[PATCH] uml ubd driver: ubd_io_lock usage fixup
[safe/jmp/linux-2.6] / arch / um / drivers / ubd_kern.c
1 /*
2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 /* 2001-09-28...2002-04-17
7  * Partition stuff by James_McMechan@hotmail.com
8  * old style ubd by setting UBD_SHIFT to 0
9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
10  * partitions have changed in 2.5
11  * 2003-01-29 more tinkering for 2.5.59-1
12  * This should now address the sysfs problems and has
13  * the symlink for devfs to allow for booting with
14  * the common /dev/ubd/discX/... names rather than
15  * only /dev/ubdN/discN this version also has lots of
16  * clean ups preparing for ubd-many.
17  * James McMechan
18  */
19
20 #define MAJOR_NR UBD_MAJOR
21 #define UBD_SHIFT 4
22
23 #include "linux/module.h"
24 #include "linux/blkdev.h"
25 #include "linux/hdreg.h"
26 #include "linux/init.h"
27 #include "linux/cdrom.h"
28 #include "linux/proc_fs.h"
29 #include "linux/ctype.h"
30 #include "linux/capability.h"
31 #include "linux/mm.h"
32 #include "linux/vmalloc.h"
33 #include "linux/blkpg.h"
34 #include "linux/genhd.h"
35 #include "linux/spinlock.h"
36 #include "linux/platform_device.h"
37 #include "asm/segment.h"
38 #include "asm/uaccess.h"
39 #include "asm/irq.h"
40 #include "asm/types.h"
41 #include "asm/tlbflush.h"
42 #include "user_util.h"
43 #include "mem_user.h"
44 #include "kern_util.h"
45 #include "kern.h"
46 #include "mconsole_kern.h"
47 #include "init.h"
48 #include "irq_user.h"
49 #include "irq_kern.h"
50 #include "ubd_user.h"
51 #include "os.h"
52 #include "mem.h"
53 #include "mem_kern.h"
54 #include "cow.h"
55
56 enum ubd_req { UBD_READ, UBD_WRITE };
57
58 struct io_thread_req {
59         enum ubd_req op;
60         int fds[2];
61         unsigned long offsets[2];
62         unsigned long long offset;
63         unsigned long length;
64         char *buffer;
65         int sectorsize;
66         unsigned long sector_mask;
67         unsigned long long cow_offset;
68         unsigned long bitmap_words[2];
69         int error;
70 };
71
72 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
73                          char **backing_file_out, int *bitmap_offset_out,
74                          unsigned long *bitmap_len_out, int *data_offset_out,
75                          int *create_cow_out);
76 extern int create_cow_file(char *cow_file, char *backing_file,
77                            struct openflags flags, int sectorsize,
78                            int alignment, int *bitmap_offset_out,
79                            unsigned long *bitmap_len_out,
80                            int *data_offset_out);
81 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
82 extern void do_io(struct io_thread_req *req);
83
84 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
85 {
86         __u64 n;
87         int bits, off;
88
89         bits = sizeof(data[0]) * 8;
90         n = bit / bits;
91         off = bit % bits;
92         return((data[n] & (1 << off)) != 0);
93 }
94
95 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
96 {
97         __u64 n;
98         int bits, off;
99
100         bits = sizeof(data[0]) * 8;
101         n = bit / bits;
102         off = bit % bits;
103         data[n] |= (1 << off);
104 }
105 /*End stuff from ubd_user.h*/
106
107 #define DRIVER_NAME "uml-blkdev"
108
109 /* Can be taken in interrupt context, and is passed to the block layer to lock
110  * the request queue. Kernel side code knows that. */
111 static DEFINE_SPINLOCK(ubd_io_lock);
112
113 static DEFINE_MUTEX(ubd_lock);
114
115 static void (*do_ubd)(void);
116
117 static int ubd_open(struct inode * inode, struct file * filp);
118 static int ubd_release(struct inode * inode, struct file * file);
119 static int ubd_ioctl(struct inode * inode, struct file * file,
120                      unsigned int cmd, unsigned long arg);
121 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
122
123 #define MAX_DEV (16)
124
125 static struct block_device_operations ubd_blops = {
126         .owner          = THIS_MODULE,
127         .open           = ubd_open,
128         .release        = ubd_release,
129         .ioctl          = ubd_ioctl,
130         .getgeo         = ubd_getgeo,
131 };
132
133 /* Protected by the queue_lock */
134 static request_queue_t *ubd_queue;
135
136 /* Protected by ubd_lock */
137 static int fake_major = MAJOR_NR;
138
139 static struct gendisk *ubd_gendisk[MAX_DEV];
140 static struct gendisk *fake_gendisk[MAX_DEV];
141
142 #ifdef CONFIG_BLK_DEV_UBD_SYNC
143 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
144                                          .cl = 1 })
145 #else
146 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
147                                          .cl = 1 })
148 #endif
149
150 /* Not protected - changed only in ubd_setup_common and then only to
151  * to enable O_SYNC.
152  */
153 static struct openflags global_openflags = OPEN_FLAGS;
154
155 struct cow {
156         /* backing file name */
157         char *file;
158         /* backing file fd */
159         int fd;
160         unsigned long *bitmap;
161         unsigned long bitmap_len;
162         int bitmap_offset;
163         int data_offset;
164 };
165
166 struct ubd {
167         /* name (and fd, below) of the file opened for writing, either the
168          * backing or the cow file. */
169         char *file;
170         int count;
171         int fd;
172         __u64 size;
173         struct openflags boot_openflags;
174         struct openflags openflags;
175         int shared;
176         int no_cow;
177         struct cow cow;
178         struct platform_device pdev;
179 };
180
181 #define DEFAULT_COW { \
182         .file =                 NULL, \
183         .fd =                   -1, \
184         .bitmap =               NULL, \
185         .bitmap_offset =        0, \
186         .data_offset =          0, \
187 }
188
189 #define DEFAULT_UBD { \
190         .file =                 NULL, \
191         .count =                0, \
192         .fd =                   -1, \
193         .size =                 -1, \
194         .boot_openflags =       OPEN_FLAGS, \
195         .openflags =            OPEN_FLAGS, \
196         .no_cow =               0, \
197         .shared =               0, \
198         .cow =                  DEFAULT_COW, \
199 }
200
201 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
202
203 static int ubd0_init(void)
204 {
205         struct ubd *ubd_dev = &ubd_devs[0];
206
207         if(ubd_dev->file == NULL)
208                 ubd_dev->file = "root_fs";
209         return(0);
210 }
211
212 __initcall(ubd0_init);
213
214 /* Only changed by fake_ide_setup which is a setup */
215 static int fake_ide = 0;
216 static struct proc_dir_entry *proc_ide_root = NULL;
217 static struct proc_dir_entry *proc_ide = NULL;
218
219 static void make_proc_ide(void)
220 {
221         proc_ide_root = proc_mkdir("ide", NULL);
222         proc_ide = proc_mkdir("ide0", proc_ide_root);
223 }
224
225 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
226                                int *eof, void *data)
227 {
228         int len;
229
230         strcpy(page, "disk\n");
231         len = strlen("disk\n");
232         len -= off;
233         if (len < count){
234                 *eof = 1;
235                 if (len <= 0) return 0;
236         }
237         else len = count;
238         *start = page + off;
239         return len;
240 }
241
242 static void make_ide_entries(char *dev_name)
243 {
244         struct proc_dir_entry *dir, *ent;
245         char name[64];
246
247         if(proc_ide_root == NULL) make_proc_ide();
248
249         dir = proc_mkdir(dev_name, proc_ide);
250         if(!dir) return;
251
252         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
253         if(!ent) return;
254         ent->nlink = 1;
255         ent->data = NULL;
256         ent->read_proc = proc_ide_read_media;
257         ent->write_proc = NULL;
258         sprintf(name,"ide0/%s", dev_name);
259         proc_symlink(dev_name, proc_ide_root, name);
260 }
261
262 static int fake_ide_setup(char *str)
263 {
264         fake_ide = 1;
265         return(1);
266 }
267
268 __setup("fake_ide", fake_ide_setup);
269
270 __uml_help(fake_ide_setup,
271 "fake_ide\n"
272 "    Create ide0 entries that map onto ubd devices.\n\n"
273 );
274
275 static int parse_unit(char **ptr)
276 {
277         char *str = *ptr, *end;
278         int n = -1;
279
280         if(isdigit(*str)) {
281                 n = simple_strtoul(str, &end, 0);
282                 if(end == str)
283                         return(-1);
284                 *ptr = end;
285         }
286         else if (('a' <= *str) && (*str <= 'z')) {
287                 n = *str - 'a';
288                 str++;
289                 *ptr = str;
290         }
291         return(n);
292 }
293
294 static int ubd_setup_common(char *str, int *index_out)
295 {
296         struct ubd *ubd_dev;
297         struct openflags flags = global_openflags;
298         char *backing_file;
299         int n, err, i;
300
301         if(index_out) *index_out = -1;
302         n = *str;
303         if(n == '='){
304                 char *end;
305                 int major;
306
307                 str++;
308                 if(!strcmp(str, "sync")){
309                         global_openflags = of_sync(global_openflags);
310                         return(0);
311                 }
312                 major = simple_strtoul(str, &end, 0);
313                 if((*end != '\0') || (end == str)){
314                         printk(KERN_ERR
315                                "ubd_setup : didn't parse major number\n");
316                         return(1);
317                 }
318
319                 err = 1;
320                 mutex_lock(&ubd_lock);
321                 if(fake_major != MAJOR_NR){
322                         printk(KERN_ERR "Can't assign a fake major twice\n");
323                         goto out1;
324                 }
325
326                 fake_major = major;
327
328                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
329                        major);
330                 err = 0;
331         out1:
332                 mutex_unlock(&ubd_lock);
333                 return(err);
334         }
335
336         n = parse_unit(&str);
337         if(n < 0){
338                 printk(KERN_ERR "ubd_setup : couldn't parse unit number "
339                        "'%s'\n", str);
340                 return(1);
341         }
342         if(n >= MAX_DEV){
343                 printk(KERN_ERR "ubd_setup : index %d out of range "
344                        "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
345                 return(1);
346         }
347
348         err = 1;
349         mutex_lock(&ubd_lock);
350
351         ubd_dev = &ubd_devs[n];
352         if(ubd_dev->file != NULL){
353                 printk(KERN_ERR "ubd_setup : device already configured\n");
354                 goto out;
355         }
356
357         if (index_out)
358                 *index_out = n;
359
360         for (i = 0; i < sizeof("rscd="); i++) {
361                 switch (*str) {
362                 case 'r':
363                         flags.w = 0;
364                         break;
365                 case 's':
366                         flags.s = 1;
367                         break;
368                 case 'd':
369                         ubd_dev->no_cow = 1;
370                         break;
371                 case 'c':
372                         ubd_dev->shared = 1;
373                         break;
374                 case '=':
375                         str++;
376                         goto break_loop;
377                 default:
378                         printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r, s, c, or d)\n");
379                         goto out;
380                 }
381                 str++;
382         }
383
384         if (*str == '=')
385                 printk(KERN_ERR "ubd_setup : Too many flags specified\n");
386         else
387                 printk(KERN_ERR "ubd_setup : Expected '='\n");
388         goto out;
389
390 break_loop:
391         err = 0;
392         backing_file = strchr(str, ',');
393
394         if (!backing_file) {
395                 backing_file = strchr(str, ':');
396         }
397
398         if(backing_file){
399                 if(ubd_dev->no_cow)
400                         printk(KERN_ERR "Can't specify both 'd' and a "
401                                "cow file\n");
402                 else {
403                         *backing_file = '\0';
404                         backing_file++;
405                 }
406         }
407         ubd_dev->file = str;
408         ubd_dev->cow.file = backing_file;
409         ubd_dev->boot_openflags = flags;
410 out:
411         mutex_unlock(&ubd_lock);
412         return(err);
413 }
414
415 static int ubd_setup(char *str)
416 {
417         ubd_setup_common(str, NULL);
418         return(1);
419 }
420
421 __setup("ubd", ubd_setup);
422 __uml_help(ubd_setup,
423 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
424 "    This is used to associate a device with a file in the underlying\n"
425 "    filesystem. When specifying two filenames, the first one is the\n"
426 "    COW name and the second is the backing file name. As separator you can\n"
427 "    use either a ':' or a ',': the first one allows writing things like;\n"
428 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
429 "    while with a ',' the shell would not expand the 2nd '~'.\n"
430 "    When using only one filename, UML will detect whether to thread it like\n"
431 "    a COW file or a backing file. To override this detection, add the 'd'\n"
432 "    flag:\n"
433 "       ubd0d=BackingFile\n"
434 "    Usually, there is a filesystem in the file, but \n"
435 "    that's not required. Swap devices containing swap files can be\n"
436 "    specified like this. Also, a file which doesn't contain a\n"
437 "    filesystem can have its contents read in the virtual \n"
438 "    machine by running 'dd' on the device. <n> must be in the range\n"
439 "    0 to 7. Appending an 'r' to the number will cause that device\n"
440 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
441 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
442 );
443
444 static int udb_setup(char *str)
445 {
446         printk("udb%s specified on command line is almost certainly a ubd -> "
447                "udb TYPO\n", str);
448         return(1);
449 }
450
451 __setup("udb", udb_setup);
452 __uml_help(udb_setup,
453 "udb\n"
454 "    This option is here solely to catch ubd -> udb typos, which can be\n"
455 "    to impossible to catch visually unless you specifically look for\n"
456 "    them.  The only result of any option starting with 'udb' is an error\n"
457 "    in the boot output.\n\n"
458 );
459
460 static int fakehd_set = 0;
461 static int fakehd(char *str)
462 {
463         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
464         fakehd_set = 1;
465         return 1;
466 }
467
468 __setup("fakehd", fakehd);
469 __uml_help(fakehd,
470 "fakehd\n"
471 "    Change the ubd device name to \"hd\".\n\n"
472 );
473
474 static void do_ubd_request(request_queue_t * q);
475
476 /* Only changed by ubd_init, which is an initcall. */
477 int thread_fd = -1;
478
479 /* Changed by ubd_handler, which is serialized because interrupts only
480  * happen on CPU 0.
481  */
482 int intr_count = 0;
483
484 /* call ubd_finish if you need to serialize */
485 static void __ubd_finish(struct request *req, int error)
486 {
487         int nsect;
488
489         if(error){
490                 end_request(req, 0);
491                 return;
492         }
493         nsect = req->current_nr_sectors;
494         req->sector += nsect;
495         req->buffer += nsect << 9;
496         req->errors = 0;
497         req->nr_sectors -= nsect;
498         req->current_nr_sectors = 0;
499         end_request(req, 1);
500 }
501
502 /* Callable only from interrupt context - otherwise you need to do
503  * spin_lock_irq()/spin_lock_irqsave() */
504 static inline void ubd_finish(struct request *req, int error)
505 {
506         spin_lock(&ubd_io_lock);
507         __ubd_finish(req, error);
508         spin_unlock(&ubd_io_lock);
509 }
510
511 /* Called without ubd_io_lock held, and only in interrupt context. */
512 static void ubd_handler(void)
513 {
514         struct io_thread_req req;
515         struct request *rq = elv_next_request(ubd_queue);
516         int n;
517
518         do_ubd = NULL;
519         intr_count++;
520         n = os_read_file(thread_fd, &req, sizeof(req));
521         if(n != sizeof(req)){
522                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
523                        "err = %d\n", os_getpid(), -n);
524                 spin_lock(&ubd_io_lock);
525                 end_request(rq, 0);
526                 spin_unlock(&ubd_io_lock);
527                 return;
528         }
529
530         ubd_finish(rq, req.error);
531         reactivate_fd(thread_fd, UBD_IRQ);      
532         spin_lock(&ubd_io_lock);
533         do_ubd_request(ubd_queue);
534         spin_unlock(&ubd_io_lock);
535 }
536
537 static irqreturn_t ubd_intr(int irq, void *dev)
538 {
539         ubd_handler();
540         return(IRQ_HANDLED);
541 }
542
543 /* Only changed by ubd_init, which is an initcall. */
544 static int io_pid = -1;
545
546 void kill_io_thread(void)
547 {
548         if(io_pid != -1)
549                 os_kill_process(io_pid, 1);
550 }
551
552 __uml_exitcall(kill_io_thread);
553
554 static int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
555 {
556         char *file;
557
558         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
559         return(os_file_size(file, size_out));
560 }
561
562 static void ubd_close_dev(struct ubd *ubd_dev)
563 {
564         os_close_file(ubd_dev->fd);
565         if(ubd_dev->cow.file == NULL)
566                 return;
567
568         os_close_file(ubd_dev->cow.fd);
569         vfree(ubd_dev->cow.bitmap);
570         ubd_dev->cow.bitmap = NULL;
571 }
572
573 static int ubd_open_dev(struct ubd *ubd_dev)
574 {
575         struct openflags flags;
576         char **back_ptr;
577         int err, create_cow, *create_ptr;
578
579         ubd_dev->openflags = ubd_dev->boot_openflags;
580         create_cow = 0;
581         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
582         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
583         ubd_dev->fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
584                                 back_ptr, &ubd_dev->cow.bitmap_offset,
585                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
586                                 create_ptr);
587
588         if((ubd_dev->fd == -ENOENT) && create_cow){
589                 ubd_dev->fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
590                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
591                                           &ubd_dev->cow.bitmap_offset,
592                                           &ubd_dev->cow.bitmap_len,
593                                           &ubd_dev->cow.data_offset);
594                 if(ubd_dev->fd >= 0){
595                         printk(KERN_INFO "Creating \"%s\" as COW file for "
596                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
597                 }
598         }
599
600         if(ubd_dev->fd < 0){
601                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
602                        -ubd_dev->fd);
603                 return(ubd_dev->fd);
604         }
605
606         if(ubd_dev->cow.file != NULL){
607                 err = -ENOMEM;
608                 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
609                 if(ubd_dev->cow.bitmap == NULL){
610                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
611                         goto error;
612                 }
613                 flush_tlb_kernel_vm();
614
615                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
616                                       ubd_dev->cow.bitmap_offset,
617                                       ubd_dev->cow.bitmap_len);
618                 if(err < 0)
619                         goto error;
620
621                 flags = ubd_dev->openflags;
622                 flags.w = 0;
623                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
624                                     NULL, NULL, NULL, NULL);
625                 if(err < 0) goto error;
626                 ubd_dev->cow.fd = err;
627         }
628         return(0);
629  error:
630         os_close_file(ubd_dev->fd);
631         return(err);
632 }
633
634 static int ubd_disk_register(int major, u64 size, int unit,
635                         struct gendisk **disk_out)
636                         
637 {
638         struct gendisk *disk;
639
640         disk = alloc_disk(1 << UBD_SHIFT);
641         if(disk == NULL)
642                 return(-ENOMEM);
643
644         disk->major = major;
645         disk->first_minor = unit << UBD_SHIFT;
646         disk->fops = &ubd_blops;
647         set_capacity(disk, size / 512);
648         if(major == MAJOR_NR)
649                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
650         else
651                 sprintf(disk->disk_name, "ubd_fake%d", unit);
652
653         /* sysfs register (not for ide fake devices) */
654         if (major == MAJOR_NR) {
655                 ubd_devs[unit].pdev.id   = unit;
656                 ubd_devs[unit].pdev.name = DRIVER_NAME;
657                 platform_device_register(&ubd_devs[unit].pdev);
658                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
659         }
660
661         disk->private_data = &ubd_devs[unit];
662         disk->queue = ubd_queue;
663         add_disk(disk);
664
665         *disk_out = disk;
666         return 0;
667 }
668
669 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
670
671 static int ubd_add(int n)
672 {
673         struct ubd *ubd_dev = &ubd_devs[n];
674         int err;
675
676         err = -ENODEV;
677         if(ubd_dev->file == NULL)
678                 goto out;
679
680         err = ubd_file_size(ubd_dev, &ubd_dev->size);
681         if(err < 0)
682                 goto out;
683
684         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
685
686         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
687         if(err)
688                 goto out;
689
690         if(fake_major != MAJOR_NR)
691                 ubd_disk_register(fake_major, ubd_dev->size, n,
692                              &fake_gendisk[n]);
693
694         /* perhaps this should also be under the "if (fake_major)" above */
695         /* using the fake_disk->disk_name and also the fakehd_set name */
696         if (fake_ide)
697                 make_ide_entries(ubd_gendisk[n]->disk_name);
698
699         err = 0;
700 out:
701         return err;
702 }
703
704 static int ubd_config(char *str)
705 {
706         int n, err;
707
708         str = kstrdup(str, GFP_KERNEL);
709         if(str == NULL){
710                 printk(KERN_ERR "ubd_config failed to strdup string\n");
711                 return(1);
712         }
713         err = ubd_setup_common(str, &n);
714         if(err){
715                 kfree(str);
716                 return(-1);
717         }
718         if(n == -1) return(0);
719
720         mutex_lock(&ubd_lock);
721         err = ubd_add(n);
722         if(err)
723                 ubd_devs[n].file = NULL;
724         mutex_unlock(&ubd_lock);
725
726         return(err);
727 }
728
729 static int ubd_get_config(char *name, char *str, int size, char **error_out)
730 {
731         struct ubd *ubd_dev;
732         int n, len = 0;
733
734         n = parse_unit(&name);
735         if((n >= MAX_DEV) || (n < 0)){
736                 *error_out = "ubd_get_config : device number out of range";
737                 return(-1);
738         }
739
740         ubd_dev = &ubd_devs[n];
741         mutex_lock(&ubd_lock);
742
743         if(ubd_dev->file == NULL){
744                 CONFIG_CHUNK(str, size, len, "", 1);
745                 goto out;
746         }
747
748         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
749
750         if(ubd_dev->cow.file != NULL){
751                 CONFIG_CHUNK(str, size, len, ",", 0);
752                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
753         }
754         else CONFIG_CHUNK(str, size, len, "", 1);
755
756  out:
757         mutex_unlock(&ubd_lock);
758         return(len);
759 }
760
761 static int ubd_id(char **str, int *start_out, int *end_out)
762 {
763         int n;
764
765         n = parse_unit(str);
766         *start_out = 0;
767         *end_out = MAX_DEV - 1;
768         return n;
769 }
770
771 static int ubd_remove(int n)
772 {
773         struct ubd *ubd_dev;
774         int err = -ENODEV;
775
776         mutex_lock(&ubd_lock);
777
778         if(ubd_gendisk[n] == NULL)
779                 goto out;
780
781         ubd_dev = &ubd_devs[n];
782
783         if(ubd_dev->file == NULL)
784                 goto out;
785
786         /* you cannot remove a open disk */
787         err = -EBUSY;
788         if(ubd_dev->count > 0)
789                 goto out;
790
791         del_gendisk(ubd_gendisk[n]);
792         put_disk(ubd_gendisk[n]);
793         ubd_gendisk[n] = NULL;
794
795         if(fake_gendisk[n] != NULL){
796                 del_gendisk(fake_gendisk[n]);
797                 put_disk(fake_gendisk[n]);
798                 fake_gendisk[n] = NULL;
799         }
800
801         platform_device_unregister(&ubd_dev->pdev);
802         *ubd_dev = ((struct ubd) DEFAULT_UBD);
803         err = 0;
804 out:
805         mutex_unlock(&ubd_lock);
806         return err;
807 }
808
809 static struct mc_device ubd_mc = {
810         .name           = "ubd",
811         .config         = ubd_config,
812         .get_config     = ubd_get_config,
813         .id             = ubd_id,
814         .remove         = ubd_remove,
815 };
816
817 static int ubd_mc_init(void)
818 {
819         mconsole_register_dev(&ubd_mc);
820         return 0;
821 }
822
823 __initcall(ubd_mc_init);
824
825 static struct platform_driver ubd_driver = {
826         .driver = {
827                 .name  = DRIVER_NAME,
828         },
829 };
830
831 int ubd_init(void)
832 {
833         int i;
834
835         if (register_blkdev(MAJOR_NR, "ubd"))
836                 return -1;
837
838         ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
839         if (!ubd_queue) {
840                 unregister_blkdev(MAJOR_NR, "ubd");
841                 return -1;
842         }
843                 
844         if (fake_major != MAJOR_NR) {
845                 char name[sizeof("ubd_nnn\0")];
846
847                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
848                 if (register_blkdev(fake_major, "ubd"))
849                         return -1;
850         }
851         platform_driver_register(&ubd_driver);
852         for (i = 0; i < MAX_DEV; i++)
853                 ubd_add(i);
854         return 0;
855 }
856
857 late_initcall(ubd_init);
858
859 int ubd_driver_init(void){
860         unsigned long stack;
861         int err;
862
863         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
864         if(global_openflags.s){
865                 printk(KERN_INFO "ubd: Synchronous mode\n");
866                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
867                  * enough. So use anyway the io thread. */
868         }
869         stack = alloc_stack(0, 0);
870         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
871                                  &thread_fd);
872         if(io_pid < 0){
873                 printk(KERN_ERR
874                        "ubd : Failed to start I/O thread (errno = %d) - "
875                        "falling back to synchronous I/O\n", -io_pid);
876                 io_pid = -1;
877                 return(0);
878         }
879         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
880                              IRQF_DISABLED, "ubd", ubd_devs);
881         if(err != 0)
882                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
883         return 0;
884 }
885
886 device_initcall(ubd_driver_init);
887
888 static int ubd_open(struct inode *inode, struct file *filp)
889 {
890         struct gendisk *disk = inode->i_bdev->bd_disk;
891         struct ubd *ubd_dev = disk->private_data;
892         int err = 0;
893
894         if(ubd_dev->count == 0){
895                 err = ubd_open_dev(ubd_dev);
896                 if(err){
897                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
898                                disk->disk_name, ubd_dev->file, -err);
899                         goto out;
900                 }
901         }
902         ubd_dev->count++;
903         set_disk_ro(disk, !ubd_dev->openflags.w);
904
905         /* This should no more be needed. And it didn't work anyway to exclude
906          * read-write remounting of filesystems.*/
907         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
908                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
909                 err = -EROFS;
910         }*/
911  out:
912         return(err);
913 }
914
915 static int ubd_release(struct inode * inode, struct file * file)
916 {
917         struct gendisk *disk = inode->i_bdev->bd_disk;
918         struct ubd *ubd_dev = disk->private_data;
919
920         if(--ubd_dev->count == 0)
921                 ubd_close_dev(ubd_dev);
922         return(0);
923 }
924
925 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
926                           __u64 *cow_offset, unsigned long *bitmap,
927                           __u64 bitmap_offset, unsigned long *bitmap_words,
928                           __u64 bitmap_len)
929 {
930         __u64 sector = io_offset >> 9;
931         int i, update_bitmap = 0;
932
933         for(i = 0; i < length >> 9; i++){
934                 if(cow_mask != NULL)
935                         ubd_set_bit(i, (unsigned char *) cow_mask);
936                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
937                         continue;
938
939                 update_bitmap = 1;
940                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
941         }
942
943         if(!update_bitmap)
944                 return;
945
946         *cow_offset = sector / (sizeof(unsigned long) * 8);
947
948         /* This takes care of the case where we're exactly at the end of the
949          * device, and *cow_offset + 1 is off the end.  So, just back it up
950          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
951          * for the original diagnosis.
952          */
953         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
954                            sizeof(unsigned long) - 1))
955                 (*cow_offset)--;
956
957         bitmap_words[0] = bitmap[*cow_offset];
958         bitmap_words[1] = bitmap[*cow_offset + 1];
959
960         *cow_offset *= sizeof(unsigned long);
961         *cow_offset += bitmap_offset;
962 }
963
964 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
965                        __u64 bitmap_offset, __u64 bitmap_len)
966 {
967         __u64 sector = req->offset >> 9;
968         int i;
969
970         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
971                 panic("Operation too long");
972
973         if(req->op == UBD_READ) {
974                 for(i = 0; i < req->length >> 9; i++){
975                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
976                                 ubd_set_bit(i, (unsigned char *)
977                                             &req->sector_mask);
978                 }
979         }
980         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
981                            &req->cow_offset, bitmap, bitmap_offset,
982                            req->bitmap_words, bitmap_len);
983 }
984
985 /* Called with ubd_io_lock held */
986 static int prepare_request(struct request *req, struct io_thread_req *io_req)
987 {
988         struct gendisk *disk = req->rq_disk;
989         struct ubd *ubd_dev = disk->private_data;
990         __u64 offset;
991         int len;
992
993         /* This should be impossible now */
994         if((rq_data_dir(req) == WRITE) && !ubd_dev->openflags.w){
995                 printk("Write attempted on readonly ubd device %s\n",
996                        disk->disk_name);
997                 end_request(req, 0);
998                 return(1);
999         }
1000
1001         offset = ((__u64) req->sector) << 9;
1002         len = req->current_nr_sectors << 9;
1003
1004         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd : ubd_dev->fd;
1005         io_req->fds[1] = ubd_dev->fd;
1006         io_req->cow_offset = -1;
1007         io_req->offset = offset;
1008         io_req->length = len;
1009         io_req->error = 0;
1010         io_req->sector_mask = 0;
1011
1012         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1013         io_req->offsets[0] = 0;
1014         io_req->offsets[1] = ubd_dev->cow.data_offset;
1015         io_req->buffer = req->buffer;
1016         io_req->sectorsize = 1 << 9;
1017
1018         if(ubd_dev->cow.file != NULL)
1019                 cowify_req(io_req, ubd_dev->cow.bitmap, ubd_dev->cow.bitmap_offset,
1020                            ubd_dev->cow.bitmap_len);
1021
1022         return(0);
1023 }
1024
1025 /* Called with ubd_io_lock held */
1026 static void do_ubd_request(request_queue_t *q)
1027 {
1028         struct io_thread_req io_req;
1029         struct request *req;
1030         int err, n;
1031
1032         if(thread_fd == -1){
1033                 while((req = elv_next_request(q)) != NULL){
1034                         err = prepare_request(req, &io_req);
1035                         if(!err){
1036                                 do_io(&io_req);
1037                                 __ubd_finish(req, io_req.error);
1038                         }
1039                 }
1040         }
1041         else {
1042                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1043                         return;
1044                 err = prepare_request(req, &io_req);
1045                 if(!err){
1046                         do_ubd = ubd_handler;
1047                         n = os_write_file(thread_fd, (char *) &io_req,
1048                                          sizeof(io_req));
1049                         if(n != sizeof(io_req))
1050                                 printk("write to io thread failed, "
1051                                        "errno = %d\n", -n);
1052                 }
1053         }
1054 }
1055
1056 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1057 {
1058         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1059
1060         geo->heads = 128;
1061         geo->sectors = 32;
1062         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1063         return 0;
1064 }
1065
1066 static int ubd_ioctl(struct inode * inode, struct file * file,
1067                      unsigned int cmd, unsigned long arg)
1068 {
1069         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1070         struct hd_driveid ubd_id = {
1071                 .cyls           = 0,
1072                 .heads          = 128,
1073                 .sectors        = 32,
1074         };
1075
1076         switch (cmd) {
1077                 struct cdrom_volctrl volume;
1078         case HDIO_GET_IDENTITY:
1079                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1080                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1081                                  sizeof(ubd_id)))
1082                         return(-EFAULT);
1083                 return(0);
1084                 
1085         case CDROMVOLREAD:
1086                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1087                         return(-EFAULT);
1088                 volume.channel0 = 255;
1089                 volume.channel1 = 255;
1090                 volume.channel2 = 255;
1091                 volume.channel3 = 255;
1092                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1093                         return(-EFAULT);
1094                 return(0);
1095         }
1096         return(-EINVAL);
1097 }
1098
1099 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1100 {
1101         struct uml_stat buf1, buf2;
1102         int err;
1103
1104         if(from_cmdline == NULL)
1105                 return 0;
1106         if(!strcmp(from_cmdline, from_cow))
1107                 return 0;
1108
1109         err = os_stat_file(from_cmdline, &buf1);
1110         if(err < 0){
1111                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1112                 return 0;
1113         }
1114         err = os_stat_file(from_cow, &buf2);
1115         if(err < 0){
1116                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1117                 return 1;
1118         }
1119         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1120                 return 0;
1121
1122         printk("Backing file mismatch - \"%s\" requested,\n"
1123                "\"%s\" specified in COW header of \"%s\"\n",
1124                from_cmdline, from_cow, cow);
1125         return 1;
1126 }
1127
1128 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1129 {
1130         unsigned long modtime;
1131         unsigned long long actual;
1132         int err;
1133
1134         err = os_file_modtime(file, &modtime);
1135         if(err < 0){
1136                 printk("Failed to get modification time of backing file "
1137                        "\"%s\", err = %d\n", file, -err);
1138                 return(err);
1139         }
1140
1141         err = os_file_size(file, &actual);
1142         if(err < 0){
1143                 printk("Failed to get size of backing file \"%s\", "
1144                        "err = %d\n", file, -err);
1145                 return(err);
1146         }
1147
1148         if(actual != size){
1149                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1150                  * the typecast.*/
1151                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1152                        "file\n", (unsigned long long) size, actual);
1153                 return(-EINVAL);
1154         }
1155         if(modtime != mtime){
1156                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1157                        "file\n", mtime, modtime);
1158                 return(-EINVAL);
1159         }
1160         return(0);
1161 }
1162
1163 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1164 {
1165         int err;
1166
1167         err = os_seek_file(fd, offset);
1168         if(err < 0)
1169                 return(err);
1170
1171         err = os_read_file(fd, buf, len);
1172         if(err < 0)
1173                 return(err);
1174
1175         return(0);
1176 }
1177
1178 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1179                   char **backing_file_out, int *bitmap_offset_out,
1180                   unsigned long *bitmap_len_out, int *data_offset_out,
1181                   int *create_cow_out)
1182 {
1183         time_t mtime;
1184         unsigned long long size;
1185         __u32 version, align;
1186         char *backing_file;
1187         int fd, err, sectorsize, asked_switch, mode = 0644;
1188
1189         fd = os_open_file(file, *openflags, mode);
1190         if (fd < 0) {
1191                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1192                         *create_cow_out = 1;
1193                 if (!openflags->w ||
1194                    ((fd != -EROFS) && (fd != -EACCES)))
1195                         return fd;
1196                 openflags->w = 0;
1197                 fd = os_open_file(file, *openflags, mode);
1198                 if (fd < 0)
1199                         return fd;
1200         }
1201
1202         if(shared)
1203                 printk("Not locking \"%s\" on the host\n", file);
1204         else {
1205                 err = os_lock_file(fd, openflags->w);
1206                 if(err < 0){
1207                         printk("Failed to lock '%s', err = %d\n", file, -err);
1208                         goto out_close;
1209                 }
1210         }
1211
1212         /* Successful return case! */
1213         if(backing_file_out == NULL)
1214                 return(fd);
1215
1216         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1217                               &size, &sectorsize, &align, bitmap_offset_out);
1218         if(err && (*backing_file_out != NULL)){
1219                 printk("Failed to read COW header from COW file \"%s\", "
1220                        "errno = %d\n", file, -err);
1221                 goto out_close;
1222         }
1223         if(err)
1224                 return(fd);
1225
1226         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1227
1228         /* Allow switching only if no mismatch. */
1229         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1230                 printk("Switching backing file to '%s'\n", *backing_file_out);
1231                 err = write_cow_header(file, fd, *backing_file_out,
1232                                        sectorsize, align, &size);
1233                 if (err) {
1234                         printk("Switch failed, errno = %d\n", -err);
1235                         goto out_close;
1236                 }
1237         } else {
1238                 *backing_file_out = backing_file;
1239                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1240                 if (err)
1241                         goto out_close;
1242         }
1243
1244         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1245                   bitmap_len_out, data_offset_out);
1246
1247         return fd;
1248  out_close:
1249         os_close_file(fd);
1250         return err;
1251 }
1252
1253 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1254                     int sectorsize, int alignment, int *bitmap_offset_out,
1255                     unsigned long *bitmap_len_out, int *data_offset_out)
1256 {
1257         int err, fd;
1258
1259         flags.c = 1;
1260         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1261         if(fd < 0){
1262                 err = fd;
1263                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1264                        -err);
1265                 goto out;
1266         }
1267
1268         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1269                             bitmap_offset_out, bitmap_len_out,
1270                             data_offset_out);
1271         if(!err)
1272                 return(fd);
1273         os_close_file(fd);
1274  out:
1275         return(err);
1276 }
1277
1278 static int update_bitmap(struct io_thread_req *req)
1279 {
1280         int n;
1281
1282         if(req->cow_offset == -1)
1283                 return(0);
1284
1285         n = os_seek_file(req->fds[1], req->cow_offset);
1286         if(n < 0){
1287                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1288                 return(1);
1289         }
1290
1291         n = os_write_file(req->fds[1], &req->bitmap_words,
1292                           sizeof(req->bitmap_words));
1293         if(n != sizeof(req->bitmap_words)){
1294                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1295                        req->fds[1]);
1296                 return(1);
1297         }
1298
1299         return(0);
1300 }
1301
1302 void do_io(struct io_thread_req *req)
1303 {
1304         char *buf;
1305         unsigned long len;
1306         int n, nsectors, start, end, bit;
1307         int err;
1308         __u64 off;
1309
1310         nsectors = req->length / req->sectorsize;
1311         start = 0;
1312         do {
1313                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1314                 end = start;
1315                 while((end < nsectors) &&
1316                       (ubd_test_bit(end, (unsigned char *)
1317                                     &req->sector_mask) == bit))
1318                         end++;
1319
1320                 off = req->offset + req->offsets[bit] +
1321                         start * req->sectorsize;
1322                 len = (end - start) * req->sectorsize;
1323                 buf = &req->buffer[start * req->sectorsize];
1324
1325                 err = os_seek_file(req->fds[bit], off);
1326                 if(err < 0){
1327                         printk("do_io - lseek failed : err = %d\n", -err);
1328                         req->error = 1;
1329                         return;
1330                 }
1331                 if(req->op == UBD_READ){
1332                         n = 0;
1333                         do {
1334                                 buf = &buf[n];
1335                                 len -= n;
1336                                 n = os_read_file(req->fds[bit], buf, len);
1337                                 if (n < 0) {
1338                                         printk("do_io - read failed, err = %d "
1339                                                "fd = %d\n", -n, req->fds[bit]);
1340                                         req->error = 1;
1341                                         return;
1342                                 }
1343                         } while((n < len) && (n != 0));
1344                         if (n < len) memset(&buf[n], 0, len - n);
1345                 } else {
1346                         n = os_write_file(req->fds[bit], buf, len);
1347                         if(n != len){
1348                                 printk("do_io - write failed err = %d "
1349                                        "fd = %d\n", -n, req->fds[bit]);
1350                                 req->error = 1;
1351                                 return;
1352                         }
1353                 }
1354
1355                 start = end;
1356         } while(start < nsectors);
1357
1358         req->error = update_bitmap(req);
1359 }
1360
1361 /* Changed in start_io_thread, which is serialized by being called only
1362  * from ubd_init, which is an initcall.
1363  */
1364 int kernel_fd = -1;
1365
1366 /* Only changed by the io thread */
1367 int io_count = 0;
1368
1369 int io_thread(void *arg)
1370 {
1371         struct io_thread_req req;
1372         int n;
1373
1374         ignore_sigwinch_sig();
1375         while(1){
1376                 n = os_read_file(kernel_fd, &req, sizeof(req));
1377                 if(n != sizeof(req)){
1378                         if(n < 0)
1379                                 printk("io_thread - read failed, fd = %d, "
1380                                        "err = %d\n", kernel_fd, -n);
1381                         else {
1382                                 printk("io_thread - short read, fd = %d, "
1383                                        "length = %d\n", kernel_fd, n);
1384                         }
1385                         continue;
1386                 }
1387                 io_count++;
1388                 do_io(&req);
1389                 n = os_write_file(kernel_fd, &req, sizeof(req));
1390                 if(n != sizeof(req))
1391                         printk("io_thread - write failed, fd = %d, err = %d\n",
1392                                kernel_fd, -n);
1393         }
1394
1395         return 0;
1396 }