int major;
300
301 str++;
302 if(!strcmp(str, "sync")){
303 global_openflags = of_sync(global_openflags);
304 return(0);
305 }
306 major = simple_strtoul(str, &end, 0);
307 if((*end != '\0') || (end == str)){
308 printk(KERN_ERR
309 "ubd_setup : didn't parse major number\n");
310 return(1);
311 }
312
313 err = 1;
314 spin_lock(&ubd_lock);
315 if(fake_major != MAJOR_NR){
316 printk(KERN_ERR "Can't assign a fake major twice\n");
317 goto out1;
318 }
319
320 fake_major = major;
321
322 printk(KERN_INFO "Setting extra ubd major number to %d\n",
323 major);
324 err = 0;
325 out1:
326 spin_unlock(&ubd_lock);
327 return(err);
328 }
329
330 n = parse_unit(&str);
331 if(n < 0){
332 printk(KERN_ERR "ubd_setup : couldn't parse unit number "
333 "'%s'\n", str);
334 return(1);
335 }
336 if(n >= MAX_DEV){
337 printk(KERN_ERR "ubd_setup : index %d out of range "
338 "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
339 return(1);
340 }
341
342 err = 1;
343 spin_lock(&ubd_lock);
344
345 dev = &ubd_dev[n];
346 if(dev->file != NULL){
347 printk(KERN_ERR "ubd_setup : device already configured\n");
348 goto out;
349 }
350
351 if (index_out)
352 *index_out = n;
353
354 for (i = 0; i < 4; i++) {
355 switch (*str) {
356 case 'r':
357 flags.w = 0;
358 break;
359 case 's':
360 flags.s = 1;
361 break;
362 case 'd':
363 dev->no_cow = 1;
364 break;
365 case '=':
366 str++;
367 goto break_loop;
368 default:
369 printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r,s or d)\n");
370 goto out;
371 }
372 str++;
373 }
374
375 if (*str == '=')
376 printk(KERN_ERR "ubd_setup : Too many flags specified\n");
377 else
378 printk(KERN_ERR "ubd_setup : Expected '='\n");
379 goto out;
380
381 break_loop:
382 err = 0;
383 backing_file = strchr(str, ',');
384
385 if (!backing_file) {
386 backing_file = strchr(str, ':');
387 }
388
389 if(backing_file){
390 if(dev->no_cow)
391 printk(KERN_ERR "Can't specify both 'd' and a "
392 "cow file\n");
393 else {
394 *backing_file = '\0';
395 backing_file++;
396 }
397 }
398 dev->file = str;
399 dev->cow.file = backing_file;
400 dev->boot_openflags = flags;
401 out:
402 spin_unlock(&ubd_lock);
403 return(err);
404 }
405
406 static int ubd_setup(char *str)
407 {
408 ubd_setup_common(str, NULL);
409 return(1);
410 }
411
412 __setup("ubd", ubd_setup);
413 __uml_help(ubd_setup,
414 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
415 " This is used to associate a device with a file in the underlying\n"
416 " filesystem. When specifying two filenames, the first one is the\n"
417 " COW name and the second is the backing file name. As separator you can\n"
418 " use either a ':' or a ',': the first one allows writing things like;\n"
419 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
420 " while with a ',' the shell would not expand the 2nd '~'.\n"
421 " When using only one filename, UML will detect whether to thread it like\n"
422 " a COW file or a backing file. To override this detection, add the 'd'\n"
423 " flag:\n"
424 " ubd0d=BackingFile\n"
425 " Usually, there is a filesystem in the file, but \n"
426 " that's not required. Swap devices containing swap files can be\n"
427 " specified like this. Also, a file which doesn't contain a\n"
428 " filesystem can have its contents read in the virtual \n"
429 " machine by running 'dd' on the device. <n> must be in the range\n"
430 " 0 to 7. Appending an 'r' to the number will cause that device\n"
431 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
432 " an 's' will cause data to be written to disk on the host immediately.\n\n"
433 );
434
435 static int udb_setup(char *str)
436 {
437 printk("udb%s specified on command line is almost certainly a ubd -> "
438 "udb TYPO\n", str);
439 return(1);
440 }
441
442 __setup("udb", udb_setup);
443 __uml_help(udb_setup,
444 "udb\n"
445 " This option is here solely to catch ubd -> udb typos, which can be\n"
446 " to impossible to catch visually unless you specifically look for\n"
447 " them. The only result of any option starting with 'udb' is an error\n"
448 " in the boot output.\n\n"
449 );
450
451 static int fakehd_set = 0;
452 static int fakehd(char *str)
453 {
454 printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
455 fakehd_set = 1;
456 return 1;
457 }
458
459 __setup("fakehd", fakehd);
460 __uml_help(fakehd,
461 "fakehd\n"
462 " Change the ubd device name to \"hd\".\n\n"
463 );
464
465 static void do_ubd_request(request_queue_t * q);
466
467 /* Only changed by ubd_init, which is an initcall. */
468 int thread_fd = -1;
469
470 /* Changed by ubd_handler, which is serialized because interrupts only
471 * happen on CPU 0.
472 */
473 int intr_count = 0;
474
475 /* call ubd_finish if you need to serialize */
476 static void __ubd_finish(struct request *req, int error)
477 {
478 int nsect;
479
480 if(error){
481 end_request(req, 0);
482 return;
483 }
484 nsect = req->current_nr_sectors;
485 req->sector += nsect;
486 req->buffer += nsect << 9;
487 req->errors = 0;
488 req->nr_sectors -= nsect;
489 req->current_nr_sectors = 0;
490 end_request(req, 1);
491 }
492
493 static inline void ubd_finish(struct request *req, int error)
494 {
495 spin_lock(&ubd_io_lock);
496 __ubd_finish(req, error);
497 spin_unlock(&ubd_io_lock);
498 }
499
500 /* Called without ubd_io_lock held */
501 static void ubd_handler(void)
502 {
503 struct io_thread_req req;
504 struct request *rq = elv_next_request(ubd_queue);
505 int n;
506
507 do_ubd = NULL;
508 intr_count++;
509 n = os_read_file(thread_fd, &req, sizeof(req));
510 if(n != sizeof(req)){
511 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
512 "err = %d\n", os_getpid(), -n);
513 spin_lock(&ubd_io_lock);
514 end_request(rq, 0);
515 spin_unlock(&ubd_io_lock);
516 return;
517 }
518
519 ubd_finish(rq, req.error);
520 reactivate_fd(thread_fd, UBD_IRQ);
521 do_ubd_request(ubd_queue);
522 }
523
524 static irqreturn_t ubd_intr(int irq, void *dev, struct pt_regs *unused)
525 {
526 ubd_handler();
527 return(IRQ_HANDLED);
528 }
529
530 /* Only changed by ubd_init, which is an initcall. */
531 static int io_pid = -1;
532
533 void kill_io_thread(void)
534 {
535 if(io_pid != -1)
536 os_kill_process(io_pid, 1);
537 }
538
539 __uml_exitcall(kill_io_thread);
540
541 static int ubd_file_size(struct ubd *dev, __u64 *size_out)
542 {
543 char *file;
544
545 file = dev->cow.file ? dev->cow.file : dev->file;
546 return(os_file_size(file, size_out));
547 }
548
549 static void ubd_close(struct ubd *dev)
550 {
551 os_close_file(dev->fd);
552 if(dev->cow.file == NULL)
553 return;
554
555 os_close_file(dev->cow.fd);
556 vfree(dev->cow.bitmap);
557 dev->cow.bitmap = NULL;
558 }
559
560 static int ubd_open_dev(struct ubd *dev)
561 {
562 struct openflags flags;
563 char **back_ptr;
564 int err, create_cow, *create_ptr;
565
566 dev->openflags = dev->boot_openflags;
567 create_cow = 0;
568 create_ptr = (dev->cow.file != NULL) ? &create_cow : NULL;
569 back_ptr = dev->no_cow ? NULL : &dev->cow.file;
570 dev->fd = open_ubd_file(dev->file, &dev->openflags, back_ptr,
571 &dev->cow.bitmap_offset, &dev->cow.bitmap_len,
572 &dev->cow.data_offset, create_ptr);
573
574 if((dev->fd == -ENOENT) && create_cow){
575 dev->fd = create_cow_file(dev->file, dev->cow.file,
576 dev->openflags, 1 << 9, PAGE_SIZE,
577 &dev->cow.bitmap_offset,
578 &dev->cow.bitmap_len,
579 &dev->cow.data_offset);
580 if(dev->fd >= 0){
581 printk(KERN_INFO "Creating \"%s\" as COW file for "
582 "\"%s\"\n", dev->file, dev->cow.file);
583 }
584 }
585
586 if(dev->fd < 0){
587 printk("Failed to open '%s', errno = %d\n", dev->file,
588 -dev->fd);
589 return(dev->fd);
590 }
591
592 if(dev->cow.file != NULL){
593 err = -ENOMEM;
594 dev->cow.bitmap = (void *) vmalloc(dev->cow.bitmap_len);
595 if(dev->cow.bitmap == NULL){
596 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
597 goto error;
598 }
599 flush_tlb_kernel_vm();
600
601 err = read_cow_bitmap(dev->fd, dev->cow.bitmap,
602 dev->cow.bitmap_offset,
603 dev->cow.bitmap_len);
604 if(err < 0)
605 goto error;
606
607 flags = dev->openflags;
608 flags.w = 0;
609 err = open_ubd_file(dev->cow.file, &flags, NULL, NULL, NULL,
610 NULL, NULL);
611 if(err < 0) goto error;
612 dev->cow.fd = err;
613 }
614 return(0);
615 error:
616 os_close_file(dev->fd);
617 return(err);
618 }
619
620 static int ubd_new_disk(int major, u64 size, int unit,
621 struct gendisk **disk_out)
622
623 {
624 struct gendisk *disk;
625 char from[sizeof("ubd/nnnnn\0")], to[sizeof("discnnnnn/disc\0")];
626 int err;
627
628 disk = alloc_disk(1 << UBD_SHIFT);
629 if(disk == NULL)
630 return(-ENOMEM);
631
632 disk->major = major;
633 disk->first_minor = unit << UBD_SHIFT;
634 disk->fops = &ubd_blops;
635 set_capacity(disk, size / 512);
636 if(major == MAJOR_NR){
637 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
638 sprintf(disk->devfs_name, "ubd/disc%d", unit);
639 sprintf(from, "ubd/%d", unit);
640 sprintf(to, "disc%d/disc", unit);
641 err = devfs_mk_symlink(from, to);
642 if(err)
643 printk("ubd_new_disk failed to make link from %s to "
644 "%s, error = %d\n", from, to, err);
645 }
646 else {
647 sprintf(disk->disk_name, "ubd_fake%d", unit);
648 sprintf(disk->devfs_name, "ubd_fake/disc%d", unit);
649 }
650
651 /* sysfs register (not for ide fake devices) */
652 if (major == MAJOR_NR) {
653 ubd_dev[unit].pdev.id = unit;
654 ubd_dev[unit].pdev.name = DRIVER_NAME;
655 platform_device_register(&ubd_dev[unit].pdev);
656 disk->driverfs_dev = &ubd_dev[unit].pdev.dev;
657 }
658
659 disk->private_data = &ubd_dev[unit];
660 disk->queue = ubd_queue;
661 add_disk(disk);
662
663 *disk_out = disk;
664 return 0;
665 }
666
667 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
668
669 static int ubd_add(int n)
670 {
671 struct ubd *dev = &ubd_dev[n];
672 int err;
673
674 err = -ENODEV;
675 if(dev->file == NULL)
676 goto out;
677
678 if (ubd_open_dev(dev))
679 goto out;
680
681 err = ubd_file_size(dev, &dev->size);
682 if(err < 0)
683 goto out_close;
684
685 dev->size = ROUND_BLOCK(dev->size);
686
687 err = ubd_new_disk(MAJOR_NR, dev->size, n, &ubd_gendisk[n]);
688 if(err)
689 goto out_close;
690
691 if(fake_major != MAJOR_NR)
692 ubd_new_disk(fake_major, dev->size, n,
693 &fake_gendisk[n]);
694
695 /* perhaps this should also be under the "if (fake_major)" above */
696 /* using the fake_disk->disk_name and also the fakehd_set name */
697 if (fake_ide)
698 make_ide_entries(ubd_gendisk[n]->disk_name);
699
700 err = 0;
701 out_close:
702 ubd_close(dev);
703 out:
704 return err;
705 }
706
707 static int ubd_config(char *str)
708 {
709 int n, err;
710
711 str = kstrdup(str, GFP_KERNEL);
712 if(str == NULL){
713 printk(KERN_ERR "ubd_config failed to strdup string\n");
714 return(1);
715 }
716 err = ubd_setup_common(str, &n);
717 if(err){
718 kfree(str);
719 return(-1);
720 }
721 if(n == -1) return(0);
722
723 spin_lock(&ubd_lock);
724 err = ubd_add(n);
725 if(err)
726 ubd_dev[n].file = NULL;
727 spin_unlock(&ubd_lock);
728
729 return(err);
730 }
731
732 static int ubd_get_config(char *name, char *str, int size, char **error_out)
733 {
734 struct ubd *dev;
735 int n, len = 0;
736
737 n = parse_unit(&name);
738 if((n >= MAX_DEV) || (n < 0)){
739 *error_out = "ubd_get_config : device number out of range";
740 return(-1);
741 }
742
743 dev = &ubd_dev[n];
744 spin_lock(&ubd_lock);
745
746 if(dev->file == NULL){
747 CONFIG_CHUNK(str, size, len, "", 1);
748 goto out;
749 }
750
751 CONFIG_CHUNK(str, size, len, dev->file, 0);
752
753 if(dev->cow.file != NULL){
754 CONFIG_CHUNK(str, size, len, ",", 0);
755 CONFIG_CHUNK(str, size, len, dev->cow.file, 1);
756 }
757 else CONFIG_CHUNK(str, size, len, "", 1);
758
759 out:
760 spin_unlock(&ubd_lock);
761 return(len);
762 }
763
764 static int ubd_id(char **str, int *start_out, int *end_out)
765 {
766 int n;
767
768 n = parse_unit(str);
769 *start_out = 0;
770 *end_out = MAX_DEV - 1;
771 return n;
772 }
773
774 static int ubd_remove(int n)
775 {
776 struct ubd *dev;
777 int err = -ENODEV;
778
779 spin_lock(&ubd_lock);
780
781 if(ubd_gendisk[n] == NULL)
782 goto out;
783
784 dev = &ubd_dev[n];
785
786 if(dev->file == NULL)
787 goto out;
788
789 /* you cannot remove a open disk */
790 err = -EBUSY;
791 if(dev->count > 0)
792 goto out;
793
794 del_gendisk(ubd_gendisk[n]);
795 put_disk(ubd_gendisk[n]);
796 ubd_gendisk[n] = NULL;
797
798 if(fake_gendisk[n] != NULL){
799 del_gendisk(fake_gendisk[n]);
800 put_disk(fake_gendisk[n]);
801 fake_gendisk[n] = NULL;
802 }
803
804 platform_device_unregister(&dev->pdev);
805 *dev = ((struct ubd) DEFAULT_UBD);
806 err = 0;
807 out:
808 spin_unlock(&ubd_lock);
809 return err;
810 }
811
812 static struct mc_device ubd_mc = {
813 .name = "ubd",
814 .config = ubd_config,
815 .get_config = ubd_get_config,
816 .id = ubd_id,
817 .remove = ubd_remove,
818 };
819
820 static int ubd_mc_init(void)
821 {
822 mconsole_register_dev(&ubd_mc);
823 return 0;
824 }
825
826 __initcall(ubd_mc_init);
827
828 static struct platform_driver ubd_driver = {
829 .driver = {
830 .name = DRIVER_NAME,
831 },
832 };
833
834 int ubd_init(void)
835 {
836 int i;
837
838 devfs_mk_dir("ubd");
839 if (register_blkdev(MAJOR_NR, "ubd"))
840 return -1;
841
842 ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
843 if (!ubd_queue) {
844 unregister_blkdev(MAJOR_NR, "ubd");
845 return -1;
846 }
847
848 if (fake_major != MAJOR_NR) {
849 char name[sizeof("ubd_nnn\0")];
850
851 snprintf(name, sizeof(name), "ubd_%d", fake_major);
852 devfs_mk_dir(name);
853 if (register_blkdev(fake_major, "ubd"))
854 return -1;
855 }
856 platform_driver_register(&ubd_driver);
857 for (i = 0; i < MAX_DEV; i++)
858 ubd_add(i);
859 return 0;
860 }
861
862 late_initcall(ubd_init);
863
864 int ubd_driver_init(void){
865 unsigned long stack;
866 int err;
867
868 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
869 if(global_openflags.s){
870 printk(KERN_INFO "ubd: Synchronous mode\n");
871 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
872 * enough. So use anyway the io thread. */
873 }
874 stack = alloc_stack(0, 0);
875 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
876 &thread_fd);
877 if(io_pid < 0){
878 printk(KERN_ERR
879 "ubd : Failed to start I/O thread (errno = %d) - "
880 "falling back to synchronous I/O\n", -io_pid);
881 io_pid = -1;
882 return(0);
883 }
884 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
885 SA_INTERRUPT, "ubd", ubd_dev);
886 if(err != 0)
887 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
888 return(err);
889 }
890
891 device_initcall(ubd_driver_init);
892
893 static int ubd_open(struct inode *inode, struct file *filp)
894 {
895 struct gendisk *disk = inode->i_bdev->bd_disk;
896 struct ubd *dev = disk->private_data;
897 int err = 0;
898
899 if(dev->count == 0){
900 err = ubd_open_dev(dev);
901 if(err){
902 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
903 disk->disk_name, dev->file, -err);
904 goto out;
905 }
906 }
907 dev->count++;
908 set_disk_ro(disk, !dev->openflags.w);
909
910 /* This should no more be needed. And it didn't work anyway to exclude
911 * read-write remounting of filesystems.*/
912 /*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
913 if(--dev->count == 0) ubd_close(dev);
914 err = -EROFS;
915 }*/
916 out:
917 return(err);
918 }
919
920 static int ubd_release(struct inode * inode, struct file * file)
921 {
922 struct gendisk *disk = inode->i_bdev->bd_disk;
923 struct ubd *dev = disk->private_data;
924
925 if(--dev->count == 0)
926 ubd_close(dev);
927 return(0);
928 }
929
930 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
931 __u64 *cow_offset, unsigned long *bitmap,
932 __u64 bitmap_offset, unsigned long *bitmap_words,
933 __u64 bitmap_len)
934 {
935 __u64 sector = io_offset >> 9;
936 int i, update_bitmap = 0;
937
938 for(i = 0; i < length >> 9; i++){
939 if(cow_mask != NULL)
940 ubd_set_bit(i, (unsigned char *) cow_mask);
941 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
942 continue;
943
944 update_bitmap = 1;
945 ubd_set_bit(sector + i, (unsigned char *) bitmap);
946 }
947
948 if(!update_bitmap)
949 return;
950
951 *cow_offset = sector / (sizeof(unsigned long) * 8);
952
953 /* This takes care of the case where we're exactly at the end of the
954 * device, and *cow_offset + 1 is off the end. So, just back it up
955 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
956 * for the original diagnosis.
957 */
958 if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
959 sizeof(unsigned long) - 1))
960 (*cow_offset)--;
961
962 bitmap_words[0] = bitmap[*cow_offset];
963 bitmap_words[1] = bitmap[*cow_offset + 1];
964
965 *cow_offset *= sizeof(unsigned long);
966 *cow_offset += bitmap_offset;
967 }
968
969 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
970 __u64 bitmap_offset, __u64 bitmap_len)
971 {
972 __u64 sector = req->offset >> 9;
973 int i;
974
975 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
976 panic("Operation too long");
977
978 if(req->op == UBD_READ) {
979 for(i = 0; i < req->length >> 9; i++){
980 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
981 ubd_set_bit(i, (unsigned char *)
982 &req->sector_mask);
983 }
984 }
985 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
986 &req->cow_offset, bitmap, bitmap_offset,
987 req->bitmap_words, bitmap_len);
988 }
989
990 /* Called with ubd_io_lock held */
991 static int prepare_request(struct request *req, struct io_thread_req *io_req)
992 {
993 struct gendisk *disk = req->rq_disk;
994 struct ubd *dev = disk->private_data;
995 __u64 offset;
996 int len;
997
998 if(req->rq_status == RQ_INACTIVE) return(1);
999
1000 /* This should be impossible now */
1001 if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
1002 printk("Write attempted on readonly ubd device %s\n",
1003 disk->disk_name);
1004 end_request(req, 0);
1005 return(1);
1006 }
1007
1008 offset = ((__u64) req->sector) << 9;
1009 len = req->current_nr_sectors << 9;
1010
1011 io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
1012 io_req->fds[1] = dev->fd;
1013 io_req->cow_offset = -1;
1014 io_req->offset = offset;
1015 io_req->length = len;
1016 io_req->error = 0;
1017 io_req->sector_mask = 0;
1018
1019 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1020 io_req->offsets[0] = 0;
1021 io_req->offsets[1] = dev->cow.data_offset;
1022 io_req->buffer = req->buffer;
1023 io_req->sectorsize = 1 << 9;
1024
1025 if(dev->cow.file != NULL)
1026 cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset,
1027 dev->cow.bitmap_len);
1028
1029 return(0);
1030 }
1031
1032 /* Called with ubd_io_lock held */
1033 static void do_ubd_request(request_queue_t *q)
1034 {
1035 struct io_thread_req io_req;
1036 struct request *req;
1037 int err, n;
1038
1039 if(thread_fd == -1){
1040 while((req = elv_next_request(q)) != NULL){
1041 err = prepare_request(req, &io_req);
1042 if(!err){
1043 do_io(&io_req);
1044 __ubd_finish(req, io_req.error);
1045 }
1046 }
1047 }
1048 else {
1049 if(do_ubd || (req = elv_next_request(q)) == NULL)
1050 return;
1051 err = prepare_request(req, &io_req);
1052 if(!err){
1053 do_ubd = ubd_handler;
1054 n = os_write_file(thread_fd, (char *) &io_req,
1055 sizeof(io_req));
1056 if(n != sizeof(io_req))
1057 printk("write to io thread failed, "
1058 "errno = %d\n", -n);
1059 }
1060 }
1061 }
1062
1063 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1064 {
1065 struct ubd *dev = bdev->bd_disk->private_data;
1066
1067 geo->heads = 128;
1068 geo->sectors = 32;
1069 geo->cylinders = dev->size / (128 * 32 * 512);
1070 return 0;
1071 }
1072
1073 static int ubd_ioctl(struct inode * inode, struct file * file,
1074 unsigned int cmd, unsigned long arg)
1075 {
1076 struct ubd *dev = inode->i_bdev->bd_disk->private_data;
1077 struct hd_driveid ubd_id = {
1078 .cyls = 0,
1079 .heads = 128,
1080 .sectors = 32,
1081 };
1082
1083 switch (cmd) {
1084 struct cdrom_volctrl volume;
1085 case HDIO_GET_IDENTITY:
1086 ubd_id.cyls = dev->size / (128 * 32 * 512);
1087 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1088 sizeof(ubd_id)))
1089 return(-EFAULT);
1090 return(0);
1091
1092 case CDROMVOLREAD:
1093 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1094 return(-EFAULT);
1095 volume.channel0 = 255;
1096 volume.channel1 = 255;
1097 volume.channel2 = 255;
1098 volume.channel3 = 255;
1099 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1100 return(-EFAULT);
1101 return(0);
1102 }
1103 return(-EINVAL);
1104 }
1105
1106 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1107 {
1108 struct uml_stat buf1, buf2;
1109 int err;
1110
1111 if(from_cmdline == NULL)
1112 return 0;
1113 if(!strcmp(from_cmdline, from_cow))
1114 return 0;
1115
1116 err = os_stat_file(from_cmdline, &buf1);
1117 if(err < 0){
1118 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1119 return 0;
1120 }
1121 err = os_stat_file(from_cow, &buf2);
1122 if(err < 0){
1123 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1124 return 1;
1125 }
1126 if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1127 return 0;
1128
1129 printk("Backing file mismatch - \"%s\" requested,\n"
1130 "\"%s\" specified in COW header of \"%s\"\n",
1131 from_cmdline, from_cow, cow);
1132 return 1;
1133 }
1134
1135 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1136 {
1137 unsigned long modtime;
1138 unsigned long long actual;
1139 int err;
1140
1141 err = os_file_modtime(file, &modtime);
1142 if(err < 0){
1143 printk("Failed to get modification time of backing file "
1144 "\"%s\", err = %d\n", file, -err);
1145 return(err);
1146 }
1147
1148 err = os_file_size(file, &actual);
1149 if(err < 0){
1150 printk("Failed to get size of backing file \"%s\", "
1151 "err = %d\n", file, -err);
1152 return(err);
1153 }
1154
1155 if(actual != size){
1156 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1157 * the typecast.*/
1158 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1159 "file\n", (unsigned long long) size, actual);
1160 return(-EINVAL);
1161 }
1162 if(modtime != mtime){
1163 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1164 "file\n", mtime, modtime);
1165 return(-EINVAL);
1166 }
1167 return(0);
1168 }
1169
1170 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1171 {
1172 int err;
1173
1174 err = os_seek_file(fd, offset);
1175 if(err < 0)
1176 return(err);
1177
1178 err = os_read_file(fd, buf, len);
1179 if(err < 0)
1180 return(err);
1181
1182 return(0);
1183 }
1184
1185 int open_ubd_file(char *file, struct openflags *openflags,
1186 char **backing_file_out, int *bitmap_offset_out,
1187 unsigned long *bitmap_len_out, int *data_offset_out,
1188 int *create_cow_out)
1189 {
1190 time_t mtime;
1191 unsigned long long size;
1192 __u32 version, align;
1193 char *backing_file;
1194 int fd, err, sectorsize, asked_switch, mode = 0644;
1195
1196 fd = os_open_file(file, *openflags, mode);
1197 if (fd < 0) {
1198 if ((fd == -ENOENT) && (create_cow_out != NULL))
1199 *create_cow_out = 1;
1200 if (!openflags->w ||
1201 ((fd != -EROFS) && (fd != -EACCES)))
1202 return fd;
1203 openflags->w = 0;
1204 fd = os_open_file(file, *openflags, mode);
1205 if (fd < 0)
1206 return fd;
1207 }
1208
1209 err = os_lock_file(fd, openflags->w);
1210 if(err < 0){
1211 printk("Failed to lock '%s', err = %d\n", file, -err);
1212 goto out_close;
1213 }
1214
1215 /* Succesful return case! */
1216 if(backing_file_out == NULL)
1217 return(fd);
1218
1219 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1220 &size, &sectorsize, &align, bitmap_offset_out);
1221 if(err && (*backing_file_out != NULL)){
1222 printk("Failed to read COW header from COW file \"%s\", "
1223 "errno = %d\n", file, -err);
1224 goto out_close;
1225 }
1226 if(err)
1227 return(fd);
1228
1229 asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1230
1231 /* Allow switching only if no mismatch. */
1232 if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1233 printk("Switching backing file to '%s'\n", *backing_file_out);
1234 err = write_cow_header(file, fd, *backing_file_out,
1235 sectorsize, align, &size);
1236 if (err) {
1237 printk("Switch failed, errno = %d\n", -err);
1238 goto out_close;
1239 }
1240 } else {
1241 *backing_file_out = backing_file;
1242 err = backing_file_mismatch(*backing_file_out, size, mtime);
1243 if (err)
1244 goto out_close;
1245 }
1246
1247 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1248 bitmap_len_out, data_offset_out);
1249
1250 return fd;
1251 out_close:
1252 os_close_file(fd);
1253 return err;
1254 }
1255
1256 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1257 int sectorsize, int alignment, int *bitmap_offset_out,
1258 unsigned long *bitmap_len_out, int *data_offset_out)
1259 {
1260 int err, fd;
1261
1262 flags.c = 1;
1263 fd = open_ubd_file(cow_file, &flags, NULL, NULL, NULL, NULL, NULL);
1264 if(fd < 0){
1265 err = fd;
1266 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1267 -err);
1268 goto out;
1269 }
1270
1271 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1272 bitmap_offset_out, bitmap_len_out,
1273 data_offset_out);
1274 if(!err)
1275 return(fd);
1276 os_close_file(fd);
1277 out:
1278 return(err);
1279 }
1280
1281 static int update_bitmap(struct io_thread_req *req)
1282 {
1283 int n;
1284
1285 if(req->cow_offset == -1)
1286 return(0);
1287
1288 n = os_seek_file(req->fds[1], req->cow_offset);
1289 if(n < 0){
1290 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1291 return(1);
1292 }
1293
1294 n = os_write_file(req->fds[1], &req->bitmap_words,
1295 sizeof(req->bitmap_words));
1296 if(n != sizeof(req->bitmap_words)){
1297 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1298 req->fds[1]);
1299 return(1);
1300 }
1301
1302 return(0);
1303 }
1304
1305 void do_io(struct io_thread_req *req)
1306 {
1307 char *buf;
1308 unsigned long len;
1309 int n, nsectors, start, end, bit;
1310 int err;
1311 __u64 off;
1312
1313 nsectors = req->length / req->sectorsize;
1314 start = 0;
1315 do {
1316 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1317 end = start;
1318 while((end < nsectors) &&
1319 (ubd_test_bit(end, (unsigned char *)
1320 &req->sector_mask) == bit))
1321 end++;
1322
1323 off = req->offset + req->offsets[bit] +
1324 start * req->sectorsize;
1325 len = (end - start) * req->sectorsize;
1326 buf = &req->buffer[start * req->sectorsize];
1327
1328 err = os_seek_file(req->fds[bit], off);
1329 if(err < 0){
1330 printk("do_io - lseek failed : err = %d\n", -err);
1331 req->error = 1;
1332 return;
1333 }
1334 if(req->op == UBD_READ){
1335 n = 0;
1336 do {
1337 buf = &buf[n];
1338 len -= n;
1339 n = os_read_file(req->fds[bit], buf, len);
1340 if (n < 0) {
1341 printk("do_io - read failed, err = %d "
1342 "fd = %d\n", -n, req->fds[bit]);
1343 req->error = 1;
1344 return;
1345 }
1346 } while((n < len) && (n != 0));
1347 if (n < len) memset(&buf[n], 0, len - n);
1348 } else {
1349 n = os_write_file(req->fds[bit], buf, len);
1350 if(n != len){
1351 printk("do_io - write failed err = %d "
1352 "fd = %d\n", -n, req->fds[bit]);
1353 req->error = 1;
1354 return;
1355 }
1356 }
1357
1358 start = end;
1359 } while(start < nsectors);
1360
1361 req->error = update_bitmap(req);
1362 }
1363
1364 /* Changed in start_io_thread, which is serialized by being called only
1365 * from ubd_init, which is an initcall.
1366 */
1367 int kernel_fd = -1;
1368
1369 /* Only changed by the io thread */
1370 int io_count = 0;
1371
1372 int io_thread(void *arg)
1373 {
1374 struct io_thread_req req;
1375 int n;
1376
1377 ignore_sigwinch_sig();
1378 while(1){
1379 n = os_read_file(kernel_fd, &req, sizeof(req));
1380 if(n != sizeof(req)){
1381 if(n < 0)
1382 printk("io_thread - read failed, fd = %d, "
1383 "err = %d\n", kernel_fd, -n);
1384 else {
1385 printk("io_thread - short read, fd = %d, "
1386 "length = %d\n", kernel_fd, n);
1387 }
1388 continue;
1389 }
1390 io_count++;
1391 do_io(&req);
1392 n = os_write_file(kernel_fd, &req, sizeof(req));
1393 if(n != sizeof(req))
1394 printk("io_thread - write failed, fd = %d, err = %d\n",
1395 kernel_fd, -n);
1396 }
1397
1398 return 0;
1399 }