Loading...
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 UBD_SHIFT 4
21
22#include <linux/module.h>
23#include <linux/init.h>
24#include <linux/blkdev.h>
25#include <linux/ata.h>
26#include <linux/hdreg.h>
27#include <linux/cdrom.h>
28#include <linux/proc_fs.h>
29#include <linux/seq_file.h>
30#include <linux/ctype.h>
31#include <linux/slab.h>
32#include <linux/vmalloc.h>
33#include <linux/platform_device.h>
34#include <linux/scatterlist.h>
35#include <asm/tlbflush.h>
36#include "kern_util.h"
37#include "mconsole_kern.h"
38#include "init.h"
39#include "irq_kern.h"
40#include "ubd.h"
41#include "os.h"
42#include "cow.h"
43
44enum ubd_req { UBD_READ, UBD_WRITE };
45
46struct io_thread_req {
47 struct request *req;
48 enum ubd_req op;
49 int fds[2];
50 unsigned long offsets[2];
51 unsigned long long offset;
52 unsigned long length;
53 char *buffer;
54 int sectorsize;
55 unsigned long sector_mask;
56 unsigned long long cow_offset;
57 unsigned long bitmap_words[2];
58 int error;
59};
60
61static inline int ubd_test_bit(__u64 bit, unsigned char *data)
62{
63 __u64 n;
64 int bits, off;
65
66 bits = sizeof(data[0]) * 8;
67 n = bit / bits;
68 off = bit % bits;
69 return (data[n] & (1 << off)) != 0;
70}
71
72static inline void ubd_set_bit(__u64 bit, unsigned char *data)
73{
74 __u64 n;
75 int bits, off;
76
77 bits = sizeof(data[0]) * 8;
78 n = bit / bits;
79 off = bit % bits;
80 data[n] |= (1 << off);
81}
82/*End stuff from ubd_user.h*/
83
84#define DRIVER_NAME "uml-blkdev"
85
86static DEFINE_MUTEX(ubd_lock);
87static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
88
89static int ubd_open(struct block_device *bdev, fmode_t mode);
90static int ubd_release(struct gendisk *disk, fmode_t mode);
91static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
92 unsigned int cmd, unsigned long arg);
93static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
94
95#define MAX_DEV (16)
96
97static const struct block_device_operations ubd_blops = {
98 .owner = THIS_MODULE,
99 .open = ubd_open,
100 .release = ubd_release,
101 .ioctl = ubd_ioctl,
102 .getgeo = ubd_getgeo,
103};
104
105/* Protected by ubd_lock */
106static int fake_major = UBD_MAJOR;
107static struct gendisk *ubd_gendisk[MAX_DEV];
108static struct gendisk *fake_gendisk[MAX_DEV];
109
110#ifdef CONFIG_BLK_DEV_UBD_SYNC
111#define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
112 .cl = 1 })
113#else
114#define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
115 .cl = 1 })
116#endif
117static struct openflags global_openflags = OPEN_FLAGS;
118
119struct cow {
120 /* backing file name */
121 char *file;
122 /* backing file fd */
123 int fd;
124 unsigned long *bitmap;
125 unsigned long bitmap_len;
126 int bitmap_offset;
127 int data_offset;
128};
129
130#define MAX_SG 64
131
132struct ubd {
133 struct list_head restart;
134 /* name (and fd, below) of the file opened for writing, either the
135 * backing or the cow file. */
136 char *file;
137 int count;
138 int fd;
139 __u64 size;
140 struct openflags boot_openflags;
141 struct openflags openflags;
142 unsigned shared:1;
143 unsigned no_cow:1;
144 struct cow cow;
145 struct platform_device pdev;
146 struct request_queue *queue;
147 spinlock_t lock;
148 struct scatterlist sg[MAX_SG];
149 struct request *request;
150 int start_sg, end_sg;
151 sector_t rq_pos;
152};
153
154#define DEFAULT_COW { \
155 .file = NULL, \
156 .fd = -1, \
157 .bitmap = NULL, \
158 .bitmap_offset = 0, \
159 .data_offset = 0, \
160}
161
162#define DEFAULT_UBD { \
163 .file = NULL, \
164 .count = 0, \
165 .fd = -1, \
166 .size = -1, \
167 .boot_openflags = OPEN_FLAGS, \
168 .openflags = OPEN_FLAGS, \
169 .no_cow = 0, \
170 .shared = 0, \
171 .cow = DEFAULT_COW, \
172 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
173 .request = NULL, \
174 .start_sg = 0, \
175 .end_sg = 0, \
176 .rq_pos = 0, \
177}
178
179/* Protected by ubd_lock */
180static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
181
182/* Only changed by fake_ide_setup which is a setup */
183static int fake_ide = 0;
184static struct proc_dir_entry *proc_ide_root = NULL;
185static struct proc_dir_entry *proc_ide = NULL;
186
187static void make_proc_ide(void)
188{
189 proc_ide_root = proc_mkdir("ide", NULL);
190 proc_ide = proc_mkdir("ide0", proc_ide_root);
191}
192
193static int fake_ide_media_proc_show(struct seq_file *m, void *v)
194{
195 seq_puts(m, "disk\n");
196 return 0;
197}
198
199static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
200{
201 return single_open(file, fake_ide_media_proc_show, NULL);
202}
203
204static const struct file_operations fake_ide_media_proc_fops = {
205 .owner = THIS_MODULE,
206 .open = fake_ide_media_proc_open,
207 .read = seq_read,
208 .llseek = seq_lseek,
209 .release = single_release,
210};
211
212static void make_ide_entries(const char *dev_name)
213{
214 struct proc_dir_entry *dir, *ent;
215 char name[64];
216
217 if(proc_ide_root == NULL) make_proc_ide();
218
219 dir = proc_mkdir(dev_name, proc_ide);
220 if(!dir) return;
221
222 ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
223 if(!ent) return;
224 snprintf(name, sizeof(name), "ide0/%s", dev_name);
225 proc_symlink(dev_name, proc_ide_root, name);
226}
227
228static int fake_ide_setup(char *str)
229{
230 fake_ide = 1;
231 return 1;
232}
233
234__setup("fake_ide", fake_ide_setup);
235
236__uml_help(fake_ide_setup,
237"fake_ide\n"
238" Create ide0 entries that map onto ubd devices.\n\n"
239);
240
241static int parse_unit(char **ptr)
242{
243 char *str = *ptr, *end;
244 int n = -1;
245
246 if(isdigit(*str)) {
247 n = simple_strtoul(str, &end, 0);
248 if(end == str)
249 return -1;
250 *ptr = end;
251 }
252 else if (('a' <= *str) && (*str <= 'z')) {
253 n = *str - 'a';
254 str++;
255 *ptr = str;
256 }
257 return n;
258}
259
260/* If *index_out == -1 at exit, the passed option was a general one;
261 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
262 * should not be freed on exit.
263 */
264static int ubd_setup_common(char *str, int *index_out, char **error_out)
265{
266 struct ubd *ubd_dev;
267 struct openflags flags = global_openflags;
268 char *backing_file;
269 int n, err = 0, i;
270
271 if(index_out) *index_out = -1;
272 n = *str;
273 if(n == '='){
274 char *end;
275 int major;
276
277 str++;
278 if(!strcmp(str, "sync")){
279 global_openflags = of_sync(global_openflags);
280 goto out1;
281 }
282
283 err = -EINVAL;
284 major = simple_strtoul(str, &end, 0);
285 if((*end != '\0') || (end == str)){
286 *error_out = "Didn't parse major number";
287 goto out1;
288 }
289
290 mutex_lock(&ubd_lock);
291 if (fake_major != UBD_MAJOR) {
292 *error_out = "Can't assign a fake major twice";
293 goto out1;
294 }
295
296 fake_major = major;
297
298 printk(KERN_INFO "Setting extra ubd major number to %d\n",
299 major);
300 err = 0;
301 out1:
302 mutex_unlock(&ubd_lock);
303 return err;
304 }
305
306 n = parse_unit(&str);
307 if(n < 0){
308 *error_out = "Couldn't parse device number";
309 return -EINVAL;
310 }
311 if(n >= MAX_DEV){
312 *error_out = "Device number out of range";
313 return 1;
314 }
315
316 err = -EBUSY;
317 mutex_lock(&ubd_lock);
318
319 ubd_dev = &ubd_devs[n];
320 if(ubd_dev->file != NULL){
321 *error_out = "Device is already configured";
322 goto out;
323 }
324
325 if (index_out)
326 *index_out = n;
327
328 err = -EINVAL;
329 for (i = 0; i < sizeof("rscd="); i++) {
330 switch (*str) {
331 case 'r':
332 flags.w = 0;
333 break;
334 case 's':
335 flags.s = 1;
336 break;
337 case 'd':
338 ubd_dev->no_cow = 1;
339 break;
340 case 'c':
341 ubd_dev->shared = 1;
342 break;
343 case '=':
344 str++;
345 goto break_loop;
346 default:
347 *error_out = "Expected '=' or flag letter "
348 "(r, s, c, or d)";
349 goto out;
350 }
351 str++;
352 }
353
354 if (*str == '=')
355 *error_out = "Too many flags specified";
356 else
357 *error_out = "Missing '='";
358 goto out;
359
360break_loop:
361 backing_file = strchr(str, ',');
362
363 if (backing_file == NULL)
364 backing_file = strchr(str, ':');
365
366 if(backing_file != NULL){
367 if(ubd_dev->no_cow){
368 *error_out = "Can't specify both 'd' and a cow file";
369 goto out;
370 }
371 else {
372 *backing_file = '\0';
373 backing_file++;
374 }
375 }
376 err = 0;
377 ubd_dev->file = str;
378 ubd_dev->cow.file = backing_file;
379 ubd_dev->boot_openflags = flags;
380out:
381 mutex_unlock(&ubd_lock);
382 return err;
383}
384
385static int ubd_setup(char *str)
386{
387 char *error;
388 int err;
389
390 err = ubd_setup_common(str, NULL, &error);
391 if(err)
392 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
393 "%s\n", str, error);
394 return 1;
395}
396
397__setup("ubd", ubd_setup);
398__uml_help(ubd_setup,
399"ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
400" This is used to associate a device with a file in the underlying\n"
401" filesystem. When specifying two filenames, the first one is the\n"
402" COW name and the second is the backing file name. As separator you can\n"
403" use either a ':' or a ',': the first one allows writing things like;\n"
404" ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
405" while with a ',' the shell would not expand the 2nd '~'.\n"
406" When using only one filename, UML will detect whether to treat it like\n"
407" a COW file or a backing file. To override this detection, add the 'd'\n"
408" flag:\n"
409" ubd0d=BackingFile\n"
410" Usually, there is a filesystem in the file, but \n"
411" that's not required. Swap devices containing swap files can be\n"
412" specified like this. Also, a file which doesn't contain a\n"
413" filesystem can have its contents read in the virtual \n"
414" machine by running 'dd' on the device. <n> must be in the range\n"
415" 0 to 7. Appending an 'r' to the number will cause that device\n"
416" to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
417" an 's' will cause data to be written to disk on the host immediately.\n"
418" 'c' will cause the device to be treated as being shared between multiple\n"
419" UMLs and file locking will be turned off - this is appropriate for a\n"
420" cluster filesystem and inappropriate at almost all other times.\n\n"
421);
422
423static int udb_setup(char *str)
424{
425 printk("udb%s specified on command line is almost certainly a ubd -> "
426 "udb TYPO\n", str);
427 return 1;
428}
429
430__setup("udb", udb_setup);
431__uml_help(udb_setup,
432"udb\n"
433" This option is here solely to catch ubd -> udb typos, which can be\n"
434" to impossible to catch visually unless you specifically look for\n"
435" them. The only result of any option starting with 'udb' is an error\n"
436" in the boot output.\n\n"
437);
438
439static void do_ubd_request(struct request_queue * q);
440
441/* Only changed by ubd_init, which is an initcall. */
442static int thread_fd = -1;
443static LIST_HEAD(restart);
444
445/* XXX - move this inside ubd_intr. */
446/* Called without dev->lock held, and only in interrupt context. */
447static void ubd_handler(void)
448{
449 struct io_thread_req *req;
450 struct ubd *ubd;
451 struct list_head *list, *next_ele;
452 unsigned long flags;
453 int n;
454
455 while(1){
456 n = os_read_file(thread_fd, &req,
457 sizeof(struct io_thread_req *));
458 if(n != sizeof(req)){
459 if(n == -EAGAIN)
460 break;
461 printk(KERN_ERR "spurious interrupt in ubd_handler, "
462 "err = %d\n", -n);
463 return;
464 }
465
466 blk_end_request(req->req, 0, req->length);
467 kfree(req);
468 }
469 reactivate_fd(thread_fd, UBD_IRQ);
470
471 list_for_each_safe(list, next_ele, &restart){
472 ubd = container_of(list, struct ubd, restart);
473 list_del_init(&ubd->restart);
474 spin_lock_irqsave(&ubd->lock, flags);
475 do_ubd_request(ubd->queue);
476 spin_unlock_irqrestore(&ubd->lock, flags);
477 }
478}
479
480static irqreturn_t ubd_intr(int irq, void *dev)
481{
482 ubd_handler();
483 return IRQ_HANDLED;
484}
485
486/* Only changed by ubd_init, which is an initcall. */
487static int io_pid = -1;
488
489static void kill_io_thread(void)
490{
491 if(io_pid != -1)
492 os_kill_process(io_pid, 1);
493}
494
495__uml_exitcall(kill_io_thread);
496
497static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
498{
499 char *file;
500 int fd;
501 int err;
502
503 __u32 version;
504 __u32 align;
505 char *backing_file;
506 time_t mtime;
507 unsigned long long size;
508 int sector_size;
509 int bitmap_offset;
510
511 if (ubd_dev->file && ubd_dev->cow.file) {
512 file = ubd_dev->cow.file;
513
514 goto out;
515 }
516
517 fd = os_open_file(ubd_dev->file, global_openflags, 0);
518 if (fd < 0)
519 return fd;
520
521 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
522 &mtime, &size, §or_size, &align, &bitmap_offset);
523 os_close_file(fd);
524
525 if(err == -EINVAL)
526 file = ubd_dev->file;
527 else
528 file = backing_file;
529
530out:
531 return os_file_size(file, size_out);
532}
533
534static int read_cow_bitmap(int fd, void *buf, int offset, int len)
535{
536 int err;
537
538 err = os_seek_file(fd, offset);
539 if (err < 0)
540 return err;
541
542 err = os_read_file(fd, buf, len);
543 if (err < 0)
544 return err;
545
546 return 0;
547}
548
549static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
550{
551 unsigned long modtime;
552 unsigned long long actual;
553 int err;
554
555 err = os_file_modtime(file, &modtime);
556 if (err < 0) {
557 printk(KERN_ERR "Failed to get modification time of backing "
558 "file \"%s\", err = %d\n", file, -err);
559 return err;
560 }
561
562 err = os_file_size(file, &actual);
563 if (err < 0) {
564 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
565 "err = %d\n", file, -err);
566 return err;
567 }
568
569 if (actual != size) {
570 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
571 * the typecast.*/
572 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
573 "vs backing file\n", (unsigned long long) size, actual);
574 return -EINVAL;
575 }
576 if (modtime != mtime) {
577 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
578 "backing file\n", mtime, modtime);
579 return -EINVAL;
580 }
581 return 0;
582}
583
584static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
585{
586 struct uml_stat buf1, buf2;
587 int err;
588
589 if (from_cmdline == NULL)
590 return 0;
591 if (!strcmp(from_cmdline, from_cow))
592 return 0;
593
594 err = os_stat_file(from_cmdline, &buf1);
595 if (err < 0) {
596 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
597 -err);
598 return 0;
599 }
600 err = os_stat_file(from_cow, &buf2);
601 if (err < 0) {
602 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
603 -err);
604 return 1;
605 }
606 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
607 return 0;
608
609 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
610 "\"%s\" specified in COW header of \"%s\"\n",
611 from_cmdline, from_cow, cow);
612 return 1;
613}
614
615static int open_ubd_file(char *file, struct openflags *openflags, int shared,
616 char **backing_file_out, int *bitmap_offset_out,
617 unsigned long *bitmap_len_out, int *data_offset_out,
618 int *create_cow_out)
619{
620 time_t mtime;
621 unsigned long long size;
622 __u32 version, align;
623 char *backing_file;
624 int fd, err, sectorsize, asked_switch, mode = 0644;
625
626 fd = os_open_file(file, *openflags, mode);
627 if (fd < 0) {
628 if ((fd == -ENOENT) && (create_cow_out != NULL))
629 *create_cow_out = 1;
630 if (!openflags->w ||
631 ((fd != -EROFS) && (fd != -EACCES)))
632 return fd;
633 openflags->w = 0;
634 fd = os_open_file(file, *openflags, mode);
635 if (fd < 0)
636 return fd;
637 }
638
639 if (shared)
640 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
641 else {
642 err = os_lock_file(fd, openflags->w);
643 if (err < 0) {
644 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
645 file, -err);
646 goto out_close;
647 }
648 }
649
650 /* Successful return case! */
651 if (backing_file_out == NULL)
652 return fd;
653
654 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
655 &size, §orsize, &align, bitmap_offset_out);
656 if (err && (*backing_file_out != NULL)) {
657 printk(KERN_ERR "Failed to read COW header from COW file "
658 "\"%s\", errno = %d\n", file, -err);
659 goto out_close;
660 }
661 if (err)
662 return fd;
663
664 asked_switch = path_requires_switch(*backing_file_out, backing_file,
665 file);
666
667 /* Allow switching only if no mismatch. */
668 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
669 mtime)) {
670 printk(KERN_ERR "Switching backing file to '%s'\n",
671 *backing_file_out);
672 err = write_cow_header(file, fd, *backing_file_out,
673 sectorsize, align, &size);
674 if (err) {
675 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
676 goto out_close;
677 }
678 } else {
679 *backing_file_out = backing_file;
680 err = backing_file_mismatch(*backing_file_out, size, mtime);
681 if (err)
682 goto out_close;
683 }
684
685 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
686 bitmap_len_out, data_offset_out);
687
688 return fd;
689 out_close:
690 os_close_file(fd);
691 return err;
692}
693
694static int create_cow_file(char *cow_file, char *backing_file,
695 struct openflags flags,
696 int sectorsize, int alignment, int *bitmap_offset_out,
697 unsigned long *bitmap_len_out, int *data_offset_out)
698{
699 int err, fd;
700
701 flags.c = 1;
702 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
703 if (fd < 0) {
704 err = fd;
705 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
706 cow_file, -err);
707 goto out;
708 }
709
710 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
711 bitmap_offset_out, bitmap_len_out,
712 data_offset_out);
713 if (!err)
714 return fd;
715 os_close_file(fd);
716 out:
717 return err;
718}
719
720static void ubd_close_dev(struct ubd *ubd_dev)
721{
722 os_close_file(ubd_dev->fd);
723 if(ubd_dev->cow.file == NULL)
724 return;
725
726 os_close_file(ubd_dev->cow.fd);
727 vfree(ubd_dev->cow.bitmap);
728 ubd_dev->cow.bitmap = NULL;
729}
730
731static int ubd_open_dev(struct ubd *ubd_dev)
732{
733 struct openflags flags;
734 char **back_ptr;
735 int err, create_cow, *create_ptr;
736 int fd;
737
738 ubd_dev->openflags = ubd_dev->boot_openflags;
739 create_cow = 0;
740 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
741 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
742
743 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
744 back_ptr, &ubd_dev->cow.bitmap_offset,
745 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
746 create_ptr);
747
748 if((fd == -ENOENT) && create_cow){
749 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
750 ubd_dev->openflags, 1 << 9, PAGE_SIZE,
751 &ubd_dev->cow.bitmap_offset,
752 &ubd_dev->cow.bitmap_len,
753 &ubd_dev->cow.data_offset);
754 if(fd >= 0){
755 printk(KERN_INFO "Creating \"%s\" as COW file for "
756 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
757 }
758 }
759
760 if(fd < 0){
761 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
762 -fd);
763 return fd;
764 }
765 ubd_dev->fd = fd;
766
767 if(ubd_dev->cow.file != NULL){
768 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
769
770 err = -ENOMEM;
771 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
772 if(ubd_dev->cow.bitmap == NULL){
773 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
774 goto error;
775 }
776 flush_tlb_kernel_vm();
777
778 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
779 ubd_dev->cow.bitmap_offset,
780 ubd_dev->cow.bitmap_len);
781 if(err < 0)
782 goto error;
783
784 flags = ubd_dev->openflags;
785 flags.w = 0;
786 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
787 NULL, NULL, NULL, NULL);
788 if(err < 0) goto error;
789 ubd_dev->cow.fd = err;
790 }
791 return 0;
792 error:
793 os_close_file(ubd_dev->fd);
794 return err;
795}
796
797static void ubd_device_release(struct device *dev)
798{
799 struct ubd *ubd_dev = dev_get_drvdata(dev);
800
801 blk_cleanup_queue(ubd_dev->queue);
802 *ubd_dev = ((struct ubd) DEFAULT_UBD);
803}
804
805static int ubd_disk_register(int major, u64 size, int unit,
806 struct gendisk **disk_out)
807{
808 struct gendisk *disk;
809
810 disk = alloc_disk(1 << UBD_SHIFT);
811 if(disk == NULL)
812 return -ENOMEM;
813
814 disk->major = major;
815 disk->first_minor = unit << UBD_SHIFT;
816 disk->fops = &ubd_blops;
817 set_capacity(disk, size / 512);
818 if (major == UBD_MAJOR)
819 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
820 else
821 sprintf(disk->disk_name, "ubd_fake%d", unit);
822
823 /* sysfs register (not for ide fake devices) */
824 if (major == UBD_MAJOR) {
825 ubd_devs[unit].pdev.id = unit;
826 ubd_devs[unit].pdev.name = DRIVER_NAME;
827 ubd_devs[unit].pdev.dev.release = ubd_device_release;
828 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
829 platform_device_register(&ubd_devs[unit].pdev);
830 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
831 }
832
833 disk->private_data = &ubd_devs[unit];
834 disk->queue = ubd_devs[unit].queue;
835 add_disk(disk);
836
837 *disk_out = disk;
838 return 0;
839}
840
841#define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
842
843static int ubd_add(int n, char **error_out)
844{
845 struct ubd *ubd_dev = &ubd_devs[n];
846 int err = 0;
847
848 if(ubd_dev->file == NULL)
849 goto out;
850
851 err = ubd_file_size(ubd_dev, &ubd_dev->size);
852 if(err < 0){
853 *error_out = "Couldn't determine size of device's file";
854 goto out;
855 }
856
857 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
858
859 INIT_LIST_HEAD(&ubd_dev->restart);
860 sg_init_table(ubd_dev->sg, MAX_SG);
861
862 err = -ENOMEM;
863 ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
864 if (ubd_dev->queue == NULL) {
865 *error_out = "Failed to initialize device queue";
866 goto out;
867 }
868 ubd_dev->queue->queuedata = ubd_dev;
869
870 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
871 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
872 if(err){
873 *error_out = "Failed to register device";
874 goto out_cleanup;
875 }
876
877 if (fake_major != UBD_MAJOR)
878 ubd_disk_register(fake_major, ubd_dev->size, n,
879 &fake_gendisk[n]);
880
881 /*
882 * Perhaps this should also be under the "if (fake_major)" above
883 * using the fake_disk->disk_name
884 */
885 if (fake_ide)
886 make_ide_entries(ubd_gendisk[n]->disk_name);
887
888 err = 0;
889out:
890 return err;
891
892out_cleanup:
893 blk_cleanup_queue(ubd_dev->queue);
894 goto out;
895}
896
897static int ubd_config(char *str, char **error_out)
898{
899 int n, ret;
900
901 /* This string is possibly broken up and stored, so it's only
902 * freed if ubd_setup_common fails, or if only general options
903 * were set.
904 */
905 str = kstrdup(str, GFP_KERNEL);
906 if (str == NULL) {
907 *error_out = "Failed to allocate memory";
908 return -ENOMEM;
909 }
910
911 ret = ubd_setup_common(str, &n, error_out);
912 if (ret)
913 goto err_free;
914
915 if (n == -1) {
916 ret = 0;
917 goto err_free;
918 }
919
920 mutex_lock(&ubd_lock);
921 ret = ubd_add(n, error_out);
922 if (ret)
923 ubd_devs[n].file = NULL;
924 mutex_unlock(&ubd_lock);
925
926out:
927 return ret;
928
929err_free:
930 kfree(str);
931 goto out;
932}
933
934static int ubd_get_config(char *name, char *str, int size, char **error_out)
935{
936 struct ubd *ubd_dev;
937 int n, len = 0;
938
939 n = parse_unit(&name);
940 if((n >= MAX_DEV) || (n < 0)){
941 *error_out = "ubd_get_config : device number out of range";
942 return -1;
943 }
944
945 ubd_dev = &ubd_devs[n];
946 mutex_lock(&ubd_lock);
947
948 if(ubd_dev->file == NULL){
949 CONFIG_CHUNK(str, size, len, "", 1);
950 goto out;
951 }
952
953 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
954
955 if(ubd_dev->cow.file != NULL){
956 CONFIG_CHUNK(str, size, len, ",", 0);
957 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
958 }
959 else CONFIG_CHUNK(str, size, len, "", 1);
960
961 out:
962 mutex_unlock(&ubd_lock);
963 return len;
964}
965
966static int ubd_id(char **str, int *start_out, int *end_out)
967{
968 int n;
969
970 n = parse_unit(str);
971 *start_out = 0;
972 *end_out = MAX_DEV - 1;
973 return n;
974}
975
976static int ubd_remove(int n, char **error_out)
977{
978 struct gendisk *disk = ubd_gendisk[n];
979 struct ubd *ubd_dev;
980 int err = -ENODEV;
981
982 mutex_lock(&ubd_lock);
983
984 ubd_dev = &ubd_devs[n];
985
986 if(ubd_dev->file == NULL)
987 goto out;
988
989 /* you cannot remove a open disk */
990 err = -EBUSY;
991 if(ubd_dev->count > 0)
992 goto out;
993
994 ubd_gendisk[n] = NULL;
995 if(disk != NULL){
996 del_gendisk(disk);
997 put_disk(disk);
998 }
999
1000 if(fake_gendisk[n] != NULL){
1001 del_gendisk(fake_gendisk[n]);
1002 put_disk(fake_gendisk[n]);
1003 fake_gendisk[n] = NULL;
1004 }
1005
1006 err = 0;
1007 platform_device_unregister(&ubd_dev->pdev);
1008out:
1009 mutex_unlock(&ubd_lock);
1010 return err;
1011}
1012
1013/* All these are called by mconsole in process context and without
1014 * ubd-specific locks. The structure itself is const except for .list.
1015 */
1016static struct mc_device ubd_mc = {
1017 .list = LIST_HEAD_INIT(ubd_mc.list),
1018 .name = "ubd",
1019 .config = ubd_config,
1020 .get_config = ubd_get_config,
1021 .id = ubd_id,
1022 .remove = ubd_remove,
1023};
1024
1025static int __init ubd_mc_init(void)
1026{
1027 mconsole_register_dev(&ubd_mc);
1028 return 0;
1029}
1030
1031__initcall(ubd_mc_init);
1032
1033static int __init ubd0_init(void)
1034{
1035 struct ubd *ubd_dev = &ubd_devs[0];
1036
1037 mutex_lock(&ubd_lock);
1038 if(ubd_dev->file == NULL)
1039 ubd_dev->file = "root_fs";
1040 mutex_unlock(&ubd_lock);
1041
1042 return 0;
1043}
1044
1045__initcall(ubd0_init);
1046
1047/* Used in ubd_init, which is an initcall */
1048static struct platform_driver ubd_driver = {
1049 .driver = {
1050 .name = DRIVER_NAME,
1051 },
1052};
1053
1054static int __init ubd_init(void)
1055{
1056 char *error;
1057 int i, err;
1058
1059 if (register_blkdev(UBD_MAJOR, "ubd"))
1060 return -1;
1061
1062 if (fake_major != UBD_MAJOR) {
1063 char name[sizeof("ubd_nnn\0")];
1064
1065 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1066 if (register_blkdev(fake_major, "ubd"))
1067 return -1;
1068 }
1069 platform_driver_register(&ubd_driver);
1070 mutex_lock(&ubd_lock);
1071 for (i = 0; i < MAX_DEV; i++){
1072 err = ubd_add(i, &error);
1073 if(err)
1074 printk(KERN_ERR "Failed to initialize ubd device %d :"
1075 "%s\n", i, error);
1076 }
1077 mutex_unlock(&ubd_lock);
1078 return 0;
1079}
1080
1081late_initcall(ubd_init);
1082
1083static int __init ubd_driver_init(void){
1084 unsigned long stack;
1085 int err;
1086
1087 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1088 if(global_openflags.s){
1089 printk(KERN_INFO "ubd: Synchronous mode\n");
1090 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1091 * enough. So use anyway the io thread. */
1092 }
1093 stack = alloc_stack(0, 0);
1094 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1095 &thread_fd);
1096 if(io_pid < 0){
1097 printk(KERN_ERR
1098 "ubd : Failed to start I/O thread (errno = %d) - "
1099 "falling back to synchronous I/O\n", -io_pid);
1100 io_pid = -1;
1101 return 0;
1102 }
1103 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1104 0, "ubd", ubd_devs);
1105 if(err != 0)
1106 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1107 return 0;
1108}
1109
1110device_initcall(ubd_driver_init);
1111
1112static int ubd_open(struct block_device *bdev, fmode_t mode)
1113{
1114 struct gendisk *disk = bdev->bd_disk;
1115 struct ubd *ubd_dev = disk->private_data;
1116 int err = 0;
1117
1118 mutex_lock(&ubd_mutex);
1119 if(ubd_dev->count == 0){
1120 err = ubd_open_dev(ubd_dev);
1121 if(err){
1122 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1123 disk->disk_name, ubd_dev->file, -err);
1124 goto out;
1125 }
1126 }
1127 ubd_dev->count++;
1128 set_disk_ro(disk, !ubd_dev->openflags.w);
1129
1130 /* This should no more be needed. And it didn't work anyway to exclude
1131 * read-write remounting of filesystems.*/
1132 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1133 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1134 err = -EROFS;
1135 }*/
1136out:
1137 mutex_unlock(&ubd_mutex);
1138 return err;
1139}
1140
1141static int ubd_release(struct gendisk *disk, fmode_t mode)
1142{
1143 struct ubd *ubd_dev = disk->private_data;
1144
1145 mutex_lock(&ubd_mutex);
1146 if(--ubd_dev->count == 0)
1147 ubd_close_dev(ubd_dev);
1148 mutex_unlock(&ubd_mutex);
1149 return 0;
1150}
1151
1152static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1153 __u64 *cow_offset, unsigned long *bitmap,
1154 __u64 bitmap_offset, unsigned long *bitmap_words,
1155 __u64 bitmap_len)
1156{
1157 __u64 sector = io_offset >> 9;
1158 int i, update_bitmap = 0;
1159
1160 for(i = 0; i < length >> 9; i++){
1161 if(cow_mask != NULL)
1162 ubd_set_bit(i, (unsigned char *) cow_mask);
1163 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1164 continue;
1165
1166 update_bitmap = 1;
1167 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1168 }
1169
1170 if(!update_bitmap)
1171 return;
1172
1173 *cow_offset = sector / (sizeof(unsigned long) * 8);
1174
1175 /* This takes care of the case where we're exactly at the end of the
1176 * device, and *cow_offset + 1 is off the end. So, just back it up
1177 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1178 * for the original diagnosis.
1179 */
1180 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1181 sizeof(unsigned long)) - 1))
1182 (*cow_offset)--;
1183
1184 bitmap_words[0] = bitmap[*cow_offset];
1185 bitmap_words[1] = bitmap[*cow_offset + 1];
1186
1187 *cow_offset *= sizeof(unsigned long);
1188 *cow_offset += bitmap_offset;
1189}
1190
1191static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1192 __u64 bitmap_offset, __u64 bitmap_len)
1193{
1194 __u64 sector = req->offset >> 9;
1195 int i;
1196
1197 if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1198 panic("Operation too long");
1199
1200 if(req->op == UBD_READ) {
1201 for(i = 0; i < req->length >> 9; i++){
1202 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1203 ubd_set_bit(i, (unsigned char *)
1204 &req->sector_mask);
1205 }
1206 }
1207 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1208 &req->cow_offset, bitmap, bitmap_offset,
1209 req->bitmap_words, bitmap_len);
1210}
1211
1212/* Called with dev->lock held */
1213static void prepare_request(struct request *req, struct io_thread_req *io_req,
1214 unsigned long long offset, int page_offset,
1215 int len, struct page *page)
1216{
1217 struct gendisk *disk = req->rq_disk;
1218 struct ubd *ubd_dev = disk->private_data;
1219
1220 io_req->req = req;
1221 io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1222 ubd_dev->fd;
1223 io_req->fds[1] = ubd_dev->fd;
1224 io_req->cow_offset = -1;
1225 io_req->offset = offset;
1226 io_req->length = len;
1227 io_req->error = 0;
1228 io_req->sector_mask = 0;
1229
1230 io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1231 io_req->offsets[0] = 0;
1232 io_req->offsets[1] = ubd_dev->cow.data_offset;
1233 io_req->buffer = page_address(page) + page_offset;
1234 io_req->sectorsize = 1 << 9;
1235
1236 if(ubd_dev->cow.file != NULL)
1237 cowify_req(io_req, ubd_dev->cow.bitmap,
1238 ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1239
1240}
1241
1242/* Called with dev->lock held */
1243static void do_ubd_request(struct request_queue *q)
1244{
1245 struct io_thread_req *io_req;
1246 struct request *req;
1247 int n;
1248
1249 while(1){
1250 struct ubd *dev = q->queuedata;
1251 if(dev->end_sg == 0){
1252 struct request *req = blk_fetch_request(q);
1253 if(req == NULL)
1254 return;
1255
1256 dev->request = req;
1257 dev->rq_pos = blk_rq_pos(req);
1258 dev->start_sg = 0;
1259 dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1260 }
1261
1262 req = dev->request;
1263 while(dev->start_sg < dev->end_sg){
1264 struct scatterlist *sg = &dev->sg[dev->start_sg];
1265
1266 io_req = kmalloc(sizeof(struct io_thread_req),
1267 GFP_ATOMIC);
1268 if(io_req == NULL){
1269 if(list_empty(&dev->restart))
1270 list_add(&dev->restart, &restart);
1271 return;
1272 }
1273 prepare_request(req, io_req,
1274 (unsigned long long)dev->rq_pos << 9,
1275 sg->offset, sg->length, sg_page(sg));
1276
1277 n = os_write_file(thread_fd, &io_req,
1278 sizeof(struct io_thread_req *));
1279 if(n != sizeof(struct io_thread_req *)){
1280 if(n != -EAGAIN)
1281 printk("write to io thread failed, "
1282 "errno = %d\n", -n);
1283 else if(list_empty(&dev->restart))
1284 list_add(&dev->restart, &restart);
1285 kfree(io_req);
1286 return;
1287 }
1288
1289 dev->rq_pos += sg->length >> 9;
1290 dev->start_sg++;
1291 }
1292 dev->end_sg = 0;
1293 dev->request = NULL;
1294 }
1295}
1296
1297static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1298{
1299 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1300
1301 geo->heads = 128;
1302 geo->sectors = 32;
1303 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1304 return 0;
1305}
1306
1307static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1308 unsigned int cmd, unsigned long arg)
1309{
1310 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1311 u16 ubd_id[ATA_ID_WORDS];
1312
1313 switch (cmd) {
1314 struct cdrom_volctrl volume;
1315 case HDIO_GET_IDENTITY:
1316 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1317 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1318 ubd_id[ATA_ID_HEADS] = 128;
1319 ubd_id[ATA_ID_SECTORS] = 32;
1320 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1321 sizeof(ubd_id)))
1322 return -EFAULT;
1323 return 0;
1324
1325 case CDROMVOLREAD:
1326 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1327 return -EFAULT;
1328 volume.channel0 = 255;
1329 volume.channel1 = 255;
1330 volume.channel2 = 255;
1331 volume.channel3 = 255;
1332 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1333 return -EFAULT;
1334 return 0;
1335 }
1336 return -EINVAL;
1337}
1338
1339static int update_bitmap(struct io_thread_req *req)
1340{
1341 int n;
1342
1343 if(req->cow_offset == -1)
1344 return 0;
1345
1346 n = os_seek_file(req->fds[1], req->cow_offset);
1347 if(n < 0){
1348 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1349 return 1;
1350 }
1351
1352 n = os_write_file(req->fds[1], &req->bitmap_words,
1353 sizeof(req->bitmap_words));
1354 if(n != sizeof(req->bitmap_words)){
1355 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1356 req->fds[1]);
1357 return 1;
1358 }
1359
1360 return 0;
1361}
1362
1363static void do_io(struct io_thread_req *req)
1364{
1365 char *buf;
1366 unsigned long len;
1367 int n, nsectors, start, end, bit;
1368 int err;
1369 __u64 off;
1370
1371 nsectors = req->length / req->sectorsize;
1372 start = 0;
1373 do {
1374 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1375 end = start;
1376 while((end < nsectors) &&
1377 (ubd_test_bit(end, (unsigned char *)
1378 &req->sector_mask) == bit))
1379 end++;
1380
1381 off = req->offset + req->offsets[bit] +
1382 start * req->sectorsize;
1383 len = (end - start) * req->sectorsize;
1384 buf = &req->buffer[start * req->sectorsize];
1385
1386 err = os_seek_file(req->fds[bit], off);
1387 if(err < 0){
1388 printk("do_io - lseek failed : err = %d\n", -err);
1389 req->error = 1;
1390 return;
1391 }
1392 if(req->op == UBD_READ){
1393 n = 0;
1394 do {
1395 buf = &buf[n];
1396 len -= n;
1397 n = os_read_file(req->fds[bit], buf, len);
1398 if (n < 0) {
1399 printk("do_io - read failed, err = %d "
1400 "fd = %d\n", -n, req->fds[bit]);
1401 req->error = 1;
1402 return;
1403 }
1404 } while((n < len) && (n != 0));
1405 if (n < len) memset(&buf[n], 0, len - n);
1406 } else {
1407 n = os_write_file(req->fds[bit], buf, len);
1408 if(n != len){
1409 printk("do_io - write failed err = %d "
1410 "fd = %d\n", -n, req->fds[bit]);
1411 req->error = 1;
1412 return;
1413 }
1414 }
1415
1416 start = end;
1417 } while(start < nsectors);
1418
1419 req->error = update_bitmap(req);
1420}
1421
1422/* Changed in start_io_thread, which is serialized by being called only
1423 * from ubd_init, which is an initcall.
1424 */
1425int kernel_fd = -1;
1426
1427/* Only changed by the io thread. XXX: currently unused. */
1428static int io_count = 0;
1429
1430int io_thread(void *arg)
1431{
1432 struct io_thread_req *req;
1433 int n;
1434
1435 ignore_sigwinch_sig();
1436 while(1){
1437 n = os_read_file(kernel_fd, &req,
1438 sizeof(struct io_thread_req *));
1439 if(n != sizeof(struct io_thread_req *)){
1440 if(n < 0)
1441 printk("io_thread - read failed, fd = %d, "
1442 "err = %d\n", kernel_fd, -n);
1443 else {
1444 printk("io_thread - short read, fd = %d, "
1445 "length = %d\n", kernel_fd, n);
1446 }
1447 continue;
1448 }
1449 io_count++;
1450 do_io(req);
1451 n = os_write_file(kernel_fd, &req,
1452 sizeof(struct io_thread_req *));
1453 if(n != sizeof(struct io_thread_req *))
1454 printk("io_thread - write failed, fd = %d, err = %d\n",
1455 kernel_fd, -n);
1456 }
1457
1458 return 0;
1459}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2018 Cambridge Greys Ltd
4 * Copyright (C) 2015-2016 Anton Ivanov (aivanov@brocade.com)
5 * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
6 */
7
8/* 2001-09-28...2002-04-17
9 * Partition stuff by James_McMechan@hotmail.com
10 * old style ubd by setting UBD_SHIFT to 0
11 * 2002-09-27...2002-10-18 massive tinkering for 2.5
12 * partitions have changed in 2.5
13 * 2003-01-29 more tinkering for 2.5.59-1
14 * This should now address the sysfs problems and has
15 * the symlink for devfs to allow for booting with
16 * the common /dev/ubd/discX/... names rather than
17 * only /dev/ubdN/discN this version also has lots of
18 * clean ups preparing for ubd-many.
19 * James McMechan
20 */
21
22#define UBD_SHIFT 4
23
24#include <linux/module.h>
25#include <linux/init.h>
26#include <linux/blkdev.h>
27#include <linux/blk-mq.h>
28#include <linux/ata.h>
29#include <linux/hdreg.h>
30#include <linux/major.h>
31#include <linux/cdrom.h>
32#include <linux/proc_fs.h>
33#include <linux/seq_file.h>
34#include <linux/ctype.h>
35#include <linux/slab.h>
36#include <linux/vmalloc.h>
37#include <linux/platform_device.h>
38#include <linux/scatterlist.h>
39#include <kern_util.h>
40#include "mconsole_kern.h"
41#include <init.h>
42#include <irq_kern.h>
43#include "ubd.h"
44#include <os.h>
45#include "cow.h"
46
47/* Max request size is determined by sector mask - 32K */
48#define UBD_MAX_REQUEST (8 * sizeof(long))
49
50struct io_desc {
51 char *buffer;
52 unsigned long length;
53 unsigned long sector_mask;
54 unsigned long long cow_offset;
55 unsigned long bitmap_words[2];
56};
57
58struct io_thread_req {
59 struct request *req;
60 int fds[2];
61 unsigned long offsets[2];
62 unsigned long long offset;
63 int sectorsize;
64 int error;
65
66 int desc_cnt;
67 /* io_desc has to be the last element of the struct */
68 struct io_desc io_desc[];
69};
70
71
72static struct io_thread_req * (*irq_req_buffer)[];
73static struct io_thread_req *irq_remainder;
74static int irq_remainder_size;
75
76static struct io_thread_req * (*io_req_buffer)[];
77static struct io_thread_req *io_remainder;
78static int io_remainder_size;
79
80
81
82static inline int ubd_test_bit(__u64 bit, unsigned char *data)
83{
84 __u64 n;
85 int bits, off;
86
87 bits = sizeof(data[0]) * 8;
88 n = bit / bits;
89 off = bit % bits;
90 return (data[n] & (1 << off)) != 0;
91}
92
93static inline void ubd_set_bit(__u64 bit, unsigned char *data)
94{
95 __u64 n;
96 int bits, off;
97
98 bits = sizeof(data[0]) * 8;
99 n = bit / bits;
100 off = bit % bits;
101 data[n] |= (1 << off);
102}
103/*End stuff from ubd_user.h*/
104
105#define DRIVER_NAME "uml-blkdev"
106
107static DEFINE_MUTEX(ubd_lock);
108
109static int ubd_ioctl(struct block_device *bdev, blk_mode_t mode,
110 unsigned int cmd, unsigned long arg);
111static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
112
113#define MAX_DEV (16)
114
115static const struct block_device_operations ubd_blops = {
116 .owner = THIS_MODULE,
117 .ioctl = ubd_ioctl,
118 .compat_ioctl = blkdev_compat_ptr_ioctl,
119 .getgeo = ubd_getgeo,
120};
121
122#ifdef CONFIG_BLK_DEV_UBD_SYNC
123#define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
124 .cl = 1 })
125#else
126#define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
127 .cl = 1 })
128#endif
129static struct openflags global_openflags = OPEN_FLAGS;
130
131struct cow {
132 /* backing file name */
133 char *file;
134 /* backing file fd */
135 int fd;
136 unsigned long *bitmap;
137 unsigned long bitmap_len;
138 int bitmap_offset;
139 int data_offset;
140};
141
142#define MAX_SG 64
143
144struct ubd {
145 /* name (and fd, below) of the file opened for writing, either the
146 * backing or the cow file. */
147 char *file;
148 char *serial;
149 int fd;
150 __u64 size;
151 struct openflags boot_openflags;
152 struct openflags openflags;
153 unsigned shared:1;
154 unsigned no_cow:1;
155 unsigned no_trim:1;
156 struct cow cow;
157 struct platform_device pdev;
158 struct gendisk *disk;
159 struct blk_mq_tag_set tag_set;
160 spinlock_t lock;
161};
162
163#define DEFAULT_COW { \
164 .file = NULL, \
165 .fd = -1, \
166 .bitmap = NULL, \
167 .bitmap_offset = 0, \
168 .data_offset = 0, \
169}
170
171#define DEFAULT_UBD { \
172 .file = NULL, \
173 .serial = NULL, \
174 .fd = -1, \
175 .size = -1, \
176 .boot_openflags = OPEN_FLAGS, \
177 .openflags = OPEN_FLAGS, \
178 .no_cow = 0, \
179 .no_trim = 0, \
180 .shared = 0, \
181 .cow = DEFAULT_COW, \
182 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
183}
184
185/* Protected by ubd_lock */
186static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
187
188static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
189 const struct blk_mq_queue_data *bd);
190
191static int fake_ide_setup(char *str)
192{
193 pr_warn("The fake_ide option has been removed\n");
194 return 1;
195}
196__setup("fake_ide", fake_ide_setup);
197
198__uml_help(fake_ide_setup,
199"fake_ide\n"
200" Obsolete stub.\n\n"
201);
202
203static int parse_unit(char **ptr)
204{
205 char *str = *ptr, *end;
206 int n = -1;
207
208 if(isdigit(*str)) {
209 n = simple_strtoul(str, &end, 0);
210 if(end == str)
211 return -1;
212 *ptr = end;
213 }
214 else if (('a' <= *str) && (*str <= 'z')) {
215 n = *str - 'a';
216 str++;
217 *ptr = str;
218 }
219 return n;
220}
221
222/* If *index_out == -1 at exit, the passed option was a general one;
223 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
224 * should not be freed on exit.
225 */
226static int ubd_setup_common(char *str, int *index_out, char **error_out)
227{
228 struct ubd *ubd_dev;
229 struct openflags flags = global_openflags;
230 char *file, *backing_file, *serial;
231 int n, err = 0, i;
232
233 if(index_out) *index_out = -1;
234 n = *str;
235 if(n == '='){
236 str++;
237 if(!strcmp(str, "sync")){
238 global_openflags = of_sync(global_openflags);
239 return err;
240 }
241
242 pr_warn("fake major not supported any more\n");
243 return 0;
244 }
245
246 n = parse_unit(&str);
247 if(n < 0){
248 *error_out = "Couldn't parse device number";
249 return -EINVAL;
250 }
251 if(n >= MAX_DEV){
252 *error_out = "Device number out of range";
253 return 1;
254 }
255
256 err = -EBUSY;
257 mutex_lock(&ubd_lock);
258
259 ubd_dev = &ubd_devs[n];
260 if(ubd_dev->file != NULL){
261 *error_out = "Device is already configured";
262 goto out;
263 }
264
265 if (index_out)
266 *index_out = n;
267
268 err = -EINVAL;
269 for (i = 0; i < sizeof("rscdt="); i++) {
270 switch (*str) {
271 case 'r':
272 flags.w = 0;
273 break;
274 case 's':
275 flags.s = 1;
276 break;
277 case 'd':
278 ubd_dev->no_cow = 1;
279 break;
280 case 'c':
281 ubd_dev->shared = 1;
282 break;
283 case 't':
284 ubd_dev->no_trim = 1;
285 break;
286 case '=':
287 str++;
288 goto break_loop;
289 default:
290 *error_out = "Expected '=' or flag letter "
291 "(r, s, c, t or d)";
292 goto out;
293 }
294 str++;
295 }
296
297 if (*str == '=')
298 *error_out = "Too many flags specified";
299 else
300 *error_out = "Missing '='";
301 goto out;
302
303break_loop:
304 file = strsep(&str, ",:");
305 if (*file == '\0')
306 file = NULL;
307
308 backing_file = strsep(&str, ",:");
309 if (backing_file && *backing_file == '\0')
310 backing_file = NULL;
311
312 serial = strsep(&str, ",:");
313 if (serial && *serial == '\0')
314 serial = NULL;
315
316 if (backing_file && ubd_dev->no_cow) {
317 *error_out = "Can't specify both 'd' and a cow file";
318 goto out;
319 }
320
321 err = 0;
322 ubd_dev->file = file;
323 ubd_dev->cow.file = backing_file;
324 ubd_dev->serial = serial;
325 ubd_dev->boot_openflags = flags;
326out:
327 mutex_unlock(&ubd_lock);
328 return err;
329}
330
331static int ubd_setup(char *str)
332{
333 char *error;
334 int err;
335
336 err = ubd_setup_common(str, NULL, &error);
337 if(err)
338 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
339 "%s\n", str, error);
340 return 1;
341}
342
343__setup("ubd", ubd_setup);
344__uml_help(ubd_setup,
345"ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n"
346" This is used to associate a device with a file in the underlying\n"
347" filesystem. When specifying two filenames, the first one is the\n"
348" COW name and the second is the backing file name. As separator you can\n"
349" use either a ':' or a ',': the first one allows writing things like;\n"
350" ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
351" while with a ',' the shell would not expand the 2nd '~'.\n"
352" When using only one filename, UML will detect whether to treat it like\n"
353" a COW file or a backing file. To override this detection, add the 'd'\n"
354" flag:\n"
355" ubd0d=BackingFile\n"
356" Usually, there is a filesystem in the file, but \n"
357" that's not required. Swap devices containing swap files can be\n"
358" specified like this. Also, a file which doesn't contain a\n"
359" filesystem can have its contents read in the virtual \n"
360" machine by running 'dd' on the device. <n> must be in the range\n"
361" 0 to 7. Appending an 'r' to the number will cause that device\n"
362" to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
363" an 's' will cause data to be written to disk on the host immediately.\n"
364" 'c' will cause the device to be treated as being shared between multiple\n"
365" UMLs and file locking will be turned off - this is appropriate for a\n"
366" cluster filesystem and inappropriate at almost all other times.\n\n"
367" 't' will disable trim/discard support on the device (enabled by default).\n\n"
368" An optional device serial number can be exposed using the serial parameter\n"
369" on the cmdline which is exposed as a sysfs entry. This is particularly\n"
370" useful when a unique number should be given to the device. Note when\n"
371" specifying a label, the filename2 must be also presented. It can be\n"
372" an empty string, in which case the backing file is not used:\n"
373" ubd0=File,,Serial\n"
374);
375
376static int udb_setup(char *str)
377{
378 printk("udb%s specified on command line is almost certainly a ubd -> "
379 "udb TYPO\n", str);
380 return 1;
381}
382
383__setup("udb", udb_setup);
384__uml_help(udb_setup,
385"udb\n"
386" This option is here solely to catch ubd -> udb typos, which can be\n"
387" to impossible to catch visually unless you specifically look for\n"
388" them. The only result of any option starting with 'udb' is an error\n"
389" in the boot output.\n\n"
390);
391
392/* Only changed by ubd_init, which is an initcall. */
393static int thread_fd = -1;
394
395/* Function to read several request pointers at a time
396* handling fractional reads if (and as) needed
397*/
398
399static int bulk_req_safe_read(
400 int fd,
401 struct io_thread_req * (*request_buffer)[],
402 struct io_thread_req **remainder,
403 int *remainder_size,
404 int max_recs
405 )
406{
407 int n = 0;
408 int res = 0;
409
410 if (*remainder_size > 0) {
411 memmove(
412 (char *) request_buffer,
413 (char *) remainder, *remainder_size
414 );
415 n = *remainder_size;
416 }
417
418 res = os_read_file(
419 fd,
420 ((char *) request_buffer) + *remainder_size,
421 sizeof(struct io_thread_req *)*max_recs
422 - *remainder_size
423 );
424 if (res > 0) {
425 n += res;
426 if ((n % sizeof(struct io_thread_req *)) > 0) {
427 /*
428 * Read somehow returned not a multiple of dword
429 * theoretically possible, but never observed in the
430 * wild, so read routine must be able to handle it
431 */
432 *remainder_size = n % sizeof(struct io_thread_req *);
433 WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
434 memmove(
435 remainder,
436 ((char *) request_buffer) +
437 (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
438 *remainder_size
439 );
440 n = n - *remainder_size;
441 }
442 } else {
443 n = res;
444 }
445 return n;
446}
447
448static void ubd_end_request(struct io_thread_req *io_req)
449{
450 if (io_req->error == BLK_STS_NOTSUPP) {
451 if (req_op(io_req->req) == REQ_OP_DISCARD)
452 blk_queue_disable_discard(io_req->req->q);
453 else if (req_op(io_req->req) == REQ_OP_WRITE_ZEROES)
454 blk_queue_disable_write_zeroes(io_req->req->q);
455 }
456 blk_mq_end_request(io_req->req, io_req->error);
457 kfree(io_req);
458}
459
460static irqreturn_t ubd_intr(int irq, void *dev)
461{
462 int len, i;
463
464 while ((len = bulk_req_safe_read(thread_fd, irq_req_buffer,
465 &irq_remainder, &irq_remainder_size,
466 UBD_REQ_BUFFER_SIZE)) >= 0) {
467 for (i = 0; i < len / sizeof(struct io_thread_req *); i++)
468 ubd_end_request((*irq_req_buffer)[i]);
469 }
470
471 if (len < 0 && len != -EAGAIN)
472 pr_err("spurious interrupt in %s, err = %d\n", __func__, len);
473 return IRQ_HANDLED;
474}
475
476/* Only changed by ubd_init, which is an initcall. */
477static int io_pid = -1;
478
479static void kill_io_thread(void)
480{
481 if(io_pid != -1)
482 os_kill_process(io_pid, 1);
483}
484
485__uml_exitcall(kill_io_thread);
486
487static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
488{
489 char *file;
490 int fd;
491 int err;
492
493 __u32 version;
494 __u32 align;
495 char *backing_file;
496 time64_t mtime;
497 unsigned long long size;
498 int sector_size;
499 int bitmap_offset;
500
501 if (ubd_dev->file && ubd_dev->cow.file) {
502 file = ubd_dev->cow.file;
503
504 goto out;
505 }
506
507 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
508 if (fd < 0)
509 return fd;
510
511 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
512 &mtime, &size, §or_size, &align, &bitmap_offset);
513 os_close_file(fd);
514
515 if(err == -EINVAL)
516 file = ubd_dev->file;
517 else
518 file = backing_file;
519
520out:
521 return os_file_size(file, size_out);
522}
523
524static int read_cow_bitmap(int fd, void *buf, int offset, int len)
525{
526 int err;
527
528 err = os_pread_file(fd, buf, len, offset);
529 if (err < 0)
530 return err;
531
532 return 0;
533}
534
535static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
536{
537 time64_t modtime;
538 unsigned long long actual;
539 int err;
540
541 err = os_file_modtime(file, &modtime);
542 if (err < 0) {
543 printk(KERN_ERR "Failed to get modification time of backing "
544 "file \"%s\", err = %d\n", file, -err);
545 return err;
546 }
547
548 err = os_file_size(file, &actual);
549 if (err < 0) {
550 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
551 "err = %d\n", file, -err);
552 return err;
553 }
554
555 if (actual != size) {
556 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
557 * the typecast.*/
558 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
559 "vs backing file\n", (unsigned long long) size, actual);
560 return -EINVAL;
561 }
562 if (modtime != mtime) {
563 printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
564 "backing file\n", mtime, modtime);
565 return -EINVAL;
566 }
567 return 0;
568}
569
570static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
571{
572 struct uml_stat buf1, buf2;
573 int err;
574
575 if (from_cmdline == NULL)
576 return 0;
577 if (!strcmp(from_cmdline, from_cow))
578 return 0;
579
580 err = os_stat_file(from_cmdline, &buf1);
581 if (err < 0) {
582 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
583 -err);
584 return 0;
585 }
586 err = os_stat_file(from_cow, &buf2);
587 if (err < 0) {
588 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
589 -err);
590 return 1;
591 }
592 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
593 return 0;
594
595 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
596 "\"%s\" specified in COW header of \"%s\"\n",
597 from_cmdline, from_cow, cow);
598 return 1;
599}
600
601static int open_ubd_file(char *file, struct openflags *openflags, int shared,
602 char **backing_file_out, int *bitmap_offset_out,
603 unsigned long *bitmap_len_out, int *data_offset_out,
604 int *create_cow_out)
605{
606 time64_t mtime;
607 unsigned long long size;
608 __u32 version, align;
609 char *backing_file;
610 int fd, err, sectorsize, asked_switch, mode = 0644;
611
612 fd = os_open_file(file, *openflags, mode);
613 if (fd < 0) {
614 if ((fd == -ENOENT) && (create_cow_out != NULL))
615 *create_cow_out = 1;
616 if (!openflags->w ||
617 ((fd != -EROFS) && (fd != -EACCES)))
618 return fd;
619 openflags->w = 0;
620 fd = os_open_file(file, *openflags, mode);
621 if (fd < 0)
622 return fd;
623 }
624
625 if (shared)
626 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
627 else {
628 err = os_lock_file(fd, openflags->w);
629 if (err < 0) {
630 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
631 file, -err);
632 goto out_close;
633 }
634 }
635
636 /* Successful return case! */
637 if (backing_file_out == NULL)
638 return fd;
639
640 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
641 &size, §orsize, &align, bitmap_offset_out);
642 if (err && (*backing_file_out != NULL)) {
643 printk(KERN_ERR "Failed to read COW header from COW file "
644 "\"%s\", errno = %d\n", file, -err);
645 goto out_close;
646 }
647 if (err)
648 return fd;
649
650 asked_switch = path_requires_switch(*backing_file_out, backing_file,
651 file);
652
653 /* Allow switching only if no mismatch. */
654 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
655 mtime)) {
656 printk(KERN_ERR "Switching backing file to '%s'\n",
657 *backing_file_out);
658 err = write_cow_header(file, fd, *backing_file_out,
659 sectorsize, align, &size);
660 if (err) {
661 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
662 goto out_close;
663 }
664 } else {
665 *backing_file_out = backing_file;
666 err = backing_file_mismatch(*backing_file_out, size, mtime);
667 if (err)
668 goto out_close;
669 }
670
671 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
672 bitmap_len_out, data_offset_out);
673
674 return fd;
675 out_close:
676 os_close_file(fd);
677 return err;
678}
679
680static int create_cow_file(char *cow_file, char *backing_file,
681 struct openflags flags,
682 int sectorsize, int alignment, int *bitmap_offset_out,
683 unsigned long *bitmap_len_out, int *data_offset_out)
684{
685 int err, fd;
686
687 flags.c = 1;
688 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
689 if (fd < 0) {
690 err = fd;
691 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
692 cow_file, -err);
693 goto out;
694 }
695
696 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
697 bitmap_offset_out, bitmap_len_out,
698 data_offset_out);
699 if (!err)
700 return fd;
701 os_close_file(fd);
702 out:
703 return err;
704}
705
706static void ubd_close_dev(struct ubd *ubd_dev)
707{
708 os_close_file(ubd_dev->fd);
709 if(ubd_dev->cow.file == NULL)
710 return;
711
712 os_close_file(ubd_dev->cow.fd);
713 vfree(ubd_dev->cow.bitmap);
714 ubd_dev->cow.bitmap = NULL;
715}
716
717static int ubd_open_dev(struct ubd *ubd_dev)
718{
719 struct openflags flags;
720 char **back_ptr;
721 int err, create_cow, *create_ptr;
722 int fd;
723
724 ubd_dev->openflags = ubd_dev->boot_openflags;
725 create_cow = 0;
726 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
727 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
728
729 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
730 back_ptr, &ubd_dev->cow.bitmap_offset,
731 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
732 create_ptr);
733
734 if((fd == -ENOENT) && create_cow){
735 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
736 ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
737 &ubd_dev->cow.bitmap_offset,
738 &ubd_dev->cow.bitmap_len,
739 &ubd_dev->cow.data_offset);
740 if(fd >= 0){
741 printk(KERN_INFO "Creating \"%s\" as COW file for "
742 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
743 }
744 }
745
746 if(fd < 0){
747 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
748 -fd);
749 return fd;
750 }
751 ubd_dev->fd = fd;
752
753 if(ubd_dev->cow.file != NULL){
754 err = -ENOMEM;
755 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
756 if(ubd_dev->cow.bitmap == NULL){
757 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
758 goto error;
759 }
760
761 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
762 ubd_dev->cow.bitmap_offset,
763 ubd_dev->cow.bitmap_len);
764 if(err < 0)
765 goto error;
766
767 flags = ubd_dev->openflags;
768 flags.w = 0;
769 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
770 NULL, NULL, NULL, NULL);
771 if(err < 0) goto error;
772 ubd_dev->cow.fd = err;
773 }
774 return 0;
775 error:
776 os_close_file(ubd_dev->fd);
777 return err;
778}
779
780static void ubd_device_release(struct device *dev)
781{
782 struct ubd *ubd_dev = container_of(dev, struct ubd, pdev.dev);
783
784 blk_mq_free_tag_set(&ubd_dev->tag_set);
785 *ubd_dev = ((struct ubd) DEFAULT_UBD);
786}
787
788static ssize_t serial_show(struct device *dev,
789 struct device_attribute *attr, char *buf)
790{
791 struct gendisk *disk = dev_to_disk(dev);
792 struct ubd *ubd_dev = disk->private_data;
793
794 if (!ubd_dev)
795 return 0;
796
797 return sprintf(buf, "%s", ubd_dev->serial);
798}
799
800static DEVICE_ATTR_RO(serial);
801
802static struct attribute *ubd_attrs[] = {
803 &dev_attr_serial.attr,
804 NULL,
805};
806
807static umode_t ubd_attrs_are_visible(struct kobject *kobj,
808 struct attribute *a, int n)
809{
810 return a->mode;
811}
812
813static const struct attribute_group ubd_attr_group = {
814 .attrs = ubd_attrs,
815 .is_visible = ubd_attrs_are_visible,
816};
817
818static const struct attribute_group *ubd_attr_groups[] = {
819 &ubd_attr_group,
820 NULL,
821};
822
823#define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
824
825static const struct blk_mq_ops ubd_mq_ops = {
826 .queue_rq = ubd_queue_rq,
827};
828
829static int ubd_add(int n, char **error_out)
830{
831 struct ubd *ubd_dev = &ubd_devs[n];
832 struct queue_limits lim = {
833 .max_segments = MAX_SG,
834 .seg_boundary_mask = PAGE_SIZE - 1,
835 .features = BLK_FEAT_WRITE_CACHE,
836 };
837 struct gendisk *disk;
838 int err = 0;
839
840 if(ubd_dev->file == NULL)
841 goto out;
842
843 if (ubd_dev->cow.file)
844 lim.max_hw_sectors = 8 * sizeof(long);
845 if (!ubd_dev->no_trim) {
846 lim.max_hw_discard_sectors = UBD_MAX_REQUEST;
847 lim.max_write_zeroes_sectors = UBD_MAX_REQUEST;
848 }
849
850 err = ubd_file_size(ubd_dev, &ubd_dev->size);
851 if(err < 0){
852 *error_out = "Couldn't determine size of device's file";
853 goto out;
854 }
855
856 err = ubd_open_dev(ubd_dev);
857 if (err) {
858 pr_err("ubd%c: Can't open \"%s\": errno = %d\n",
859 'a' + n, ubd_dev->file, -err);
860 goto out;
861 }
862
863 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
864
865 ubd_dev->tag_set.ops = &ubd_mq_ops;
866 ubd_dev->tag_set.queue_depth = 64;
867 ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
868 ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
869 ubd_dev->tag_set.driver_data = ubd_dev;
870 ubd_dev->tag_set.nr_hw_queues = 1;
871
872 err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
873 if (err)
874 goto out_close;
875
876 disk = blk_mq_alloc_disk(&ubd_dev->tag_set, &lim, ubd_dev);
877 if (IS_ERR(disk)) {
878 err = PTR_ERR(disk);
879 goto out_cleanup_tags;
880 }
881
882 disk->major = UBD_MAJOR;
883 disk->first_minor = n << UBD_SHIFT;
884 disk->minors = 1 << UBD_SHIFT;
885 disk->fops = &ubd_blops;
886 set_capacity(disk, ubd_dev->size / 512);
887 sprintf(disk->disk_name, "ubd%c", 'a' + n);
888 disk->private_data = ubd_dev;
889 set_disk_ro(disk, !ubd_dev->openflags.w);
890
891 ubd_dev->pdev.id = n;
892 ubd_dev->pdev.name = DRIVER_NAME;
893 ubd_dev->pdev.dev.release = ubd_device_release;
894 dev_set_drvdata(&ubd_dev->pdev.dev, ubd_dev);
895 platform_device_register(&ubd_dev->pdev);
896
897 err = device_add_disk(&ubd_dev->pdev.dev, disk, ubd_attr_groups);
898 if (err)
899 goto out_cleanup_disk;
900
901 ubd_dev->disk = disk;
902
903 return 0;
904
905out_cleanup_disk:
906 put_disk(disk);
907out_cleanup_tags:
908 blk_mq_free_tag_set(&ubd_dev->tag_set);
909out_close:
910 ubd_close_dev(ubd_dev);
911out:
912 return err;
913}
914
915static int ubd_config(char *str, char **error_out)
916{
917 int n, ret;
918
919 /* This string is possibly broken up and stored, so it's only
920 * freed if ubd_setup_common fails, or if only general options
921 * were set.
922 */
923 str = kstrdup(str, GFP_KERNEL);
924 if (str == NULL) {
925 *error_out = "Failed to allocate memory";
926 return -ENOMEM;
927 }
928
929 ret = ubd_setup_common(str, &n, error_out);
930 if (ret)
931 goto err_free;
932
933 if (n == -1) {
934 ret = 0;
935 goto err_free;
936 }
937
938 mutex_lock(&ubd_lock);
939 ret = ubd_add(n, error_out);
940 if (ret)
941 ubd_devs[n].file = NULL;
942 mutex_unlock(&ubd_lock);
943
944out:
945 return ret;
946
947err_free:
948 kfree(str);
949 goto out;
950}
951
952static int ubd_get_config(char *name, char *str, int size, char **error_out)
953{
954 struct ubd *ubd_dev;
955 int n, len = 0;
956
957 n = parse_unit(&name);
958 if((n >= MAX_DEV) || (n < 0)){
959 *error_out = "ubd_get_config : device number out of range";
960 return -1;
961 }
962
963 ubd_dev = &ubd_devs[n];
964 mutex_lock(&ubd_lock);
965
966 if(ubd_dev->file == NULL){
967 CONFIG_CHUNK(str, size, len, "", 1);
968 goto out;
969 }
970
971 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
972
973 if(ubd_dev->cow.file != NULL){
974 CONFIG_CHUNK(str, size, len, ",", 0);
975 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
976 }
977 else CONFIG_CHUNK(str, size, len, "", 1);
978
979 out:
980 mutex_unlock(&ubd_lock);
981 return len;
982}
983
984static int ubd_id(char **str, int *start_out, int *end_out)
985{
986 int n;
987
988 n = parse_unit(str);
989 *start_out = 0;
990 *end_out = MAX_DEV - 1;
991 return n;
992}
993
994static int ubd_remove(int n, char **error_out)
995{
996 struct ubd *ubd_dev;
997 int err = -ENODEV;
998
999 mutex_lock(&ubd_lock);
1000
1001 ubd_dev = &ubd_devs[n];
1002
1003 if(ubd_dev->file == NULL)
1004 goto out;
1005
1006 if (ubd_dev->disk) {
1007 /* you cannot remove a open disk */
1008 err = -EBUSY;
1009 if (disk_openers(ubd_dev->disk))
1010 goto out;
1011
1012 del_gendisk(ubd_dev->disk);
1013 ubd_close_dev(ubd_dev);
1014 put_disk(ubd_dev->disk);
1015 }
1016
1017 err = 0;
1018 platform_device_unregister(&ubd_dev->pdev);
1019out:
1020 mutex_unlock(&ubd_lock);
1021 return err;
1022}
1023
1024/* All these are called by mconsole in process context and without
1025 * ubd-specific locks. The structure itself is const except for .list.
1026 */
1027static struct mc_device ubd_mc = {
1028 .list = LIST_HEAD_INIT(ubd_mc.list),
1029 .name = "ubd",
1030 .config = ubd_config,
1031 .get_config = ubd_get_config,
1032 .id = ubd_id,
1033 .remove = ubd_remove,
1034};
1035
1036static int __init ubd_mc_init(void)
1037{
1038 mconsole_register_dev(&ubd_mc);
1039 return 0;
1040}
1041
1042__initcall(ubd_mc_init);
1043
1044static int __init ubd0_init(void)
1045{
1046 struct ubd *ubd_dev = &ubd_devs[0];
1047
1048 mutex_lock(&ubd_lock);
1049 if(ubd_dev->file == NULL)
1050 ubd_dev->file = "root_fs";
1051 mutex_unlock(&ubd_lock);
1052
1053 return 0;
1054}
1055
1056__initcall(ubd0_init);
1057
1058/* Used in ubd_init, which is an initcall */
1059static struct platform_driver ubd_driver = {
1060 .driver = {
1061 .name = DRIVER_NAME,
1062 },
1063};
1064
1065static int __init ubd_init(void)
1066{
1067 char *error;
1068 int i, err;
1069
1070 if (register_blkdev(UBD_MAJOR, "ubd"))
1071 return -1;
1072
1073 irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1074 sizeof(struct io_thread_req *),
1075 GFP_KERNEL
1076 );
1077 irq_remainder = 0;
1078
1079 if (irq_req_buffer == NULL) {
1080 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1081 return -ENOMEM;
1082 }
1083 io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1084 sizeof(struct io_thread_req *),
1085 GFP_KERNEL
1086 );
1087
1088 io_remainder = 0;
1089
1090 if (io_req_buffer == NULL) {
1091 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1092 return -ENOMEM;
1093 }
1094 platform_driver_register(&ubd_driver);
1095 mutex_lock(&ubd_lock);
1096 for (i = 0; i < MAX_DEV; i++){
1097 err = ubd_add(i, &error);
1098 if(err)
1099 printk(KERN_ERR "Failed to initialize ubd device %d :"
1100 "%s\n", i, error);
1101 }
1102 mutex_unlock(&ubd_lock);
1103 return 0;
1104}
1105
1106late_initcall(ubd_init);
1107
1108static int __init ubd_driver_init(void){
1109 unsigned long stack;
1110 int err;
1111
1112 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1113 if(global_openflags.s){
1114 printk(KERN_INFO "ubd: Synchronous mode\n");
1115 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1116 * enough. So use anyway the io thread. */
1117 }
1118 stack = alloc_stack(0, 0);
1119 io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd);
1120 if(io_pid < 0){
1121 printk(KERN_ERR
1122 "ubd : Failed to start I/O thread (errno = %d) - "
1123 "falling back to synchronous I/O\n", -io_pid);
1124 io_pid = -1;
1125 return 0;
1126 }
1127 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1128 0, "ubd", ubd_devs);
1129 if(err < 0)
1130 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1131 return 0;
1132}
1133
1134device_initcall(ubd_driver_init);
1135
1136static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1137 __u64 *cow_offset, unsigned long *bitmap,
1138 __u64 bitmap_offset, unsigned long *bitmap_words,
1139 __u64 bitmap_len)
1140{
1141 __u64 sector = io_offset >> SECTOR_SHIFT;
1142 int i, update_bitmap = 0;
1143
1144 for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1145 if(cow_mask != NULL)
1146 ubd_set_bit(i, (unsigned char *) cow_mask);
1147 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1148 continue;
1149
1150 update_bitmap = 1;
1151 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1152 }
1153
1154 if(!update_bitmap)
1155 return;
1156
1157 *cow_offset = sector / (sizeof(unsigned long) * 8);
1158
1159 /* This takes care of the case where we're exactly at the end of the
1160 * device, and *cow_offset + 1 is off the end. So, just back it up
1161 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1162 * for the original diagnosis.
1163 */
1164 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1165 sizeof(unsigned long)) - 1))
1166 (*cow_offset)--;
1167
1168 bitmap_words[0] = bitmap[*cow_offset];
1169 bitmap_words[1] = bitmap[*cow_offset + 1];
1170
1171 *cow_offset *= sizeof(unsigned long);
1172 *cow_offset += bitmap_offset;
1173}
1174
1175static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1176 unsigned long offset, unsigned long *bitmap,
1177 __u64 bitmap_offset, __u64 bitmap_len)
1178{
1179 __u64 sector = offset >> SECTOR_SHIFT;
1180 int i;
1181
1182 if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1183 panic("Operation too long");
1184
1185 if (req_op(req->req) == REQ_OP_READ) {
1186 for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1187 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1188 ubd_set_bit(i, (unsigned char *)
1189 &segment->sector_mask);
1190 }
1191 } else {
1192 cowify_bitmap(offset, segment->length, &segment->sector_mask,
1193 &segment->cow_offset, bitmap, bitmap_offset,
1194 segment->bitmap_words, bitmap_len);
1195 }
1196}
1197
1198static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1199 struct request *req)
1200{
1201 struct bio_vec bvec;
1202 struct req_iterator iter;
1203 int i = 0;
1204 unsigned long byte_offset = io_req->offset;
1205 enum req_op op = req_op(req);
1206
1207 if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1208 io_req->io_desc[0].buffer = NULL;
1209 io_req->io_desc[0].length = blk_rq_bytes(req);
1210 } else {
1211 rq_for_each_segment(bvec, req, iter) {
1212 BUG_ON(i >= io_req->desc_cnt);
1213
1214 io_req->io_desc[i].buffer = bvec_virt(&bvec);
1215 io_req->io_desc[i].length = bvec.bv_len;
1216 i++;
1217 }
1218 }
1219
1220 if (dev->cow.file) {
1221 for (i = 0; i < io_req->desc_cnt; i++) {
1222 cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1223 dev->cow.bitmap, dev->cow.bitmap_offset,
1224 dev->cow.bitmap_len);
1225 byte_offset += io_req->io_desc[i].length;
1226 }
1227
1228 }
1229}
1230
1231static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1232 int desc_cnt)
1233{
1234 struct io_thread_req *io_req;
1235 int i;
1236
1237 io_req = kmalloc(sizeof(*io_req) +
1238 (desc_cnt * sizeof(struct io_desc)),
1239 GFP_ATOMIC);
1240 if (!io_req)
1241 return NULL;
1242
1243 io_req->req = req;
1244 if (dev->cow.file)
1245 io_req->fds[0] = dev->cow.fd;
1246 else
1247 io_req->fds[0] = dev->fd;
1248 io_req->error = 0;
1249 io_req->sectorsize = SECTOR_SIZE;
1250 io_req->fds[1] = dev->fd;
1251 io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1252 io_req->offsets[0] = 0;
1253 io_req->offsets[1] = dev->cow.data_offset;
1254
1255 for (i = 0 ; i < desc_cnt; i++) {
1256 io_req->io_desc[i].sector_mask = 0;
1257 io_req->io_desc[i].cow_offset = -1;
1258 }
1259
1260 return io_req;
1261}
1262
1263static int ubd_submit_request(struct ubd *dev, struct request *req)
1264{
1265 int segs = 0;
1266 struct io_thread_req *io_req;
1267 int ret;
1268 enum req_op op = req_op(req);
1269
1270 if (op == REQ_OP_FLUSH)
1271 segs = 0;
1272 else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1273 segs = 1;
1274 else
1275 segs = blk_rq_nr_phys_segments(req);
1276
1277 io_req = ubd_alloc_req(dev, req, segs);
1278 if (!io_req)
1279 return -ENOMEM;
1280
1281 io_req->desc_cnt = segs;
1282 if (segs)
1283 ubd_map_req(dev, io_req, req);
1284
1285 ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1286 if (ret != sizeof(io_req)) {
1287 if (ret != -EAGAIN)
1288 pr_err("write to io thread failed: %d\n", -ret);
1289 kfree(io_req);
1290 }
1291 return ret;
1292}
1293
1294static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1295 const struct blk_mq_queue_data *bd)
1296{
1297 struct ubd *ubd_dev = hctx->queue->queuedata;
1298 struct request *req = bd->rq;
1299 int ret = 0, res = BLK_STS_OK;
1300
1301 blk_mq_start_request(req);
1302
1303 spin_lock_irq(&ubd_dev->lock);
1304
1305 switch (req_op(req)) {
1306 case REQ_OP_FLUSH:
1307 case REQ_OP_READ:
1308 case REQ_OP_WRITE:
1309 case REQ_OP_DISCARD:
1310 case REQ_OP_WRITE_ZEROES:
1311 ret = ubd_submit_request(ubd_dev, req);
1312 break;
1313 default:
1314 WARN_ON_ONCE(1);
1315 res = BLK_STS_NOTSUPP;
1316 }
1317
1318 spin_unlock_irq(&ubd_dev->lock);
1319
1320 if (ret < 0) {
1321 if (ret == -ENOMEM)
1322 res = BLK_STS_RESOURCE;
1323 else
1324 res = BLK_STS_DEV_RESOURCE;
1325 }
1326
1327 return res;
1328}
1329
1330static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1331{
1332 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1333
1334 geo->heads = 128;
1335 geo->sectors = 32;
1336 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1337 return 0;
1338}
1339
1340static int ubd_ioctl(struct block_device *bdev, blk_mode_t mode,
1341 unsigned int cmd, unsigned long arg)
1342{
1343 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1344 u16 ubd_id[ATA_ID_WORDS];
1345
1346 switch (cmd) {
1347 struct cdrom_volctrl volume;
1348 case HDIO_GET_IDENTITY:
1349 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1350 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1351 ubd_id[ATA_ID_HEADS] = 128;
1352 ubd_id[ATA_ID_SECTORS] = 32;
1353 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1354 sizeof(ubd_id)))
1355 return -EFAULT;
1356 return 0;
1357
1358 case CDROMVOLREAD:
1359 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1360 return -EFAULT;
1361 volume.channel0 = 255;
1362 volume.channel1 = 255;
1363 volume.channel2 = 255;
1364 volume.channel3 = 255;
1365 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1366 return -EFAULT;
1367 return 0;
1368 }
1369 return -EINVAL;
1370}
1371
1372static int map_error(int error_code)
1373{
1374 switch (error_code) {
1375 case 0:
1376 return BLK_STS_OK;
1377 case ENOSYS:
1378 case EOPNOTSUPP:
1379 return BLK_STS_NOTSUPP;
1380 case ENOSPC:
1381 return BLK_STS_NOSPC;
1382 }
1383 return BLK_STS_IOERR;
1384}
1385
1386/*
1387 * Everything from here onwards *IS NOT PART OF THE KERNEL*
1388 *
1389 * The following functions are part of UML hypervisor code.
1390 * All functions from here onwards are executed as a helper
1391 * thread and are not allowed to execute any kernel functions.
1392 *
1393 * Any communication must occur strictly via shared memory and IPC.
1394 *
1395 * Do not add printks, locks, kernel memory operations, etc - it
1396 * will result in unpredictable behaviour and/or crashes.
1397 */
1398
1399static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1400{
1401 int n;
1402
1403 if (segment->cow_offset == -1)
1404 return map_error(0);
1405
1406 n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1407 sizeof(segment->bitmap_words), segment->cow_offset);
1408 if (n != sizeof(segment->bitmap_words))
1409 return map_error(-n);
1410
1411 return map_error(0);
1412}
1413
1414static void do_io(struct io_thread_req *req, struct io_desc *desc)
1415{
1416 char *buf = NULL;
1417 unsigned long len;
1418 int n, nsectors, start, end, bit;
1419 __u64 off;
1420
1421 /* FLUSH is really a special case, we cannot "case" it with others */
1422
1423 if (req_op(req->req) == REQ_OP_FLUSH) {
1424 /* fds[0] is always either the rw image or our cow file */
1425 req->error = map_error(-os_sync_file(req->fds[0]));
1426 return;
1427 }
1428
1429 nsectors = desc->length / req->sectorsize;
1430 start = 0;
1431 do {
1432 bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1433 end = start;
1434 while((end < nsectors) &&
1435 (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1436 end++;
1437
1438 off = req->offset + req->offsets[bit] +
1439 start * req->sectorsize;
1440 len = (end - start) * req->sectorsize;
1441 if (desc->buffer != NULL)
1442 buf = &desc->buffer[start * req->sectorsize];
1443
1444 switch (req_op(req->req)) {
1445 case REQ_OP_READ:
1446 n = 0;
1447 do {
1448 buf = &buf[n];
1449 len -= n;
1450 n = os_pread_file(req->fds[bit], buf, len, off);
1451 if (n < 0) {
1452 req->error = map_error(-n);
1453 return;
1454 }
1455 } while((n < len) && (n != 0));
1456 if (n < len) memset(&buf[n], 0, len - n);
1457 break;
1458 case REQ_OP_WRITE:
1459 n = os_pwrite_file(req->fds[bit], buf, len, off);
1460 if(n != len){
1461 req->error = map_error(-n);
1462 return;
1463 }
1464 break;
1465 case REQ_OP_DISCARD:
1466 n = os_falloc_punch(req->fds[bit], off, len);
1467 if (n) {
1468 req->error = map_error(-n);
1469 return;
1470 }
1471 break;
1472 case REQ_OP_WRITE_ZEROES:
1473 n = os_falloc_zeroes(req->fds[bit], off, len);
1474 if (n) {
1475 req->error = map_error(-n);
1476 return;
1477 }
1478 break;
1479 default:
1480 WARN_ON_ONCE(1);
1481 req->error = BLK_STS_NOTSUPP;
1482 return;
1483 }
1484
1485 start = end;
1486 } while(start < nsectors);
1487
1488 req->offset += len;
1489 req->error = update_bitmap(req, desc);
1490}
1491
1492/* Changed in start_io_thread, which is serialized by being called only
1493 * from ubd_init, which is an initcall.
1494 */
1495int kernel_fd = -1;
1496
1497/* Only changed by the io thread. XXX: currently unused. */
1498static int io_count;
1499
1500int io_thread(void *arg)
1501{
1502 int n, count, written, res;
1503
1504 os_set_pdeathsig();
1505 os_fix_helper_signals();
1506
1507 while(1){
1508 n = bulk_req_safe_read(
1509 kernel_fd,
1510 io_req_buffer,
1511 &io_remainder,
1512 &io_remainder_size,
1513 UBD_REQ_BUFFER_SIZE
1514 );
1515 if (n <= 0) {
1516 if (n == -EAGAIN)
1517 ubd_read_poll(-1);
1518
1519 continue;
1520 }
1521
1522 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1523 struct io_thread_req *req = (*io_req_buffer)[count];
1524 int i;
1525
1526 io_count++;
1527 for (i = 0; !req->error && i < req->desc_cnt; i++)
1528 do_io(req, &(req->io_desc[i]));
1529
1530 }
1531
1532 written = 0;
1533
1534 do {
1535 res = os_write_file(kernel_fd,
1536 ((char *) io_req_buffer) + written,
1537 n - written);
1538 if (res >= 0) {
1539 written += res;
1540 }
1541 if (written < n) {
1542 ubd_write_poll(-1);
1543 }
1544 } while (written < n);
1545 }
1546
1547 return 0;
1548}