Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-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/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 "kern_util.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 struct request *req;
62 enum ubd_req op;
63 int fds[2];
64 unsigned long offsets[2];
65 unsigned long long offset;
66 unsigned long length;
67 char *buffer;
68 int sectorsize;
69 unsigned long sector_mask;
70 unsigned long long cow_offset;
71 unsigned long bitmap_words[2];
72 int error;
73 };
74
75 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
76 {
77 __u64 n;
78 int bits, off;
79
80 bits = sizeof(data[0]) * 8;
81 n = bit / bits;
82 off = bit % bits;
83 return (data[n] & (1 << off)) != 0;
84 }
85
86 static inline void ubd_set_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 data[n] |= (1 << off);
95 }
96 /*End stuff from ubd_user.h*/
97
98 #define DRIVER_NAME "uml-blkdev"
99
100 static DEFINE_MUTEX(ubd_lock);
101
102 static int ubd_open(struct inode * inode, struct file * filp);
103 static int ubd_release(struct inode * inode, struct file * file);
104 static int ubd_ioctl(struct inode * inode, struct file * file,
105 unsigned int cmd, unsigned long arg);
106 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
107
108 #define MAX_DEV (16)
109
110 static struct block_device_operations ubd_blops = {
111 .owner = THIS_MODULE,
112 .open = ubd_open,
113 .release = ubd_release,
114 .ioctl = ubd_ioctl,
115 .getgeo = ubd_getgeo,
116 };
117
118 /* Protected by ubd_lock */
119 static int fake_major = MAJOR_NR;
120 static struct gendisk *ubd_gendisk[MAX_DEV];
121 static struct gendisk *fake_gendisk[MAX_DEV];
122
123 #ifdef CONFIG_BLK_DEV_UBD_SYNC
124 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
125 .cl = 1 })
126 #else
127 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
128 .cl = 1 })
129 #endif
130 static struct openflags global_openflags = OPEN_FLAGS;
131
132 struct cow {
133 /* backing file name */
134 char *file;
135 /* backing file fd */
136 int fd;
137 unsigned long *bitmap;
138 unsigned long bitmap_len;
139 int bitmap_offset;
140 int data_offset;
141 };
142
143 #define MAX_SG 64
144
145 struct ubd {
146 struct list_head restart;
147 /* name (and fd, below) of the file opened for writing, either the
148 * backing or the cow file. */
149 char *file;
150 int count;
151 int fd;
152 __u64 size;
153 struct openflags boot_openflags;
154 struct openflags openflags;
155 unsigned shared:1;
156 unsigned no_cow:1;
157 struct cow cow;
158 struct platform_device pdev;
159 struct request_queue *queue;
160 spinlock_t lock;
161 struct scatterlist sg[MAX_SG];
162 struct request *request;
163 int start_sg, end_sg;
164 };
165
166 #define DEFAULT_COW { \
167 .file = NULL, \
168 .fd = -1, \
169 .bitmap = NULL, \
170 .bitmap_offset = 0, \
171 .data_offset = 0, \
172 }
173
174 #define DEFAULT_UBD { \
175 .file = NULL, \
176 .count = 0, \
177 .fd = -1, \
178 .size = -1, \
179 .boot_openflags = OPEN_FLAGS, \
180 .openflags = OPEN_FLAGS, \
181 .no_cow = 0, \
182 .shared = 0, \
183 .cow = DEFAULT_COW, \
184 .lock = SPIN_LOCK_UNLOCKED, \
185 .request = NULL, \
186 .start_sg = 0, \
187 .end_sg = 0, \
188 }
189
190 /* Protected by ubd_lock */
191 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
192
193 /* Only changed by fake_ide_setup which is a setup */
194 static int fake_ide = 0;
195 static struct proc_dir_entry *proc_ide_root = NULL;
196 static struct proc_dir_entry *proc_ide = NULL;
197
198 static void make_proc_ide(void)
199 {
200 proc_ide_root = proc_mkdir("ide", NULL);
201 proc_ide = proc_mkdir("ide0", proc_ide_root);
202 }
203
204 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
205 int *eof, void *data)
206 {
207 int len;
208
209 strcpy(page, "disk\n");
210 len = strlen("disk\n");
211 len -= off;
212 if (len < count){
213 *eof = 1;
214 if (len <= 0) return 0;
215 }
216 else len = count;
217 *start = page + off;
218 return len;
219 }
220
221 static void make_ide_entries(const char *dev_name)
222 {
223 struct proc_dir_entry *dir, *ent;
224 char name[64];
225
226 if(proc_ide_root == NULL) make_proc_ide();
227
228 dir = proc_mkdir(dev_name, proc_ide);
229 if(!dir) return;
230
231 ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
232 if(!ent) return;
233 ent->data = NULL;
234 ent->read_proc = proc_ide_read_media;
235 ent->write_proc = NULL;
236 snprintf(name, sizeof(name), "ide0/%s", dev_name);
237 proc_symlink(dev_name, proc_ide_root, name);
238 }
239
240 static int fake_ide_setup(char *str)
241 {
242 fake_ide = 1;
243 return 1;
244 }
245
246 __setup("fake_ide", fake_ide_setup);
247
248 __uml_help(fake_ide_setup,
249 "fake_ide\n"
250 " Create ide0 entries that map onto ubd devices.\n\n"
251 );
252
253 static int parse_unit(char **ptr)
254 {
255 char *str = *ptr, *end;
256 int n = -1;
257
258 if(isdigit(*str)) {
259 n = simple_strtoul(str, &end, 0);
260 if(end == str)
261 return -1;
262 *ptr = end;
263 }
264 else if (('a' <= *str) && (*str <= 'z')) {
265 n = *str - 'a';
266 str++;
267 *ptr = str;
268 }
269 return n;
270 }
271
272 /* If *index_out == -1 at exit, the passed option was a general one;
273 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
274 * should not be freed on exit.
275 */
276 static int ubd_setup_common(char *str, int *index_out, char **error_out)
277 {
278 struct ubd *ubd_dev;
279 struct openflags flags = global_openflags;
280 char *backing_file;
281 int n, err = 0, i;
282
283 if(index_out) *index_out = -1;
284 n = *str;
285 if(n == '='){
286 char *end;
287 int major;
288
289 str++;
290 if(!strcmp(str, "sync")){
291 global_openflags = of_sync(global_openflags);
292 goto out1;
293 }
294
295 err = -EINVAL;
296 major = simple_strtoul(str, &end, 0);
297 if((*end != '\0') || (end == str)){
298 *error_out = "Didn't parse major number";
299 goto out1;
300 }
301
302 mutex_lock(&ubd_lock);
303 if(fake_major != MAJOR_NR){
304 *error_out = "Can't assign a fake major twice";
305 goto out1;
306 }
307
308 fake_major = major;
309
310 printk(KERN_INFO "Setting extra ubd major number to %d\n",
311 major);
312 err = 0;
313 out1:
314 mutex_unlock(&ubd_lock);
315 return err;
316 }
317
318 n = parse_unit(&str);
319 if(n < 0){
320 *error_out = "Couldn't parse device number";
321 return -EINVAL;
322 }
323 if(n >= MAX_DEV){
324 *error_out = "Device number out of range";
325 return 1;
326 }
327
328 err = -EBUSY;
329 mutex_lock(&ubd_lock);
330
331 ubd_dev = &ubd_devs[n];
332 if(ubd_dev->file != NULL){
333 *error_out = "Device is already configured";
334 goto out;
335 }
336
337 if (index_out)
338 *index_out = n;
339
340 err = -EINVAL;
341 for (i = 0; i < sizeof("rscd="); i++) {
342 switch (*str) {
343 case 'r':
344 flags.w = 0;
345 break;
346 case 's':
347 flags.s = 1;
348 break;
349 case 'd':
350 ubd_dev->no_cow = 1;
351 break;
352 case 'c':
353 ubd_dev->shared = 1;
354 break;
355 case '=':
356 str++;
357 goto break_loop;
358 default:
359 *error_out = "Expected '=' or flag letter "
360 "(r, s, c, or d)";
361 goto out;
362 }
363 str++;
364 }
365
366 if (*str == '=')
367 *error_out = "Too many flags specified";
368 else
369 *error_out = "Missing '='";
370 goto out;
371
372 break_loop:
373 backing_file = strchr(str, ',');
374
375 if (backing_file == NULL)
376 backing_file = strchr(str, ':');
377
378 if(backing_file != NULL){
379 if(ubd_dev->no_cow){
380 *error_out = "Can't specify both 'd' and a cow file";
381 goto out;
382 }
383 else {
384 *backing_file = '\0';
385 backing_file++;
386 }
387 }
388 err = 0;
389 ubd_dev->file = str;
390 ubd_dev->cow.file = backing_file;
391 ubd_dev->boot_openflags = flags;
392 out:
393 mutex_unlock(&ubd_lock);
394 return err;
395 }
396
397 static int ubd_setup(char *str)
398 {
399 char *error;
400 int err;
401
402 err = ubd_setup_common(str, NULL, &error);
403 if(err)
404 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
405 "%s\n", str, error);
406 return 1;
407 }
408
409 __setup("ubd", ubd_setup);
410 __uml_help(ubd_setup,
411 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
412 " This is used to associate a device with a file in the underlying\n"
413 " filesystem. When specifying two filenames, the first one is the\n"
414 " COW name and the second is the backing file name. As separator you can\n"
415 " use either a ':' or a ',': the first one allows writing things like;\n"
416 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
417 " while with a ',' the shell would not expand the 2nd '~'.\n"
418 " When using only one filename, UML will detect whether to treat it like\n"
419 " a COW file or a backing file. To override this detection, add the 'd'\n"
420 " flag:\n"
421 " ubd0d=BackingFile\n"
422 " Usually, there is a filesystem in the file, but \n"
423 " that's not required. Swap devices containing swap files can be\n"
424 " specified like this. Also, a file which doesn't contain a\n"
425 " filesystem can have its contents read in the virtual \n"
426 " machine by running 'dd' on the device. <n> must be in the range\n"
427 " 0 to 7. Appending an 'r' to the number will cause that device\n"
428 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
429 " an 's' will cause data to be written to disk on the host immediately.\n"
430 " 'c' will cause the device to be treated as being shared between multiple\n"
431 " UMLs and file locking will be turned off - this is appropriate for a\n"
432 " cluster filesystem and inappropriate at almost all other times.\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 void do_ubd_request(struct request_queue * q);
452
453 /* Only changed by ubd_init, which is an initcall. */
454 static int thread_fd = -1;
455
456 static void ubd_end_request(struct request *req, int bytes, int error)
457 {
458 blk_end_request(req, error, bytes);
459 }
460
461 /* Callable only from interrupt context - otherwise you need to do
462 * spin_lock_irq()/spin_lock_irqsave() */
463 static inline void ubd_finish(struct request *req, int bytes)
464 {
465 if(bytes < 0){
466 ubd_end_request(req, 0, -EIO);
467 return;
468 }
469 ubd_end_request(req, bytes, 0);
470 }
471
472 static LIST_HEAD(restart);
473
474 /* XXX - move this inside ubd_intr. */
475 /* Called without dev->lock held, and only in interrupt context. */
476 static void ubd_handler(void)
477 {
478 struct io_thread_req *req;
479 struct request *rq;
480 struct ubd *ubd;
481 struct list_head *list, *next_ele;
482 unsigned long flags;
483 int n;
484
485 while(1){
486 n = os_read_file(thread_fd, &req,
487 sizeof(struct io_thread_req *));
488 if(n != sizeof(req)){
489 if(n == -EAGAIN)
490 break;
491 printk(KERN_ERR "spurious interrupt in ubd_handler, "
492 "err = %d\n", -n);
493 return;
494 }
495
496 rq = req->req;
497 rq->nr_sectors -= req->length >> 9;
498 if(rq->nr_sectors == 0)
499 ubd_finish(rq, rq->hard_nr_sectors << 9);
500 kfree(req);
501 }
502 reactivate_fd(thread_fd, UBD_IRQ);
503
504 list_for_each_safe(list, next_ele, &restart){
505 ubd = container_of(list, struct ubd, restart);
506 list_del_init(&ubd->restart);
507 spin_lock_irqsave(&ubd->lock, flags);
508 do_ubd_request(ubd->queue);
509 spin_unlock_irqrestore(&ubd->lock, flags);
510 }
511 }
512
513 static irqreturn_t ubd_intr(int irq, void *dev)
514 {
515 ubd_handler();
516 return IRQ_HANDLED;
517 }
518
519 /* Only changed by ubd_init, which is an initcall. */
520 static int io_pid = -1;
521
522 static void kill_io_thread(void)
523 {
524 if(io_pid != -1)
525 os_kill_process(io_pid, 1);
526 }
527
528 __uml_exitcall(kill_io_thread);
529
530 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
531 {
532 char *file;
533
534 file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
535 return os_file_size(file, size_out);
536 }
537
538 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
539 {
540 int err;
541
542 err = os_seek_file(fd, offset);
543 if (err < 0)
544 return err;
545
546 err = os_read_file(fd, buf, len);
547 if (err < 0)
548 return err;
549
550 return 0;
551 }
552
553 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
554 {
555 unsigned long modtime;
556 unsigned long long actual;
557 int err;
558
559 err = os_file_modtime(file, &modtime);
560 if (err < 0) {
561 printk(KERN_ERR "Failed to get modification time of backing "
562 "file \"%s\", err = %d\n", file, -err);
563 return err;
564 }
565
566 err = os_file_size(file, &actual);
567 if (err < 0) {
568 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
569 "err = %d\n", file, -err);
570 return err;
571 }
572
573 if (actual != size) {
574 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
575 * the typecast.*/
576 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
577 "vs backing file\n", (unsigned long long) size, actual);
578 return -EINVAL;
579 }
580 if (modtime != mtime) {
581 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
582 "backing file\n", mtime, modtime);
583 return -EINVAL;
584 }
585 return 0;
586 }
587
588 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
589 {
590 struct uml_stat buf1, buf2;
591 int err;
592
593 if (from_cmdline == NULL)
594 return 0;
595 if (!strcmp(from_cmdline, from_cow))
596 return 0;
597
598 err = os_stat_file(from_cmdline, &buf1);
599 if (err < 0) {
600 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
601 -err);
602 return 0;
603 }
604 err = os_stat_file(from_cow, &buf2);
605 if (err < 0) {
606 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
607 -err);
608 return 1;
609 }
610 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
611 return 0;
612
613 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
614 "\"%s\" specified in COW header of \"%s\"\n",
615 from_cmdline, from_cow, cow);
616 return 1;
617 }
618
619 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
620 char **backing_file_out, int *bitmap_offset_out,
621 unsigned long *bitmap_len_out, int *data_offset_out,
622 int *create_cow_out)
623 {
624 time_t mtime;
625 unsigned long long size;
626 __u32 version, align;
627 char *backing_file;
628 int fd, err, sectorsize, asked_switch, mode = 0644;
629
630 fd = os_open_file(file, *openflags, mode);
631 if (fd < 0) {
632 if ((fd == -ENOENT) && (create_cow_out != NULL))
633 *create_cow_out = 1;
634 if (!openflags->w ||
635 ((fd != -EROFS) && (fd != -EACCES)))
636 return fd;
637 openflags->w = 0;
638 fd = os_open_file(file, *openflags, mode);
639 if (fd < 0)
640 return fd;
641 }
642
643 if (shared)
644 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
645 else {
646 err = os_lock_file(fd, openflags->w);
647 if (err < 0) {
648 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
649 file, -err);
650 goto out_close;
651 }
652 }
653
654 /* Successful return case! */
655 if (backing_file_out == NULL)
656 return fd;
657
658 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
659 &size, &sectorsize, &align, bitmap_offset_out);
660 if (err && (*backing_file_out != NULL)) {
661 printk(KERN_ERR "Failed to read COW header from COW file "
662 "\"%s\", errno = %d\n", file, -err);
663 goto out_close;
664 }
665 if (err)
666 return fd;
667
668 asked_switch = path_requires_switch(*backing_file_out, backing_file,
669 file);
670
671 /* Allow switching only if no mismatch. */
672 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
673 mtime)) {
674 printk(KERN_ERR "Switching backing file to '%s'\n",
675 *backing_file_out);
676 err = write_cow_header(file, fd, *backing_file_out,
677 sectorsize, align, &size);
678 if (err) {
679 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
680 goto out_close;
681 }
682 } else {
683 *backing_file_out = backing_file;
684 err = backing_file_mismatch(*backing_file_out, size, mtime);
685 if (err)
686 goto out_close;
687 }
688
689 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
690 bitmap_len_out, data_offset_out);
691
692 return fd;
693 out_close:
694 os_close_file(fd);
695 return err;
696 }
697
698 static int create_cow_file(char *cow_file, char *backing_file,
699 struct openflags flags,
700 int sectorsize, int alignment, int *bitmap_offset_out,
701 unsigned long *bitmap_len_out, int *data_offset_out)
702 {
703 int err, fd;
704
705 flags.c = 1;
706 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
707 if (fd < 0) {
708 err = fd;
709 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
710 cow_file, -err);
711 goto out;
712 }
713
714 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
715 bitmap_offset_out, bitmap_len_out,
716 data_offset_out);
717 if (!err)
718 return fd;
719 os_close_file(fd);
720 out:
721 return err;
722 }
723
724 static void ubd_close_dev(struct ubd *ubd_dev)
725 {
726 os_close_file(ubd_dev->fd);
727 if(ubd_dev->cow.file == NULL)
728 return;
729
730 os_close_file(ubd_dev->cow.fd);
731 vfree(ubd_dev->cow.bitmap);
732 ubd_dev->cow.bitmap = NULL;
733 }
734
735 static int ubd_open_dev(struct ubd *ubd_dev)
736 {
737 struct openflags flags;
738 char **back_ptr;
739 int err, create_cow, *create_ptr;
740 int fd;
741
742 ubd_dev->openflags = ubd_dev->boot_openflags;
743 create_cow = 0;
744 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
745 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
746
747 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
748 back_ptr, &ubd_dev->cow.bitmap_offset,
749 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
750 create_ptr);
751
752 if((fd == -ENOENT) && create_cow){
753 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
754 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
755 &ubd_dev->cow.bitmap_offset,
756 &ubd_dev->cow.bitmap_len,
757 &ubd_dev->cow.data_offset);
758 if(fd >= 0){
759 printk(KERN_INFO "Creating \"%s\" as COW file for "
760 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
761 }
762 }
763
764 if(fd < 0){
765 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
766 -fd);
767 return fd;
768 }
769 ubd_dev->fd = fd;
770
771 if(ubd_dev->cow.file != NULL){
772 blk_queue_max_sectors(ubd_dev->queue, 8 * sizeof(long));
773
774 err = -ENOMEM;
775 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
776 if(ubd_dev->cow.bitmap == NULL){
777 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
778 goto error;
779 }
780 flush_tlb_kernel_vm();
781
782 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
783 ubd_dev->cow.bitmap_offset,
784 ubd_dev->cow.bitmap_len);
785 if(err < 0)
786 goto error;
787
788 flags = ubd_dev->openflags;
789 flags.w = 0;
790 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
791 NULL, NULL, NULL, NULL);
792 if(err < 0) goto error;
793 ubd_dev->cow.fd = err;
794 }
795 return 0;
796 error:
797 os_close_file(ubd_dev->fd);
798 return err;
799 }
800
801 static void ubd_device_release(struct device *dev)
802 {
803 struct ubd *ubd_dev = dev->driver_data;
804
805 blk_cleanup_queue(ubd_dev->queue);
806 *ubd_dev = ((struct ubd) DEFAULT_UBD);
807 }
808
809 static int ubd_disk_register(int major, u64 size, int unit,
810 struct gendisk **disk_out)
811 {
812 struct gendisk *disk;
813
814 disk = alloc_disk(1 << UBD_SHIFT);
815 if(disk == NULL)
816 return -ENOMEM;
817
818 disk->major = major;
819 disk->first_minor = unit << UBD_SHIFT;
820 disk->fops = &ubd_blops;
821 set_capacity(disk, size / 512);
822 if(major == MAJOR_NR)
823 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
824 else
825 sprintf(disk->disk_name, "ubd_fake%d", unit);
826
827 /* sysfs register (not for ide fake devices) */
828 if (major == MAJOR_NR) {
829 ubd_devs[unit].pdev.id = unit;
830 ubd_devs[unit].pdev.name = DRIVER_NAME;
831 ubd_devs[unit].pdev.dev.release = ubd_device_release;
832 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
833 platform_device_register(&ubd_devs[unit].pdev);
834 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
835 }
836
837 disk->private_data = &ubd_devs[unit];
838 disk->queue = ubd_devs[unit].queue;
839 add_disk(disk);
840
841 *disk_out = disk;
842 return 0;
843 }
844
845 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
846
847 static int ubd_add(int n, char **error_out)
848 {
849 struct ubd *ubd_dev = &ubd_devs[n];
850 int err = 0;
851
852 if(ubd_dev->file == NULL)
853 goto out;
854
855 err = ubd_file_size(ubd_dev, &ubd_dev->size);
856 if(err < 0){
857 *error_out = "Couldn't determine size of device's file";
858 goto out;
859 }
860
861 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
862
863 INIT_LIST_HEAD(&ubd_dev->restart);
864 sg_init_table(ubd_dev->sg, MAX_SG);
865
866 err = -ENOMEM;
867 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
868 if (ubd_dev->queue == NULL) {
869 *error_out = "Failed to initialize device queue";
870 goto out;
871 }
872 ubd_dev->queue->queuedata = ubd_dev;
873
874 blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG);
875 err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
876 if(err){
877 *error_out = "Failed to register device";
878 goto out_cleanup;
879 }
880
881 if(fake_major != MAJOR_NR)
882 ubd_disk_register(fake_major, ubd_dev->size, n,
883 &fake_gendisk[n]);
884
885 /*
886 * Perhaps this should also be under the "if (fake_major)" above
887 * using the fake_disk->disk_name
888 */
889 if (fake_ide)
890 make_ide_entries(ubd_gendisk[n]->disk_name);
891
892 err = 0;
893 out:
894 return err;
895
896 out_cleanup:
897 blk_cleanup_queue(ubd_dev->queue);
898 goto out;
899 }
900
901 static int ubd_config(char *str, char **error_out)
902 {
903 int n, ret;
904
905 /* This string is possibly broken up and stored, so it's only
906 * freed if ubd_setup_common fails, or if only general options
907 * were set.
908 */
909 str = kstrdup(str, GFP_KERNEL);
910 if (str == NULL) {
911 *error_out = "Failed to allocate memory";
912 return -ENOMEM;
913 }
914
915 ret = ubd_setup_common(str, &n, error_out);
916 if (ret)
917 goto err_free;
918
919 if (n == -1) {
920 ret = 0;
921 goto err_free;
922 }
923
924 mutex_lock(&ubd_lock);
925 ret = ubd_add(n, error_out);
926 if (ret)
927 ubd_devs[n].file = NULL;
928 mutex_unlock(&ubd_lock);
929
930 out:
931 return ret;
932
933 err_free:
934 kfree(str);
935 goto out;
936 }
937
938 static int ubd_get_config(char *name, char *str, int size, char **error_out)
939 {
940 struct ubd *ubd_dev;
941 int n, len = 0;
942
943 n = parse_unit(&name);
944 if((n >= MAX_DEV) || (n < 0)){
945 *error_out = "ubd_get_config : device number out of range";
946 return -1;
947 }
948
949 ubd_dev = &ubd_devs[n];
950 mutex_lock(&ubd_lock);
951
952 if(ubd_dev->file == NULL){
953 CONFIG_CHUNK(str, size, len, "", 1);
954 goto out;
955 }
956
957 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
958
959 if(ubd_dev->cow.file != NULL){
960 CONFIG_CHUNK(str, size, len, ",", 0);
961 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
962 }
963 else CONFIG_CHUNK(str, size, len, "", 1);
964
965 out:
966 mutex_unlock(&ubd_lock);
967 return len;
968 }
969
970 static int ubd_id(char **str, int *start_out, int *end_out)
971 {
972 int n;
973
974 n = parse_unit(str);
975 *start_out = 0;
976 *end_out = MAX_DEV - 1;
977 return n;
978 }
979
980 static int ubd_remove(int n, char **error_out)
981 {
982 struct gendisk *disk = ubd_gendisk[n];
983 struct ubd *ubd_dev;
984 int err = -ENODEV;
985
986 mutex_lock(&ubd_lock);
987
988 ubd_dev = &ubd_devs[n];
989
990 if(ubd_dev->file == NULL)
991 goto out;
992
993 /* you cannot remove a open disk */
994 err = -EBUSY;
995 if(ubd_dev->count > 0)
996 goto out;
997
998 ubd_gendisk[n] = NULL;
999 if(disk != NULL){
1000 del_gendisk(disk);
1001 put_disk(disk);
1002 }
1003
1004 if(fake_gendisk[n] != NULL){
1005 del_gendisk(fake_gendisk[n]);
1006 put_disk(fake_gendisk[n]);
1007 fake_gendisk[n] = NULL;
1008 }
1009
1010 err = 0;
1011 platform_device_unregister(&ubd_dev->pdev);
1012 out:
1013 mutex_unlock(&ubd_lock);
1014 return err;
1015 }
1016
1017 /* All these are called by mconsole in process context and without
1018 * ubd-specific locks. The structure itself is const except for .list.
1019 */
1020 static struct mc_device ubd_mc = {
1021 .list = LIST_HEAD_INIT(ubd_mc.list),
1022 .name = "ubd",
1023 .config = ubd_config,
1024 .get_config = ubd_get_config,
1025 .id = ubd_id,
1026 .remove = ubd_remove,
1027 };
1028
1029 static int __init ubd_mc_init(void)
1030 {
1031 mconsole_register_dev(&ubd_mc);
1032 return 0;
1033 }
1034
1035 __initcall(ubd_mc_init);
1036
1037 static int __init ubd0_init(void)
1038 {
1039 struct ubd *ubd_dev = &ubd_devs[0];
1040
1041 mutex_lock(&ubd_lock);
1042 if(ubd_dev->file == NULL)
1043 ubd_dev->file = "root_fs";
1044 mutex_unlock(&ubd_lock);
1045
1046 return 0;
1047 }
1048
1049 __initcall(ubd0_init);
1050
1051 /* Used in ubd_init, which is an initcall */
1052 static struct platform_driver ubd_driver = {
1053 .driver = {
1054 .name = DRIVER_NAME,
1055 },
1056 };
1057
1058 static int __init ubd_init(void)
1059 {
1060 char *error;
1061 int i, err;
1062
1063 if (register_blkdev(MAJOR_NR, "ubd"))
1064 return -1;
1065
1066 if (fake_major != MAJOR_NR) {
1067 char name[sizeof("ubd_nnn\0")];
1068
1069 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1070 if (register_blkdev(fake_major, "ubd"))
1071 return -1;
1072 }
1073 platform_driver_register(&ubd_driver);
1074 mutex_lock(&ubd_lock);
1075 for (i = 0; i < MAX_DEV; i++){
1076 err = ubd_add(i, &error);
1077 if(err)
1078 printk(KERN_ERR "Failed to initialize ubd device %d :"
1079 "%s\n", i, error);
1080 }
1081 mutex_unlock(&ubd_lock);
1082 return 0;
1083 }
1084
1085 late_initcall(ubd_init);
1086
1087 static int __init ubd_driver_init(void){
1088 unsigned long stack;
1089 int err;
1090
1091 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1092 if(global_openflags.s){
1093 printk(KERN_INFO "ubd: Synchronous mode\n");
1094 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1095 * enough. So use anyway the io thread. */
1096 }
1097 stack = alloc_stack(0, 0);
1098 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1099 &thread_fd);
1100 if(io_pid < 0){
1101 printk(KERN_ERR
1102 "ubd : Failed to start I/O thread (errno = %d) - "
1103 "falling back to synchronous I/O\n", -io_pid);
1104 io_pid = -1;
1105 return 0;
1106 }
1107 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1108 IRQF_DISABLED, "ubd", ubd_devs);
1109 if(err != 0)
1110 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1111 return 0;
1112 }
1113
1114 device_initcall(ubd_driver_init);
1115
1116 static int ubd_open(struct inode *inode, struct file *filp)
1117 {
1118 struct gendisk *disk = inode->i_bdev->bd_disk;
1119 struct ubd *ubd_dev = disk->private_data;
1120 int err = 0;
1121
1122 if(ubd_dev->count == 0){
1123 err = ubd_open_dev(ubd_dev);
1124 if(err){
1125 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1126 disk->disk_name, ubd_dev->file, -err);
1127 goto out;
1128 }
1129 }
1130 ubd_dev->count++;
1131 set_disk_ro(disk, !ubd_dev->openflags.w);
1132
1133 /* This should no more be needed. And it didn't work anyway to exclude
1134 * read-write remounting of filesystems.*/
1135 /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1136 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1137 err = -EROFS;
1138 }*/
1139 out:
1140 return err;
1141 }
1142
1143 static int ubd_release(struct inode * inode, struct file * file)
1144 {
1145 struct gendisk *disk = inode->i_bdev->bd_disk;
1146 struct ubd *ubd_dev = disk->private_data;
1147
1148 if(--ubd_dev->count == 0)
1149 ubd_close_dev(ubd_dev);
1150 return 0;
1151 }
1152
1153 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1154 __u64 *cow_offset, unsigned long *bitmap,
1155 __u64 bitmap_offset, unsigned long *bitmap_words,
1156 __u64 bitmap_len)
1157 {
1158 __u64 sector = io_offset >> 9;
1159 int i, update_bitmap = 0;
1160
1161 for(i = 0; i < length >> 9; i++){
1162 if(cow_mask != NULL)
1163 ubd_set_bit(i, (unsigned char *) cow_mask);
1164 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1165 continue;
1166
1167 update_bitmap = 1;
1168 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1169 }
1170
1171 if(!update_bitmap)
1172 return;
1173
1174 *cow_offset = sector / (sizeof(unsigned long) * 8);
1175
1176 /* This takes care of the case where we're exactly at the end of the
1177 * device, and *cow_offset + 1 is off the end. So, just back it up
1178 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1179 * for the original diagnosis.
1180 */
1181 if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
1182 sizeof(unsigned long) - 1))
1183 (*cow_offset)--;
1184
1185 bitmap_words[0] = bitmap[*cow_offset];
1186 bitmap_words[1] = bitmap[*cow_offset + 1];
1187
1188 *cow_offset *= sizeof(unsigned long);
1189 *cow_offset += bitmap_offset;
1190 }
1191
1192 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1193 __u64 bitmap_offset, __u64 bitmap_len)
1194 {
1195 __u64 sector = req->offset >> 9;
1196 int i;
1197
1198 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1199 panic("Operation too long");
1200
1201 if(req->op == UBD_READ) {
1202 for(i = 0; i < req->length >> 9; i++){
1203 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1204 ubd_set_bit(i, (unsigned char *)
1205 &req->sector_mask);
1206 }
1207 }
1208 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1209 &req->cow_offset, bitmap, bitmap_offset,
1210 req->bitmap_words, bitmap_len);
1211 }
1212
1213 /* Called with dev->lock held */
1214 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1215 unsigned long long offset, int page_offset,
1216 int len, struct page *page)
1217 {
1218 struct gendisk *disk = req->rq_disk;
1219 struct ubd *ubd_dev = disk->private_data;
1220
1221 io_req->req = req;
1222 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1223 ubd_dev->fd;
1224 io_req->fds[1] = ubd_dev->fd;
1225 io_req->cow_offset = -1;
1226 io_req->offset = offset;
1227 io_req->length = len;
1228 io_req->error = 0;
1229 io_req->sector_mask = 0;
1230
1231 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1232 io_req->offsets[0] = 0;
1233 io_req->offsets[1] = ubd_dev->cow.data_offset;
1234 io_req->buffer = page_address(page) + page_offset;
1235 io_req->sectorsize = 1 << 9;
1236
1237 if(ubd_dev->cow.file != NULL)
1238 cowify_req(io_req, ubd_dev->cow.bitmap,
1239 ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1240
1241 }
1242
1243 /* Called with dev->lock held */
1244 static void do_ubd_request(struct request_queue *q)
1245 {
1246 struct io_thread_req *io_req;
1247 struct request *req;
1248 int n, last_sectors;
1249
1250 while(1){
1251 struct ubd *dev = q->queuedata;
1252 if(dev->end_sg == 0){
1253 struct request *req = elv_next_request(q);
1254 if(req == NULL)
1255 return;
1256
1257 dev->request = req;
1258 blkdev_dequeue_request(req);
1259 dev->start_sg = 0;
1260 dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1261 }
1262
1263 req = dev->request;
1264 last_sectors = 0;
1265 while(dev->start_sg < dev->end_sg){
1266 struct scatterlist *sg = &dev->sg[dev->start_sg];
1267
1268 req->sector += last_sectors;
1269 io_req = kmalloc(sizeof(struct io_thread_req),
1270 GFP_ATOMIC);
1271 if(io_req == NULL){
1272 if(list_empty(&dev->restart))
1273 list_add(&dev->restart, &restart);
1274 return;
1275 }
1276 prepare_request(req, io_req,
1277 (unsigned long long) req->sector << 9,
1278 sg->offset, sg->length, sg_page(sg));
1279
1280 last_sectors = sg->length >> 9;
1281 n = os_write_file(thread_fd, &io_req,
1282 sizeof(struct io_thread_req *));
1283 if(n != sizeof(struct io_thread_req *)){
1284 if(n != -EAGAIN)
1285 printk("write to io thread failed, "
1286 "errno = %d\n", -n);
1287 else if(list_empty(&dev->restart))
1288 list_add(&dev->restart, &restart);
1289 kfree(io_req);
1290 return;
1291 }
1292
1293 dev->start_sg++;
1294 }
1295 dev->end_sg = 0;
1296 dev->request = NULL;
1297 }
1298 }
1299
1300 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1301 {
1302 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1303
1304 geo->heads = 128;
1305 geo->sectors = 32;
1306 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1307 return 0;
1308 }
1309
1310 static int ubd_ioctl(struct inode * inode, struct file * file,
1311 unsigned int cmd, unsigned long arg)
1312 {
1313 struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1314 struct hd_driveid ubd_id = {
1315 .cyls = 0,
1316 .heads = 128,
1317 .sectors = 32,
1318 };
1319
1320 switch (cmd) {
1321 struct cdrom_volctrl volume;
1322 case HDIO_GET_IDENTITY:
1323 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1324 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1325 sizeof(ubd_id)))
1326 return -EFAULT;
1327 return 0;
1328
1329 case CDROMVOLREAD:
1330 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1331 return -EFAULT;
1332 volume.channel0 = 255;
1333 volume.channel1 = 255;
1334 volume.channel2 = 255;
1335 volume.channel3 = 255;
1336 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1337 return -EFAULT;
1338 return 0;
1339 }
1340 return -EINVAL;
1341 }
1342
1343 static int update_bitmap(struct io_thread_req *req)
1344 {
1345 int n;
1346
1347 if(req->cow_offset == -1)
1348 return 0;
1349
1350 n = os_seek_file(req->fds[1], req->cow_offset);
1351 if(n < 0){
1352 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1353 return 1;
1354 }
1355
1356 n = os_write_file(req->fds[1], &req->bitmap_words,
1357 sizeof(req->bitmap_words));
1358 if(n != sizeof(req->bitmap_words)){
1359 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1360 req->fds[1]);
1361 return 1;
1362 }
1363
1364 return 0;
1365 }
1366
1367 static void do_io(struct io_thread_req *req)
1368 {
1369 char *buf;
1370 unsigned long len;
1371 int n, nsectors, start, end, bit;
1372 int err;
1373 __u64 off;
1374
1375 nsectors = req->length / req->sectorsize;
1376 start = 0;
1377 do {
1378 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1379 end = start;
1380 while((end < nsectors) &&
1381 (ubd_test_bit(end, (unsigned char *)
1382 &req->sector_mask) == bit))
1383 end++;
1384
1385 off = req->offset + req->offsets[bit] +
1386 start * req->sectorsize;
1387 len = (end - start) * req->sectorsize;
1388 buf = &req->buffer[start * req->sectorsize];
1389
1390 err = os_seek_file(req->fds[bit], off);
1391 if(err < 0){
1392 printk("do_io - lseek failed : err = %d\n", -err);
1393 req->error = 1;
1394 return;
1395 }
1396 if(req->op == UBD_READ){
1397 n = 0;
1398 do {
1399 buf = &buf[n];
1400 len -= n;
1401 n = os_read_file(req->fds[bit], buf, len);
1402 if (n < 0) {
1403 printk("do_io - read failed, err = %d "
1404 "fd = %d\n", -n, req->fds[bit]);
1405 req->error = 1;
1406 return;
1407 }
1408 } while((n < len) && (n != 0));
1409 if (n < len) memset(&buf[n], 0, len - n);
1410 } else {
1411 n = os_write_file(req->fds[bit], buf, len);
1412 if(n != len){
1413 printk("do_io - write failed err = %d "
1414 "fd = %d\n", -n, req->fds[bit]);
1415 req->error = 1;
1416 return;
1417 }
1418 }
1419
1420 start = end;
1421 } while(start < nsectors);
1422
1423 req->error = update_bitmap(req);
1424 }
1425
1426 /* Changed in start_io_thread, which is serialized by being called only
1427 * from ubd_init, which is an initcall.
1428 */
1429 int kernel_fd = -1;
1430
1431 /* Only changed by the io thread. XXX: currently unused. */
1432 static int io_count = 0;
1433
1434 int io_thread(void *arg)
1435 {
1436 struct io_thread_req *req;
1437 int n;
1438
1439 ignore_sigwinch_sig();
1440 while(1){
1441 n = os_read_file(kernel_fd, &req,
1442 sizeof(struct io_thread_req *));
1443 if(n != sizeof(struct io_thread_req *)){
1444 if(n < 0)
1445 printk("io_thread - read failed, fd = %d, "
1446 "err = %d\n", kernel_fd, -n);
1447 else {
1448 printk("io_thread - short read, fd = %d, "
1449 "length = %d\n", kernel_fd, n);
1450 }
1451 continue;
1452 }
1453 io_count++;
1454 do_io(req);
1455 n = os_write_file(kernel_fd, &req,
1456 sizeof(struct io_thread_req *));
1457 if(n != sizeof(struct io_thread_req *))
1458 printk("io_thread - write failed, fd = %d, err = %d\n",
1459 kernel_fd, -n);
1460 }
1461
1462 return 0;
1463 }
This page took 0.060829 seconds and 6 git commands to generate.