Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[deliverable/linux.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/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 "user_util.h"
43 #include "mem_user.h"
44 #include "kern_util.h"
45 #include "kern.h"
46 #include "mconsole_kern.h"
47 #include "init.h"
48 #include "irq_user.h"
49 #include "irq_kern.h"
50 #include "ubd_user.h"
51 #include "os.h"
52 #include "mem.h"
53 #include "mem_kern.h"
54 #include "cow.h"
55
56 enum ubd_req { UBD_READ, UBD_WRITE };
57
58 struct io_thread_req {
59 struct request *req;
60 enum ubd_req op;
61 int fds[2];
62 unsigned long offsets[2];
63 unsigned long long offset;
64 unsigned long length;
65 char *buffer;
66 int sectorsize;
67 unsigned long sector_mask;
68 unsigned long long cow_offset;
69 unsigned long bitmap_words[2];
70 int error;
71 };
72
73 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
74 char **backing_file_out, int *bitmap_offset_out,
75 unsigned long *bitmap_len_out, int *data_offset_out,
76 int *create_cow_out);
77 extern int create_cow_file(char *cow_file, char *backing_file,
78 struct openflags flags, int sectorsize,
79 int alignment, int *bitmap_offset_out,
80 unsigned long *bitmap_len_out,
81 int *data_offset_out);
82 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
83 extern void do_io(struct io_thread_req *req);
84
85 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
86 {
87 __u64 n;
88 int bits, off;
89
90 bits = sizeof(data[0]) * 8;
91 n = bit / bits;
92 off = bit % bits;
93 return((data[n] & (1 << off)) != 0);
94 }
95
96 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
97 {
98 __u64 n;
99 int bits, off;
100
101 bits = sizeof(data[0]) * 8;
102 n = bit / bits;
103 off = bit % bits;
104 data[n] |= (1 << off);
105 }
106 /*End stuff from ubd_user.h*/
107
108 #define DRIVER_NAME "uml-blkdev"
109
110 static DEFINE_MUTEX(ubd_lock);
111
112 static int ubd_open(struct inode * inode, struct file * filp);
113 static int ubd_release(struct inode * inode, struct file * file);
114 static int ubd_ioctl(struct inode * inode, struct file * file,
115 unsigned int cmd, unsigned long arg);
116 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
117
118 #define MAX_DEV (16)
119
120 static struct block_device_operations ubd_blops = {
121 .owner = THIS_MODULE,
122 .open = ubd_open,
123 .release = ubd_release,
124 .ioctl = ubd_ioctl,
125 .getgeo = ubd_getgeo,
126 };
127
128 /* Protected by ubd_lock */
129 static int fake_major = MAJOR_NR;
130 static struct gendisk *ubd_gendisk[MAX_DEV];
131 static struct gendisk *fake_gendisk[MAX_DEV];
132
133 #ifdef CONFIG_BLK_DEV_UBD_SYNC
134 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
135 .cl = 1 })
136 #else
137 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
138 .cl = 1 })
139 #endif
140 static struct openflags global_openflags = OPEN_FLAGS;
141
142 struct cow {
143 /* backing file name */
144 char *file;
145 /* backing file fd */
146 int fd;
147 unsigned long *bitmap;
148 unsigned long bitmap_len;
149 int bitmap_offset;
150 int data_offset;
151 };
152
153 struct ubd {
154 /* name (and fd, below) of the file opened for writing, either the
155 * backing or the cow file. */
156 char *file;
157 int count;
158 int fd;
159 __u64 size;
160 struct openflags boot_openflags;
161 struct openflags openflags;
162 unsigned shared:1;
163 unsigned no_cow:1;
164 struct cow cow;
165 struct platform_device pdev;
166 struct request_queue *queue;
167 spinlock_t lock;
168 int active;
169 };
170
171 #define DEFAULT_COW { \
172 .file = NULL, \
173 .fd = -1, \
174 .bitmap = NULL, \
175 .bitmap_offset = 0, \
176 .data_offset = 0, \
177 }
178
179 #define DEFAULT_UBD { \
180 .file = NULL, \
181 .count = 0, \
182 .fd = -1, \
183 .size = -1, \
184 .boot_openflags = OPEN_FLAGS, \
185 .openflags = OPEN_FLAGS, \
186 .no_cow = 0, \
187 .shared = 0, \
188 .cow = DEFAULT_COW, \
189 .lock = SPIN_LOCK_UNLOCKED, \
190 .active = 0, \
191 }
192
193 /* Protected by ubd_lock */
194 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
195
196 /* Only changed by fake_ide_setup which is a setup */
197 static int fake_ide = 0;
198 static struct proc_dir_entry *proc_ide_root = NULL;
199 static struct proc_dir_entry *proc_ide = NULL;
200
201 static void make_proc_ide(void)
202 {
203 proc_ide_root = proc_mkdir("ide", NULL);
204 proc_ide = proc_mkdir("ide0", proc_ide_root);
205 }
206
207 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
208 int *eof, void *data)
209 {
210 int len;
211
212 strcpy(page, "disk\n");
213 len = strlen("disk\n");
214 len -= off;
215 if (len < count){
216 *eof = 1;
217 if (len <= 0) return 0;
218 }
219 else len = count;
220 *start = page + off;
221 return len;
222 }
223
224 static void make_ide_entries(char *dev_name)
225 {
226 struct proc_dir_entry *dir, *ent;
227 char name[64];
228
229 if(proc_ide_root == NULL) make_proc_ide();
230
231 dir = proc_mkdir(dev_name, proc_ide);
232 if(!dir) return;
233
234 ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
235 if(!ent) return;
236 ent->data = NULL;
237 ent->read_proc = proc_ide_read_media;
238 ent->write_proc = NULL;
239 sprintf(name,"ide0/%s", dev_name);
240 proc_symlink(dev_name, proc_ide_root, name);
241 }
242
243 static int fake_ide_setup(char *str)
244 {
245 fake_ide = 1;
246 return(1);
247 }
248
249 __setup("fake_ide", fake_ide_setup);
250
251 __uml_help(fake_ide_setup,
252 "fake_ide\n"
253 " Create ide0 entries that map onto ubd devices.\n\n"
254 );
255
256 static int parse_unit(char **ptr)
257 {
258 char *str = *ptr, *end;
259 int n = -1;
260
261 if(isdigit(*str)) {
262 n = simple_strtoul(str, &end, 0);
263 if(end == str)
264 return(-1);
265 *ptr = end;
266 }
267 else if (('a' <= *str) && (*str <= 'z')) {
268 n = *str - 'a';
269 str++;
270 *ptr = str;
271 }
272 return(n);
273 }
274
275 /* If *index_out == -1 at exit, the passed option was a general one;
276 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
277 * should not be freed on exit.
278 */
279 static int ubd_setup_common(char *str, int *index_out, char **error_out)
280 {
281 struct ubd *ubd_dev;
282 struct openflags flags = global_openflags;
283 char *backing_file;
284 int n, err = 0, i;
285
286 if(index_out) *index_out = -1;
287 n = *str;
288 if(n == '='){
289 char *end;
290 int major;
291
292 str++;
293 if(!strcmp(str, "sync")){
294 global_openflags = of_sync(global_openflags);
295 goto out1;
296 }
297
298 err = -EINVAL;
299 major = simple_strtoul(str, &end, 0);
300 if((*end != '\0') || (end == str)){
301 *error_out = "Didn't parse major number";
302 goto out1;
303 }
304
305 mutex_lock(&ubd_lock);
306 if(fake_major != MAJOR_NR){
307 *error_out = "Can't assign a fake major twice";
308 goto out1;
309 }
310
311 fake_major = major;
312
313 printk(KERN_INFO "Setting extra ubd major number to %d\n",
314 major);
315 err = 0;
316 out1:
317 mutex_unlock(&ubd_lock);
318 return err;
319 }
320
321 n = parse_unit(&str);
322 if(n < 0){
323 *error_out = "Couldn't parse device number";
324 return -EINVAL;
325 }
326 if(n >= MAX_DEV){
327 *error_out = "Device number out of range";
328 return 1;
329 }
330
331 err = -EBUSY;
332 mutex_lock(&ubd_lock);
333
334 ubd_dev = &ubd_devs[n];
335 if(ubd_dev->file != NULL){
336 *error_out = "Device is already configured";
337 goto out;
338 }
339
340 if (index_out)
341 *index_out = n;
342
343 err = -EINVAL;
344 for (i = 0; i < sizeof("rscd="); i++) {
345 switch (*str) {
346 case 'r':
347 flags.w = 0;
348 break;
349 case 's':
350 flags.s = 1;
351 break;
352 case 'd':
353 ubd_dev->no_cow = 1;
354 break;
355 case 'c':
356 ubd_dev->shared = 1;
357 break;
358 case '=':
359 str++;
360 goto break_loop;
361 default:
362 *error_out = "Expected '=' or flag letter "
363 "(r, s, c, or d)";
364 goto out;
365 }
366 str++;
367 }
368
369 if (*str == '=')
370 *error_out = "Too many flags specified";
371 else
372 *error_out = "Missing '='";
373 goto out;
374
375 break_loop:
376 backing_file = strchr(str, ',');
377
378 if (backing_file == NULL)
379 backing_file = strchr(str, ':');
380
381 if(backing_file != NULL){
382 if(ubd_dev->no_cow){
383 *error_out = "Can't specify both 'd' and a cow file";
384 goto out;
385 }
386 else {
387 *backing_file = '\0';
388 backing_file++;
389 }
390 }
391 err = 0;
392 ubd_dev->file = str;
393 ubd_dev->cow.file = backing_file;
394 ubd_dev->boot_openflags = flags;
395 out:
396 mutex_unlock(&ubd_lock);
397 return err;
398 }
399
400 static int ubd_setup(char *str)
401 {
402 char *error;
403 int err;
404
405 err = ubd_setup_common(str, NULL, &error);
406 if(err)
407 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
408 "%s\n", str, error);
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 treat 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 /* call ubd_finish if you need to serialize */
471 static void __ubd_finish(struct request *req, int error)
472 {
473 int nsect;
474
475 if(error){
476 end_request(req, 0);
477 return;
478 }
479 nsect = req->current_nr_sectors;
480 req->sector += nsect;
481 req->buffer += nsect << 9;
482 req->errors = 0;
483 req->nr_sectors -= nsect;
484 req->current_nr_sectors = 0;
485 end_request(req, 1);
486 }
487
488 /* Callable only from interrupt context - otherwise you need to do
489 * spin_lock_irq()/spin_lock_irqsave() */
490 static inline void ubd_finish(struct request *req, int error)
491 {
492 struct ubd *dev = req->rq_disk->private_data;
493
494 spin_lock(&dev->lock);
495 __ubd_finish(req, error);
496 spin_unlock(&dev->lock);
497 }
498
499 /* XXX - move this inside ubd_intr. */
500 /* Called without dev->lock held, and only in interrupt context. */
501 static void ubd_handler(void)
502 {
503 struct io_thread_req req;
504 struct request *rq;
505 struct ubd *dev;
506 int n;
507
508 n = os_read_file(thread_fd, &req, sizeof(req));
509 if(n != sizeof(req)){
510 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
511 "err = %d\n", os_getpid(), -n);
512 return;
513 }
514
515 rq = req.req;
516 dev = rq->rq_disk->private_data;
517 dev->active = 0;
518
519 ubd_finish(rq, req.error);
520 reactivate_fd(thread_fd, UBD_IRQ);
521 spin_lock(&dev->lock);
522 do_ubd_request(dev->queue);
523 spin_unlock(&dev->lock);
524 }
525
526 static irqreturn_t ubd_intr(int irq, void *dev)
527 {
528 ubd_handler();
529 return(IRQ_HANDLED);
530 }
531
532 /* Only changed by ubd_init, which is an initcall. */
533 static int io_pid = -1;
534
535 void kill_io_thread(void)
536 {
537 if(io_pid != -1)
538 os_kill_process(io_pid, 1);
539 }
540
541 __uml_exitcall(kill_io_thread);
542
543 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
544 {
545 char *file;
546
547 file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
548 return(os_file_size(file, size_out));
549 }
550
551 static void ubd_close_dev(struct ubd *ubd_dev)
552 {
553 os_close_file(ubd_dev->fd);
554 if(ubd_dev->cow.file == NULL)
555 return;
556
557 os_close_file(ubd_dev->cow.fd);
558 vfree(ubd_dev->cow.bitmap);
559 ubd_dev->cow.bitmap = NULL;
560 }
561
562 static int ubd_open_dev(struct ubd *ubd_dev)
563 {
564 struct openflags flags;
565 char **back_ptr;
566 int err, create_cow, *create_ptr;
567 int fd;
568
569 ubd_dev->openflags = ubd_dev->boot_openflags;
570 create_cow = 0;
571 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
572 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
573
574 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
575 back_ptr, &ubd_dev->cow.bitmap_offset,
576 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
577 create_ptr);
578
579 if((fd == -ENOENT) && create_cow){
580 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
581 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
582 &ubd_dev->cow.bitmap_offset,
583 &ubd_dev->cow.bitmap_len,
584 &ubd_dev->cow.data_offset);
585 if(fd >= 0){
586 printk(KERN_INFO "Creating \"%s\" as COW file for "
587 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
588 }
589 }
590
591 if(fd < 0){
592 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
593 -fd);
594 return fd;
595 }
596 ubd_dev->fd = fd;
597
598 if(ubd_dev->cow.file != NULL){
599 err = -ENOMEM;
600 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
601 if(ubd_dev->cow.bitmap == NULL){
602 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
603 goto error;
604 }
605 flush_tlb_kernel_vm();
606
607 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
608 ubd_dev->cow.bitmap_offset,
609 ubd_dev->cow.bitmap_len);
610 if(err < 0)
611 goto error;
612
613 flags = ubd_dev->openflags;
614 flags.w = 0;
615 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
616 NULL, NULL, NULL, NULL);
617 if(err < 0) goto error;
618 ubd_dev->cow.fd = err;
619 }
620 return(0);
621 error:
622 os_close_file(ubd_dev->fd);
623 return(err);
624 }
625
626 static int ubd_disk_register(int major, u64 size, int unit,
627 struct gendisk **disk_out)
628 {
629 struct gendisk *disk;
630
631 disk = alloc_disk(1 << UBD_SHIFT);
632 if(disk == NULL)
633 return(-ENOMEM);
634
635 disk->major = major;
636 disk->first_minor = unit << UBD_SHIFT;
637 disk->fops = &ubd_blops;
638 set_capacity(disk, size / 512);
639 if(major == MAJOR_NR)
640 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
641 else
642 sprintf(disk->disk_name, "ubd_fake%d", unit);
643
644 /* sysfs register (not for ide fake devices) */
645 if (major == MAJOR_NR) {
646 ubd_devs[unit].pdev.id = unit;
647 ubd_devs[unit].pdev.name = DRIVER_NAME;
648 platform_device_register(&ubd_devs[unit].pdev);
649 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
650 }
651
652 disk->private_data = &ubd_devs[unit];
653 disk->queue = ubd_devs[unit].queue;
654 add_disk(disk);
655
656 *disk_out = disk;
657 return 0;
658 }
659
660 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
661
662 static int ubd_add(int n, char **error_out)
663 {
664 struct ubd *ubd_dev = &ubd_devs[n];
665 int err = 0;
666
667 if(ubd_dev->file == NULL)
668 goto out;
669
670 err = ubd_file_size(ubd_dev, &ubd_dev->size);
671 if(err < 0){
672 *error_out = "Couldn't determine size of device's file";
673 goto out;
674 }
675
676 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
677
678 err = -ENOMEM;
679 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
680 if (ubd_dev->queue == NULL) {
681 *error_out = "Failed to initialize device queue";
682 goto out;
683 }
684 ubd_dev->queue->queuedata = ubd_dev;
685
686 err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
687 if(err){
688 *error_out = "Failed to register device";
689 goto out_cleanup;
690 }
691
692 if(fake_major != MAJOR_NR)
693 ubd_disk_register(fake_major, ubd_dev->size, n,
694 &fake_gendisk[n]);
695
696 /* perhaps this should also be under the "if (fake_major)" above */
697 /* using the fake_disk->disk_name and also the fakehd_set name */
698 if (fake_ide)
699 make_ide_entries(ubd_gendisk[n]->disk_name);
700
701 err = 0;
702 out:
703 return err;
704
705 out_cleanup:
706 blk_cleanup_queue(ubd_dev->queue);
707 goto out;
708 }
709
710 static int ubd_config(char *str, char **error_out)
711 {
712 int n, ret;
713
714 /* This string is possibly broken up and stored, so it's only
715 * freed if ubd_setup_common fails, or if only general options
716 * were set.
717 */
718 str = kstrdup(str, GFP_KERNEL);
719 if (str == NULL) {
720 *error_out = "Failed to allocate memory";
721 return -ENOMEM;
722 }
723
724 ret = ubd_setup_common(str, &n, error_out);
725 if (ret)
726 goto err_free;
727
728 if (n == -1) {
729 ret = 0;
730 goto err_free;
731 }
732
733 mutex_lock(&ubd_lock);
734 ret = ubd_add(n, error_out);
735 if (ret)
736 ubd_devs[n].file = NULL;
737 mutex_unlock(&ubd_lock);
738
739 out:
740 return ret;
741
742 err_free:
743 kfree(str);
744 goto out;
745 }
746
747 static int ubd_get_config(char *name, char *str, int size, char **error_out)
748 {
749 struct ubd *ubd_dev;
750 int n, len = 0;
751
752 n = parse_unit(&name);
753 if((n >= MAX_DEV) || (n < 0)){
754 *error_out = "ubd_get_config : device number out of range";
755 return(-1);
756 }
757
758 ubd_dev = &ubd_devs[n];
759 mutex_lock(&ubd_lock);
760
761 if(ubd_dev->file == NULL){
762 CONFIG_CHUNK(str, size, len, "", 1);
763 goto out;
764 }
765
766 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
767
768 if(ubd_dev->cow.file != NULL){
769 CONFIG_CHUNK(str, size, len, ",", 0);
770 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
771 }
772 else CONFIG_CHUNK(str, size, len, "", 1);
773
774 out:
775 mutex_unlock(&ubd_lock);
776 return(len);
777 }
778
779 static int ubd_id(char **str, int *start_out, int *end_out)
780 {
781 int n;
782
783 n = parse_unit(str);
784 *start_out = 0;
785 *end_out = MAX_DEV - 1;
786 return n;
787 }
788
789 static int ubd_remove(int n, char **error_out)
790 {
791 struct ubd *ubd_dev;
792 int err = -ENODEV;
793
794 mutex_lock(&ubd_lock);
795
796 if(ubd_gendisk[n] == NULL)
797 goto out;
798
799 ubd_dev = &ubd_devs[n];
800
801 if(ubd_dev->file == NULL)
802 goto out;
803
804 /* you cannot remove a open disk */
805 err = -EBUSY;
806 if(ubd_dev->count > 0)
807 goto out;
808
809 del_gendisk(ubd_gendisk[n]);
810 put_disk(ubd_gendisk[n]);
811 ubd_gendisk[n] = NULL;
812
813 if(fake_gendisk[n] != NULL){
814 del_gendisk(fake_gendisk[n]);
815 put_disk(fake_gendisk[n]);
816 fake_gendisk[n] = NULL;
817 }
818
819 blk_cleanup_queue(ubd_dev->queue);
820 platform_device_unregister(&ubd_dev->pdev);
821 *ubd_dev = ((struct ubd) DEFAULT_UBD);
822 err = 0;
823 out:
824 mutex_unlock(&ubd_lock);
825 return err;
826 }
827
828 /* All these are called by mconsole in process context and without
829 * ubd-specific locks. The structure itself is const except for .list.
830 */
831 static struct mc_device ubd_mc = {
832 .list = LIST_HEAD_INIT(ubd_mc.list),
833 .name = "ubd",
834 .config = ubd_config,
835 .get_config = ubd_get_config,
836 .id = ubd_id,
837 .remove = ubd_remove,
838 };
839
840 static int __init ubd_mc_init(void)
841 {
842 mconsole_register_dev(&ubd_mc);
843 return 0;
844 }
845
846 __initcall(ubd_mc_init);
847
848 static int __init ubd0_init(void)
849 {
850 struct ubd *ubd_dev = &ubd_devs[0];
851
852 mutex_lock(&ubd_lock);
853 if(ubd_dev->file == NULL)
854 ubd_dev->file = "root_fs";
855 mutex_unlock(&ubd_lock);
856
857 return(0);
858 }
859
860 __initcall(ubd0_init);
861
862 /* Used in ubd_init, which is an initcall */
863 static struct platform_driver ubd_driver = {
864 .driver = {
865 .name = DRIVER_NAME,
866 },
867 };
868
869 static int __init ubd_init(void)
870 {
871 char *error;
872 int i, err;
873
874 if (register_blkdev(MAJOR_NR, "ubd"))
875 return -1;
876
877 if (fake_major != MAJOR_NR) {
878 char name[sizeof("ubd_nnn\0")];
879
880 snprintf(name, sizeof(name), "ubd_%d", fake_major);
881 if (register_blkdev(fake_major, "ubd"))
882 return -1;
883 }
884 platform_driver_register(&ubd_driver);
885 mutex_lock(&ubd_lock);
886 for (i = 0; i < MAX_DEV; i++){
887 err = ubd_add(i, &error);
888 if(err)
889 printk(KERN_ERR "Failed to initialize ubd device %d :"
890 "%s\n", i, error);
891 }
892 mutex_unlock(&ubd_lock);
893 return 0;
894 }
895
896 late_initcall(ubd_init);
897
898 static int __init ubd_driver_init(void){
899 unsigned long stack;
900 int err;
901
902 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
903 if(global_openflags.s){
904 printk(KERN_INFO "ubd: Synchronous mode\n");
905 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
906 * enough. So use anyway the io thread. */
907 }
908 stack = alloc_stack(0, 0);
909 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
910 &thread_fd);
911 if(io_pid < 0){
912 printk(KERN_ERR
913 "ubd : Failed to start I/O thread (errno = %d) - "
914 "falling back to synchronous I/O\n", -io_pid);
915 io_pid = -1;
916 return(0);
917 }
918 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
919 IRQF_DISABLED, "ubd", ubd_devs);
920 if(err != 0)
921 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
922 return 0;
923 }
924
925 device_initcall(ubd_driver_init);
926
927 static int ubd_open(struct inode *inode, struct file *filp)
928 {
929 struct gendisk *disk = inode->i_bdev->bd_disk;
930 struct ubd *ubd_dev = disk->private_data;
931 int err = 0;
932
933 if(ubd_dev->count == 0){
934 err = ubd_open_dev(ubd_dev);
935 if(err){
936 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
937 disk->disk_name, ubd_dev->file, -err);
938 goto out;
939 }
940 }
941 ubd_dev->count++;
942 set_disk_ro(disk, !ubd_dev->openflags.w);
943
944 /* This should no more be needed. And it didn't work anyway to exclude
945 * read-write remounting of filesystems.*/
946 /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
947 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
948 err = -EROFS;
949 }*/
950 out:
951 return(err);
952 }
953
954 static int ubd_release(struct inode * inode, struct file * file)
955 {
956 struct gendisk *disk = inode->i_bdev->bd_disk;
957 struct ubd *ubd_dev = disk->private_data;
958
959 if(--ubd_dev->count == 0)
960 ubd_close_dev(ubd_dev);
961 return(0);
962 }
963
964 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
965 __u64 *cow_offset, unsigned long *bitmap,
966 __u64 bitmap_offset, unsigned long *bitmap_words,
967 __u64 bitmap_len)
968 {
969 __u64 sector = io_offset >> 9;
970 int i, update_bitmap = 0;
971
972 for(i = 0; i < length >> 9; i++){
973 if(cow_mask != NULL)
974 ubd_set_bit(i, (unsigned char *) cow_mask);
975 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
976 continue;
977
978 update_bitmap = 1;
979 ubd_set_bit(sector + i, (unsigned char *) bitmap);
980 }
981
982 if(!update_bitmap)
983 return;
984
985 *cow_offset = sector / (sizeof(unsigned long) * 8);
986
987 /* This takes care of the case where we're exactly at the end of the
988 * device, and *cow_offset + 1 is off the end. So, just back it up
989 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
990 * for the original diagnosis.
991 */
992 if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
993 sizeof(unsigned long) - 1))
994 (*cow_offset)--;
995
996 bitmap_words[0] = bitmap[*cow_offset];
997 bitmap_words[1] = bitmap[*cow_offset + 1];
998
999 *cow_offset *= sizeof(unsigned long);
1000 *cow_offset += bitmap_offset;
1001 }
1002
1003 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1004 __u64 bitmap_offset, __u64 bitmap_len)
1005 {
1006 __u64 sector = req->offset >> 9;
1007 int i;
1008
1009 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1010 panic("Operation too long");
1011
1012 if(req->op == UBD_READ) {
1013 for(i = 0; i < req->length >> 9; i++){
1014 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1015 ubd_set_bit(i, (unsigned char *)
1016 &req->sector_mask);
1017 }
1018 }
1019 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1020 &req->cow_offset, bitmap, bitmap_offset,
1021 req->bitmap_words, bitmap_len);
1022 }
1023
1024 /* Called with dev->lock held */
1025 static int prepare_request(struct request *req, struct io_thread_req *io_req)
1026 {
1027 struct gendisk *disk = req->rq_disk;
1028 struct ubd *ubd_dev = disk->private_data;
1029 __u64 offset;
1030 int len;
1031
1032 /* This should be impossible now */
1033 if((rq_data_dir(req) == WRITE) && !ubd_dev->openflags.w){
1034 printk("Write attempted on readonly ubd device %s\n",
1035 disk->disk_name);
1036 end_request(req, 0);
1037 return(1);
1038 }
1039
1040 offset = ((__u64) req->sector) << 9;
1041 len = req->current_nr_sectors << 9;
1042
1043 io_req->req = req;
1044 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd : ubd_dev->fd;
1045 io_req->fds[1] = ubd_dev->fd;
1046 io_req->cow_offset = -1;
1047 io_req->offset = offset;
1048 io_req->length = len;
1049 io_req->error = 0;
1050 io_req->sector_mask = 0;
1051
1052 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1053 io_req->offsets[0] = 0;
1054 io_req->offsets[1] = ubd_dev->cow.data_offset;
1055 io_req->buffer = req->buffer;
1056 io_req->sectorsize = 1 << 9;
1057
1058 if(ubd_dev->cow.file != NULL)
1059 cowify_req(io_req, ubd_dev->cow.bitmap, ubd_dev->cow.bitmap_offset,
1060 ubd_dev->cow.bitmap_len);
1061
1062 return(0);
1063 }
1064
1065 /* Called with dev->lock held */
1066 static void do_ubd_request(request_queue_t *q)
1067 {
1068 struct io_thread_req io_req;
1069 struct request *req;
1070 int err, n;
1071
1072 if(thread_fd == -1){
1073 while((req = elv_next_request(q)) != NULL){
1074 err = prepare_request(req, &io_req);
1075 if(!err){
1076 do_io(&io_req);
1077 __ubd_finish(req, io_req.error);
1078 }
1079 }
1080 }
1081 else {
1082 struct ubd *dev = q->queuedata;
1083 if(dev->active || (req = elv_next_request(q)) == NULL)
1084 return;
1085 err = prepare_request(req, &io_req);
1086 if(!err){
1087 dev->active = 1;
1088 n = os_write_file(thread_fd, (char *) &io_req,
1089 sizeof(io_req));
1090 if(n != sizeof(io_req))
1091 printk("write to io thread failed, "
1092 "errno = %d\n", -n);
1093 }
1094 }
1095 }
1096
1097 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1098 {
1099 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1100
1101 geo->heads = 128;
1102 geo->sectors = 32;
1103 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1104 return 0;
1105 }
1106
1107 static int ubd_ioctl(struct inode * inode, struct file * file,
1108 unsigned int cmd, unsigned long arg)
1109 {
1110 struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1111 struct hd_driveid ubd_id = {
1112 .cyls = 0,
1113 .heads = 128,
1114 .sectors = 32,
1115 };
1116
1117 switch (cmd) {
1118 struct cdrom_volctrl volume;
1119 case HDIO_GET_IDENTITY:
1120 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1121 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1122 sizeof(ubd_id)))
1123 return(-EFAULT);
1124 return(0);
1125
1126 case CDROMVOLREAD:
1127 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1128 return(-EFAULT);
1129 volume.channel0 = 255;
1130 volume.channel1 = 255;
1131 volume.channel2 = 255;
1132 volume.channel3 = 255;
1133 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1134 return(-EFAULT);
1135 return(0);
1136 }
1137 return(-EINVAL);
1138 }
1139
1140 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1141 {
1142 struct uml_stat buf1, buf2;
1143 int err;
1144
1145 if(from_cmdline == NULL)
1146 return 0;
1147 if(!strcmp(from_cmdline, from_cow))
1148 return 0;
1149
1150 err = os_stat_file(from_cmdline, &buf1);
1151 if(err < 0){
1152 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1153 return 0;
1154 }
1155 err = os_stat_file(from_cow, &buf2);
1156 if(err < 0){
1157 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1158 return 1;
1159 }
1160 if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1161 return 0;
1162
1163 printk("Backing file mismatch - \"%s\" requested,\n"
1164 "\"%s\" specified in COW header of \"%s\"\n",
1165 from_cmdline, from_cow, cow);
1166 return 1;
1167 }
1168
1169 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1170 {
1171 unsigned long modtime;
1172 unsigned long long actual;
1173 int err;
1174
1175 err = os_file_modtime(file, &modtime);
1176 if(err < 0){
1177 printk("Failed to get modification time of backing file "
1178 "\"%s\", err = %d\n", file, -err);
1179 return(err);
1180 }
1181
1182 err = os_file_size(file, &actual);
1183 if(err < 0){
1184 printk("Failed to get size of backing file \"%s\", "
1185 "err = %d\n", file, -err);
1186 return(err);
1187 }
1188
1189 if(actual != size){
1190 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1191 * the typecast.*/
1192 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1193 "file\n", (unsigned long long) size, actual);
1194 return(-EINVAL);
1195 }
1196 if(modtime != mtime){
1197 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1198 "file\n", mtime, modtime);
1199 return(-EINVAL);
1200 }
1201 return(0);
1202 }
1203
1204 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1205 {
1206 int err;
1207
1208 err = os_seek_file(fd, offset);
1209 if(err < 0)
1210 return(err);
1211
1212 err = os_read_file(fd, buf, len);
1213 if(err < 0)
1214 return(err);
1215
1216 return(0);
1217 }
1218
1219 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1220 char **backing_file_out, int *bitmap_offset_out,
1221 unsigned long *bitmap_len_out, int *data_offset_out,
1222 int *create_cow_out)
1223 {
1224 time_t mtime;
1225 unsigned long long size;
1226 __u32 version, align;
1227 char *backing_file;
1228 int fd, err, sectorsize, asked_switch, mode = 0644;
1229
1230 fd = os_open_file(file, *openflags, mode);
1231 if (fd < 0) {
1232 if ((fd == -ENOENT) && (create_cow_out != NULL))
1233 *create_cow_out = 1;
1234 if (!openflags->w ||
1235 ((fd != -EROFS) && (fd != -EACCES)))
1236 return fd;
1237 openflags->w = 0;
1238 fd = os_open_file(file, *openflags, mode);
1239 if (fd < 0)
1240 return fd;
1241 }
1242
1243 if(shared)
1244 printk("Not locking \"%s\" on the host\n", file);
1245 else {
1246 err = os_lock_file(fd, openflags->w);
1247 if(err < 0){
1248 printk("Failed to lock '%s', err = %d\n", file, -err);
1249 goto out_close;
1250 }
1251 }
1252
1253 /* Successful return case! */
1254 if(backing_file_out == NULL)
1255 return(fd);
1256
1257 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1258 &size, &sectorsize, &align, bitmap_offset_out);
1259 if(err && (*backing_file_out != NULL)){
1260 printk("Failed to read COW header from COW file \"%s\", "
1261 "errno = %d\n", file, -err);
1262 goto out_close;
1263 }
1264 if(err)
1265 return(fd);
1266
1267 asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1268
1269 /* Allow switching only if no mismatch. */
1270 if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1271 printk("Switching backing file to '%s'\n", *backing_file_out);
1272 err = write_cow_header(file, fd, *backing_file_out,
1273 sectorsize, align, &size);
1274 if (err) {
1275 printk("Switch failed, errno = %d\n", -err);
1276 goto out_close;
1277 }
1278 } else {
1279 *backing_file_out = backing_file;
1280 err = backing_file_mismatch(*backing_file_out, size, mtime);
1281 if (err)
1282 goto out_close;
1283 }
1284
1285 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1286 bitmap_len_out, data_offset_out);
1287
1288 return fd;
1289 out_close:
1290 os_close_file(fd);
1291 return err;
1292 }
1293
1294 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1295 int sectorsize, int alignment, int *bitmap_offset_out,
1296 unsigned long *bitmap_len_out, int *data_offset_out)
1297 {
1298 int err, fd;
1299
1300 flags.c = 1;
1301 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1302 if(fd < 0){
1303 err = fd;
1304 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1305 -err);
1306 goto out;
1307 }
1308
1309 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1310 bitmap_offset_out, bitmap_len_out,
1311 data_offset_out);
1312 if(!err)
1313 return(fd);
1314 os_close_file(fd);
1315 out:
1316 return(err);
1317 }
1318
1319 static int update_bitmap(struct io_thread_req *req)
1320 {
1321 int n;
1322
1323 if(req->cow_offset == -1)
1324 return(0);
1325
1326 n = os_seek_file(req->fds[1], req->cow_offset);
1327 if(n < 0){
1328 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1329 return(1);
1330 }
1331
1332 n = os_write_file(req->fds[1], &req->bitmap_words,
1333 sizeof(req->bitmap_words));
1334 if(n != sizeof(req->bitmap_words)){
1335 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1336 req->fds[1]);
1337 return(1);
1338 }
1339
1340 return(0);
1341 }
1342
1343 void do_io(struct io_thread_req *req)
1344 {
1345 char *buf;
1346 unsigned long len;
1347 int n, nsectors, start, end, bit;
1348 int err;
1349 __u64 off;
1350
1351 nsectors = req->length / req->sectorsize;
1352 start = 0;
1353 do {
1354 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1355 end = start;
1356 while((end < nsectors) &&
1357 (ubd_test_bit(end, (unsigned char *)
1358 &req->sector_mask) == bit))
1359 end++;
1360
1361 off = req->offset + req->offsets[bit] +
1362 start * req->sectorsize;
1363 len = (end - start) * req->sectorsize;
1364 buf = &req->buffer[start * req->sectorsize];
1365
1366 err = os_seek_file(req->fds[bit], off);
1367 if(err < 0){
1368 printk("do_io - lseek failed : err = %d\n", -err);
1369 req->error = 1;
1370 return;
1371 }
1372 if(req->op == UBD_READ){
1373 n = 0;
1374 do {
1375 buf = &buf[n];
1376 len -= n;
1377 n = os_read_file(req->fds[bit], buf, len);
1378 if (n < 0) {
1379 printk("do_io - read failed, err = %d "
1380 "fd = %d\n", -n, req->fds[bit]);
1381 req->error = 1;
1382 return;
1383 }
1384 } while((n < len) && (n != 0));
1385 if (n < len) memset(&buf[n], 0, len - n);
1386 } else {
1387 n = os_write_file(req->fds[bit], buf, len);
1388 if(n != len){
1389 printk("do_io - write failed err = %d "
1390 "fd = %d\n", -n, req->fds[bit]);
1391 req->error = 1;
1392 return;
1393 }
1394 }
1395
1396 start = end;
1397 } while(start < nsectors);
1398
1399 req->error = update_bitmap(req);
1400 }
1401
1402 /* Changed in start_io_thread, which is serialized by being called only
1403 * from ubd_init, which is an initcall.
1404 */
1405 int kernel_fd = -1;
1406
1407 /* Only changed by the io thread. XXX: currently unused. */
1408 static int io_count = 0;
1409
1410 int io_thread(void *arg)
1411 {
1412 struct io_thread_req req;
1413 int n;
1414
1415 ignore_sigwinch_sig();
1416 while(1){
1417 n = os_read_file(kernel_fd, &req, sizeof(req));
1418 if(n != sizeof(req)){
1419 if(n < 0)
1420 printk("io_thread - read failed, fd = %d, "
1421 "err = %d\n", kernel_fd, -n);
1422 else {
1423 printk("io_thread - short read, fd = %d, "
1424 "length = %d\n", kernel_fd, n);
1425 }
1426 continue;
1427 }
1428 io_count++;
1429 do_io(&req);
1430 n = os_write_file(kernel_fd, &req, sizeof(req));
1431 if(n != sizeof(req))
1432 printk("io_thread - write failed, fd = %d, err = %d\n",
1433 kernel_fd, -n);
1434 }
1435
1436 return 0;
1437 }
This page took 0.06065 seconds and 6 git commands to generate.