[PATCH] uml ubd driver: do not store error codes as ->fd
[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 /* XXX - this made sense in 2.4 days, now it's only used as a boolean, and
116  * probably it doesn't make sense even for that. */
117 static int do_ubd;
118
119 static int ubd_open(struct inode * inode, struct file * filp);
120 static int ubd_release(struct inode * inode, struct file * file);
121 static int ubd_ioctl(struct inode * inode, struct file * file,
122                      unsigned int cmd, unsigned long arg);
123 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
124
125 #define MAX_DEV (16)
126
127 static struct block_device_operations ubd_blops = {
128         .owner          = THIS_MODULE,
129         .open           = ubd_open,
130         .release        = ubd_release,
131         .ioctl          = ubd_ioctl,
132         .getgeo         = ubd_getgeo,
133 };
134
135 /* Protected by the queue_lock */
136 static request_queue_t *ubd_queue;
137
138 /* Protected by ubd_lock */
139 static int fake_major = MAJOR_NR;
140
141 static struct gendisk *ubd_gendisk[MAX_DEV];
142 static struct gendisk *fake_gendisk[MAX_DEV];
143
144 #ifdef CONFIG_BLK_DEV_UBD_SYNC
145 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
146                                          .cl = 1 })
147 #else
148 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
149                                          .cl = 1 })
150 #endif
151
152 /* Not protected - changed only in ubd_setup_common and then only to
153  * to enable O_SYNC.
154  */
155 static struct openflags global_openflags = OPEN_FLAGS;
156
157 struct cow {
158         /* backing file name */
159         char *file;
160         /* backing file fd */
161         int fd;
162         unsigned long *bitmap;
163         unsigned long bitmap_len;
164         int bitmap_offset;
165         int data_offset;
166 };
167
168 struct ubd {
169         /* name (and fd, below) of the file opened for writing, either the
170          * backing or the cow file. */
171         char *file;
172         int count;
173         int fd;
174         __u64 size;
175         struct openflags boot_openflags;
176         struct openflags openflags;
177         unsigned shared:1;
178         unsigned no_cow:1;
179         struct cow cow;
180         struct platform_device pdev;
181 };
182
183 #define DEFAULT_COW { \
184         .file =                 NULL, \
185         .fd =                   -1, \
186         .bitmap =               NULL, \
187         .bitmap_offset =        0, \
188         .data_offset =          0, \
189 }
190
191 #define DEFAULT_UBD { \
192         .file =                 NULL, \
193         .count =                0, \
194         .fd =                   -1, \
195         .size =                 -1, \
196         .boot_openflags =       OPEN_FLAGS, \
197         .openflags =            OPEN_FLAGS, \
198         .no_cow =               0, \
199         .shared =               0, \
200         .cow =                  DEFAULT_COW, \
201 }
202
203 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
204
205 static int ubd0_init(void)
206 {
207         struct ubd *ubd_dev = &ubd_devs[0];
208
209         if(ubd_dev->file == NULL)
210                 ubd_dev->file = "root_fs";
211         return(0);
212 }
213
214 __initcall(ubd0_init);
215
216 /* Only changed by fake_ide_setup which is a setup */
217 static int fake_ide = 0;
218 static struct proc_dir_entry *proc_ide_root = NULL;
219 static struct proc_dir_entry *proc_ide = NULL;
220
221 static void make_proc_ide(void)
222 {
223         proc_ide_root = proc_mkdir("ide", NULL);
224         proc_ide = proc_mkdir("ide0", proc_ide_root);
225 }
226
227 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
228                                int *eof, void *data)
229 {
230         int len;
231
232         strcpy(page, "disk\n");
233         len = strlen("disk\n");
234         len -= off;
235         if (len < count){
236                 *eof = 1;
237                 if (len <= 0) return 0;
238         }
239         else len = count;
240         *start = page + off;
241         return len;
242 }
243
244 static void make_ide_entries(char *dev_name)
245 {
246         struct proc_dir_entry *dir, *ent;
247         char name[64];
248
249         if(proc_ide_root == NULL) make_proc_ide();
250
251         dir = proc_mkdir(dev_name, proc_ide);
252         if(!dir) return;
253
254         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
255         if(!ent) return;
256         ent->nlink = 1;
257         ent->data = NULL;
258         ent->read_proc = proc_ide_read_media;
259         ent->write_proc = NULL;
260         sprintf(name,"ide0/%s", dev_name);
261         proc_symlink(dev_name, proc_ide_root, name);
262 }
263
264 static int fake_ide_setup(char *str)
265 {
266         fake_ide = 1;
267         return(1);
268 }
269
270 __setup("fake_ide", fake_ide_setup);
271
272 __uml_help(fake_ide_setup,
273 "fake_ide\n"
274 "    Create ide0 entries that map onto ubd devices.\n\n"
275 );
276
277 static int parse_unit(char **ptr)
278 {
279         char *str = *ptr, *end;
280         int n = -1;
281
282         if(isdigit(*str)) {
283                 n = simple_strtoul(str, &end, 0);
284                 if(end == str)
285                         return(-1);
286                 *ptr = end;
287         }
288         else if (('a' <= *str) && (*str <= 'z')) {
289                 n = *str - 'a';
290                 str++;
291                 *ptr = str;
292         }
293         return(n);
294 }
295
296 static int ubd_setup_common(char *str, int *index_out)
297 {
298         struct ubd *ubd_dev;
299         struct openflags flags = global_openflags;
300         char *backing_file;
301         int n, err, i;
302
303         if(index_out) *index_out = -1;
304         n = *str;
305         if(n == '='){
306                 char *end;
307                 int major;
308
309                 str++;
310                 if(!strcmp(str, "sync")){
311                         global_openflags = of_sync(global_openflags);
312                         return(0);
313                 }
314                 major = simple_strtoul(str, &end, 0);
315                 if((*end != '\0') || (end == str)){
316                         printk(KERN_ERR
317                                "ubd_setup : didn't parse major number\n");
318                         return(1);
319                 }
320
321                 err = 1;
322                 mutex_lock(&ubd_lock);
323                 if(fake_major != MAJOR_NR){
324                         printk(KERN_ERR "Can't assign a fake major twice\n");
325                         goto out1;
326                 }
327
328                 fake_major = major;
329
330                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
331                        major);
332                 err = 0;
333         out1:
334                 mutex_unlock(&ubd_lock);
335                 return(err);
336         }
337
338         n = parse_unit(&str);
339         if(n < 0){
340                 printk(KERN_ERR "ubd_setup : couldn't parse unit number "
341                        "'%s'\n", str);
342                 return(1);
343         }
344         if(n >= MAX_DEV){
345                 printk(KERN_ERR "ubd_setup : index %d out of range "
346                        "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
347                 return(1);
348         }
349
350         err = 1;
351         mutex_lock(&ubd_lock);
352
353         ubd_dev = &ubd_devs[n];
354         if(ubd_dev->file != NULL){
355                 printk(KERN_ERR "ubd_setup : device already configured\n");
356                 goto out;
357         }
358
359         if (index_out)
360                 *index_out = n;
361
362         for (i = 0; i < sizeof("rscd="); i++) {
363                 switch (*str) {
364                 case 'r':
365                         flags.w = 0;
366                         break;
367                 case 's':
368                         flags.s = 1;
369                         break;
370                 case 'd':
371                         ubd_dev->no_cow = 1;
372                         break;
373                 case 'c':
374                         ubd_dev->shared = 1;
375                         break;
376                 case '=':
377                         str++;
378                         goto break_loop;
379                 default:
380                         printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r, s, c, or d)\n");
381                         goto out;
382                 }
383                 str++;
384         }
385
386         if (*str == '=')
387                 printk(KERN_ERR "ubd_setup : Too many flags specified\n");
388         else
389                 printk(KERN_ERR "ubd_setup : Expected '='\n");
390         goto out;
391
392 break_loop:
393         err = 0;
394         backing_file = strchr(str, ',');
395
396         if (!backing_file) {
397                 backing_file = strchr(str, ':');
398         }
399
400         if(backing_file){
401                 if(ubd_dev->no_cow)
402                         printk(KERN_ERR "Can't specify both 'd' and a "
403                                "cow file\n");
404                 else {
405                         *backing_file = '\0';
406                         backing_file++;
407                 }
408         }
409         ubd_dev->file = str;
410         ubd_dev->cow.file = backing_file;
411         ubd_dev->boot_openflags = flags;
412 out:
413         mutex_unlock(&ubd_lock);
414         return(err);
415 }
416
417 static int ubd_setup(char *str)
418 {
419         ubd_setup_common(str, NULL);
420         return(1);
421 }
422
423 __setup("ubd", ubd_setup);
424 __uml_help(ubd_setup,
425 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
426 "    This is used to associate a device with a file in the underlying\n"
427 "    filesystem. When specifying two filenames, the first one is the\n"
428 "    COW name and the second is the backing file name. As separator you can\n"
429 "    use either a ':' or a ',': the first one allows writing things like;\n"
430 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
431 "    while with a ',' the shell would not expand the 2nd '~'.\n"
432 "    When using only one filename, UML will detect whether to thread it like\n"
433 "    a COW file or a backing file. To override this detection, add the 'd'\n"
434 "    flag:\n"
435 "       ubd0d=BackingFile\n"
436 "    Usually, there is a filesystem in the file, but \n"
437 "    that's not required. Swap devices containing swap files can be\n"
438 "    specified like this. Also, a file which doesn't contain a\n"
439 "    filesystem can have its contents read in the virtual \n"
440 "    machine by running 'dd' on the device. <n> must be in the range\n"
441 "    0 to 7. Appending an 'r' to the number will cause that device\n"
442 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
443 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
444 );
445
446 static int udb_setup(char *str)
447 {
448         printk("udb%s specified on command line is almost certainly a ubd -> "
449                "udb TYPO\n", str);
450         return(1);
451 }
452
453 __setup("udb", udb_setup);
454 __uml_help(udb_setup,
455 "udb\n"
456 "    This option is here solely to catch ubd -> udb typos, which can be\n"
457 "    to impossible to catch visually unless you specifically look for\n"
458 "    them.  The only result of any option starting with 'udb' is an error\n"
459 "    in the boot output.\n\n"
460 );
461
462 static int fakehd_set = 0;
463 static int fakehd(char *str)
464 {
465         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
466         fakehd_set = 1;
467         return 1;
468 }
469
470 __setup("fakehd", fakehd);
471 __uml_help(fakehd,
472 "fakehd\n"
473 "    Change the ubd device name to \"hd\".\n\n"
474 );
475
476 static void do_ubd_request(request_queue_t * q);
477
478 /* Only changed by ubd_init, which is an initcall. */
479 int thread_fd = -1;
480
481 /* Changed by ubd_handler, which is serialized because interrupts only
482  * happen on CPU 0.
483  */
484 int intr_count = 0;
485
486 /* call ubd_finish if you need to serialize */
487 static void __ubd_finish(struct request *req, int error)
488 {
489         int nsect;
490
491         if(error){
492                 end_request(req, 0);
493                 return;
494         }
495         nsect = req->current_nr_sectors;
496         req->sector += nsect;
497         req->buffer += nsect << 9;
498         req->errors = 0;
499         req->nr_sectors -= nsect;
500         req->current_nr_sectors = 0;
501         end_request(req, 1);
502 }
503
504 /* Callable only from interrupt context - otherwise you need to do
505  * spin_lock_irq()/spin_lock_irqsave() */
506 static inline void ubd_finish(struct request *req, int error)
507 {
508         spin_lock(&ubd_io_lock);
509         __ubd_finish(req, error);
510         spin_unlock(&ubd_io_lock);
511 }
512
513 /* XXX - move this inside ubd_intr. */
514 /* Called without ubd_io_lock held, and only in interrupt context. */
515 static void ubd_handler(void)
516 {
517         struct io_thread_req req;
518         struct request *rq = elv_next_request(ubd_queue);
519         int n;
520
521         do_ubd = 0;
522         intr_count++;
523         n = os_read_file(thread_fd, &req, sizeof(req));
524         if(n != sizeof(req)){
525                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
526                        "err = %d\n", os_getpid(), -n);
527                 spin_lock(&ubd_io_lock);
528                 end_request(rq, 0);
529                 spin_unlock(&ubd_io_lock);
530                 return;
531         }
532
533         ubd_finish(rq, req.error);
534         reactivate_fd(thread_fd, UBD_IRQ);      
535         spin_lock(&ubd_io_lock);
536         do_ubd_request(ubd_queue);
537         spin_unlock(&ubd_io_lock);
538 }
539
540 static irqreturn_t ubd_intr(int irq, void *dev)
541 {
542         ubd_handler();
543         return(IRQ_HANDLED);
544 }
545
546 /* Only changed by ubd_init, which is an initcall. */
547 static int io_pid = -1;
548
549 void kill_io_thread(void)
550 {
551         if(io_pid != -1)
552                 os_kill_process(io_pid, 1);
553 }
554
555 __uml_exitcall(kill_io_thread);
556
557 static int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
558 {
559         char *file;
560
561         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
562         return(os_file_size(file, size_out));
563 }
564
565 static void ubd_close_dev(struct ubd *ubd_dev)
566 {
567         os_close_file(ubd_dev->fd);
568         if(ubd_dev->cow.file == NULL)
569                 return;
570
571         os_close_file(ubd_dev->cow.fd);
572         vfree(ubd_dev->cow.bitmap);
573         ubd_dev->cow.bitmap = NULL;
574 }
575
576 static int ubd_open_dev(struct ubd *ubd_dev)
577 {
578         struct openflags flags;
579         char **back_ptr;
580         int err, create_cow, *create_ptr;
581         int fd;
582
583         ubd_dev->openflags = ubd_dev->boot_openflags;
584         create_cow = 0;
585         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
586         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
587
588         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
589                                 back_ptr, &ubd_dev->cow.bitmap_offset,
590                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
591                                 create_ptr);
592
593         if((fd == -ENOENT) && create_cow){
594                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
595                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
596                                           &ubd_dev->cow.bitmap_offset,
597                                           &ubd_dev->cow.bitmap_len,
598                                           &ubd_dev->cow.data_offset);
599                 if(fd >= 0){
600                         printk(KERN_INFO "Creating \"%s\" as COW file for "
601                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
602                 }
603         }
604
605         if(fd < 0){
606                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
607                        -fd);
608                 return fd;
609         }
610         ubd_dev->fd = fd;
611
612         if(ubd_dev->cow.file != NULL){
613                 err = -ENOMEM;
614                 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
615                 if(ubd_dev->cow.bitmap == NULL){
616                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
617                         goto error;
618                 }
619                 flush_tlb_kernel_vm();
620
621                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
622                                       ubd_dev->cow.bitmap_offset,
623                                       ubd_dev->cow.bitmap_len);
624                 if(err < 0)
625                         goto error;
626
627                 flags = ubd_dev->openflags;
628                 flags.w = 0;
629                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
630                                     NULL, NULL, NULL, NULL);
631                 if(err < 0) goto error;
632                 ubd_dev->cow.fd = err;
633         }
634         return(0);
635  error:
636         os_close_file(ubd_dev->fd);
637         return(err);
638 }
639
640 static int ubd_disk_register(int major, u64 size, int unit,
641                         struct gendisk **disk_out)
642                         
643 {
644         struct gendisk *disk;
645
646         disk = alloc_disk(1 << UBD_SHIFT);
647         if(disk == NULL)
648                 return(-ENOMEM);
649
650         disk->major = major;
651         disk->first_minor = unit << UBD_SHIFT;
652         disk->fops = &ubd_blops;
653         set_capacity(disk, size / 512);
654         if(major == MAJOR_NR)
655                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
656         else
657                 sprintf(disk->disk_name, "ubd_fake%d", unit);
658
659         /* sysfs register (not for ide fake devices) */
660         if (major == MAJOR_NR) {
661                 ubd_devs[unit].pdev.id   = unit;
662                 ubd_devs[unit].pdev.name = DRIVER_NAME;
663                 platform_device_register(&ubd_devs[unit].pdev);
664                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
665         }
666
667         disk->private_data = &ubd_devs[unit];
668         disk->queue = ubd_queue;
669         add_disk(disk);
670
671         *disk_out = disk;
672         return 0;
673 }
674
675 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
676
677 static int ubd_add(int n)
678 {
679         struct ubd *ubd_dev = &ubd_devs[n];
680         int err;
681
682         err = -ENODEV;
683         if(ubd_dev->file == NULL)
684                 goto out;
685
686         err = ubd_file_size(ubd_dev, &ubd_dev->size);
687         if(err < 0)
688                 goto out;
689
690         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
691
692         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
693         if(err)
694                 goto out;
695
696         if(fake_major != MAJOR_NR)
697                 ubd_disk_register(fake_major, ubd_dev->size, n,
698                              &fake_gendisk[n]);
699
700         /* perhaps this should also be under the "if (fake_major)" above */
701         /* using the fake_disk->disk_name and also the fakehd_set name */
702         if (fake_ide)
703                 make_ide_entries(ubd_gendisk[n]->disk_name);
704
705         err = 0;
706 out:
707         return err;
708 }
709
710 static int ubd_config(char *str)
711 {
712         int n, ret;
713
714         str = kstrdup(str, GFP_KERNEL);
715         if (str == NULL) {
716                 printk(KERN_ERR "ubd_config failed to strdup string\n");
717                 ret = 1;
718                 goto out;
719         }
720         ret = ubd_setup_common(str, &n);
721         if (ret) {
722                 ret = -1;
723                 goto err_free;
724         }
725         if (n == -1) {
726                 ret = 0;
727                 goto out;
728         }
729
730         mutex_lock(&ubd_lock);
731         ret = ubd_add(n);
732         if (ret)
733                 ubd_devs[n].file = NULL;
734         mutex_unlock(&ubd_lock);
735
736 out:
737         return ret;
738
739 err_free:
740         kfree(str);
741         goto out;
742 }
743
744 static int ubd_get_config(char *name, char *str, int size, char **error_out)
745 {
746         struct ubd *ubd_dev;
747         int n, len = 0;
748
749         n = parse_unit(&name);
750         if((n >= MAX_DEV) || (n < 0)){
751                 *error_out = "ubd_get_config : device number out of range";
752                 return(-1);
753         }
754
755         ubd_dev = &ubd_devs[n];
756         mutex_lock(&ubd_lock);
757
758         if(ubd_dev->file == NULL){
759                 CONFIG_CHUNK(str, size, len, "", 1);
760                 goto out;
761         }
762
763         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
764
765         if(ubd_dev->cow.file != NULL){
766                 CONFIG_CHUNK(str, size, len, ",", 0);
767                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
768         }
769         else CONFIG_CHUNK(str, size, len, "", 1);
770
771  out:
772         mutex_unlock(&ubd_lock);
773         return(len);
774 }
775
776 static int ubd_id(char **str, int *start_out, int *end_out)
777 {
778         int n;
779
780         n = parse_unit(str);
781         *start_out = 0;
782         *end_out = MAX_DEV - 1;
783         return n;
784 }
785
786 static int ubd_remove(int n)
787 {
788         struct ubd *ubd_dev;
789         int err = -ENODEV;
790
791         mutex_lock(&ubd_lock);
792
793         if(ubd_gendisk[n] == NULL)
794                 goto out;
795
796         ubd_dev = &ubd_devs[n];
797
798         if(ubd_dev->file == NULL)
799                 goto out;
800
801         /* you cannot remove a open disk */
802         err = -EBUSY;
803         if(ubd_dev->count > 0)
804                 goto out;
805
806         del_gendisk(ubd_gendisk[n]);
807         put_disk(ubd_gendisk[n]);
808         ubd_gendisk[n] = NULL;
809
810         if(fake_gendisk[n] != NULL){
811                 del_gendisk(fake_gendisk[n]);
812                 put_disk(fake_gendisk[n]);
813                 fake_gendisk[n] = NULL;
814         }
815
816         platform_device_unregister(&ubd_dev->pdev);
817         *ubd_dev = ((struct ubd) DEFAULT_UBD);
818         err = 0;
819 out:
820         mutex_unlock(&ubd_lock);
821         return err;
822 }
823
824 static struct mc_device ubd_mc = {
825         .name           = "ubd",
826         .config         = ubd_config,
827         .get_config     = ubd_get_config,
828         .id             = ubd_id,
829         .remove         = ubd_remove,
830 };
831
832 static int ubd_mc_init(void)
833 {
834         mconsole_register_dev(&ubd_mc);
835         return 0;
836 }
837
838 __initcall(ubd_mc_init);
839
840 static struct platform_driver ubd_driver = {
841         .driver = {
842                 .name  = DRIVER_NAME,
843         },
844 };
845
846 int ubd_init(void)
847 {
848         int i;
849
850         if (register_blkdev(MAJOR_NR, "ubd"))
851                 return -1;
852
853         ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
854         if (!ubd_queue) {
855                 unregister_blkdev(MAJOR_NR, "ubd");
856                 return -1;
857         }
858                 
859         if (fake_major != MAJOR_NR) {
860                 char name[sizeof("ubd_nnn\0")];
861
862                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
863                 if (register_blkdev(fake_major, "ubd"))
864                         return -1;
865         }
866         platform_driver_register(&ubd_driver);
867         for (i = 0; i < MAX_DEV; i++)
868                 ubd_add(i);
869         return 0;
870 }
871
872 late_initcall(ubd_init);
873
874 int ubd_driver_init(void){
875         unsigned long stack;
876         int err;
877
878         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
879         if(global_openflags.s){
880                 printk(KERN_INFO "ubd: Synchronous mode\n");
881                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
882                  * enough. So use anyway the io thread. */
883         }
884         stack = alloc_stack(0, 0);
885         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
886                                  &thread_fd);
887         if(io_pid < 0){
888                 printk(KERN_ERR
889                        "ubd : Failed to start I/O thread (errno = %d) - "
890                        "falling back to synchronous I/O\n", -io_pid);
891                 io_pid = -1;
892                 return(0);
893         }
894         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
895                              IRQF_DISABLED, "ubd", ubd_devs);
896         if(err != 0)
897                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
898         return 0;
899 }
900
901 device_initcall(ubd_driver_init);
902
903 static int ubd_open(struct inode *inode, struct file *filp)
904 {
905         struct gendisk *disk = inode->i_bdev->bd_disk;
906         struct ubd *ubd_dev = disk->private_data;
907         int err = 0;
908
909         if(ubd_dev->count == 0){
910                 err = ubd_open_dev(ubd_dev);
911                 if(err){
912                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
913                                disk->disk_name, ubd_dev->file, -err);
914                         goto out;
915                 }
916         }
917         ubd_dev->count++;
918         set_disk_ro(disk, !ubd_dev->openflags.w);
919
920         /* This should no more be needed. And it didn't work anyway to exclude
921          * read-write remounting of filesystems.*/
922         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
923                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
924                 err = -EROFS;
925         }*/
926  out:
927         return(err);
928 }
929
930 static int ubd_release(struct inode * inode, struct file * file)
931 {
932         struct gendisk *disk = inode->i_bdev->bd_disk;
933         struct ubd *ubd_dev = disk->private_data;
934
935         if(--ubd_dev->count == 0)
936                 ubd_close_dev(ubd_dev);
937         return(0);
938 }
939
940 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
941                           __u64 *cow_offset, unsigned long *bitmap,
942                           __u64 bitmap_offset, unsigned long *bitmap_words,
943                           __u64 bitmap_len)
944 {
945         __u64 sector = io_offset >> 9;
946         int i, update_bitmap = 0;
947
948         for(i = 0; i < length >> 9; i++){
949                 if(cow_mask != NULL)
950                         ubd_set_bit(i, (unsigned char *) cow_mask);
951                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
952                         continue;
953
954                 update_bitmap = 1;
955                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
956         }
957
958         if(!update_bitmap)
959                 return;
960
961         *cow_offset = sector / (sizeof(unsigned long) * 8);
962
963         /* This takes care of the case where we're exactly at the end of the
964          * device, and *cow_offset + 1 is off the end.  So, just back it up
965          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
966          * for the original diagnosis.
967          */
968         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
969                            sizeof(unsigned long) - 1))
970                 (*cow_offset)--;
971
972         bitmap_words[0] = bitmap[*cow_offset];
973         bitmap_words[1] = bitmap[*cow_offset + 1];
974
975         *cow_offset *= sizeof(unsigned long);
976         *cow_offset += bitmap_offset;
977 }
978
979 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
980                        __u64 bitmap_offset, __u64 bitmap_len)
981 {
982         __u64 sector = req->offset >> 9;
983         int i;
984
985         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
986                 panic("Operation too long");
987
988         if(req->op == UBD_READ) {
989                 for(i = 0; i < req->length >> 9; i++){
990                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
991                                 ubd_set_bit(i, (unsigned char *)
992                                             &req->sector_mask);
993                 }
994         }
995         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
996                            &req->cow_offset, bitmap, bitmap_offset,
997                            req->bitmap_words, bitmap_len);
998 }
999
1000 /* Called with ubd_io_lock held */
1001 static int prepare_request(struct request *req, struct io_thread_req *io_req)
1002 {
1003         struct gendisk *disk = req->rq_disk;
1004         struct ubd *ubd_dev = disk->private_data;
1005         __u64 offset;
1006         int len;
1007
1008         /* This should be impossible now */
1009         if((rq_data_dir(req) == WRITE) && !ubd_dev->openflags.w){
1010                 printk("Write attempted on readonly ubd device %s\n",
1011                        disk->disk_name);
1012                 end_request(req, 0);
1013                 return(1);
1014         }
1015
1016         offset = ((__u64) req->sector) << 9;
1017         len = req->current_nr_sectors << 9;
1018
1019         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd : ubd_dev->fd;
1020         io_req->fds[1] = ubd_dev->fd;
1021         io_req->cow_offset = -1;
1022         io_req->offset = offset;
1023         io_req->length = len;
1024         io_req->error = 0;
1025         io_req->sector_mask = 0;
1026
1027         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1028         io_req->offsets[0] = 0;
1029         io_req->offsets[1] = ubd_dev->cow.data_offset;
1030         io_req->buffer = req->buffer;
1031         io_req->sectorsize = 1 << 9;
1032
1033         if(ubd_dev->cow.file != NULL)
1034                 cowify_req(io_req, ubd_dev->cow.bitmap, ubd_dev->cow.bitmap_offset,
1035                            ubd_dev->cow.bitmap_len);
1036
1037         return(0);
1038 }
1039
1040 /* Called with ubd_io_lock held */
1041 static void do_ubd_request(request_queue_t *q)
1042 {
1043         struct io_thread_req io_req;
1044         struct request *req;
1045         int err, n;
1046
1047         if(thread_fd == -1){
1048                 while((req = elv_next_request(q)) != NULL){
1049                         err = prepare_request(req, &io_req);
1050                         if(!err){
1051                                 do_io(&io_req);
1052                                 __ubd_finish(req, io_req.error);
1053                         }
1054                 }
1055         }
1056         else {
1057                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1058                         return;
1059                 err = prepare_request(req, &io_req);
1060                 if(!err){
1061                         do_ubd = 1;
1062                         n = os_write_file(thread_fd, (char *) &io_req,
1063                                          sizeof(io_req));
1064                         if(n != sizeof(io_req))
1065                                 printk("write to io thread failed, "
1066                                        "errno = %d\n", -n);
1067                 }
1068         }
1069 }
1070
1071 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1072 {
1073         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1074
1075         geo->heads = 128;
1076         geo->sectors = 32;
1077         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1078         return 0;
1079 }
1080
1081 static int ubd_ioctl(struct inode * inode, struct file * file,
1082                      unsigned int cmd, unsigned long arg)
1083 {
1084         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1085         struct hd_driveid ubd_id = {
1086                 .cyls           = 0,
1087                 .heads          = 128,
1088                 .sectors        = 32,
1089         };
1090
1091         switch (cmd) {
1092                 struct cdrom_volctrl volume;
1093         case HDIO_GET_IDENTITY:
1094                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1095                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1096                                  sizeof(ubd_id)))
1097                         return(-EFAULT);
1098                 return(0);
1099                 
1100         case CDROMVOLREAD:
1101                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1102                         return(-EFAULT);
1103                 volume.channel0 = 255;
1104                 volume.channel1 = 255;
1105                 volume.channel2 = 255;
1106                 volume.channel3 = 255;
1107                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1108                         return(-EFAULT);
1109                 return(0);
1110         }
1111         return(-EINVAL);
1112 }
1113
1114 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1115 {
1116         struct uml_stat buf1, buf2;
1117         int err;
1118
1119         if(from_cmdline == NULL)
1120                 return 0;
1121         if(!strcmp(from_cmdline, from_cow))
1122                 return 0;
1123
1124         err = os_stat_file(from_cmdline, &buf1);
1125         if(err < 0){
1126                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1127                 return 0;
1128         }
1129         err = os_stat_file(from_cow, &buf2);
1130         if(err < 0){
1131                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1132                 return 1;
1133         }
1134         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1135                 return 0;
1136
1137         printk("Backing file mismatch - \"%s\" requested,\n"
1138                "\"%s\" specified in COW header of \"%s\"\n",
1139                from_cmdline, from_cow, cow);
1140         return 1;
1141 }
1142
1143 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1144 {
1145         unsigned long modtime;
1146         unsigned long long actual;
1147         int err;
1148
1149         err = os_file_modtime(file, &modtime);
1150         if(err < 0){
1151                 printk("Failed to get modification time of backing file "
1152                        "\"%s\", err = %d\n", file, -err);
1153                 return(err);
1154         }
1155
1156         err = os_file_size(file, &actual);
1157         if(err < 0){
1158                 printk("Failed to get size of backing file \"%s\", "
1159                        "err = %d\n", file, -err);
1160                 return(err);
1161         }
1162
1163         if(actual != size){
1164                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1165                  * the typecast.*/
1166                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1167                        "file\n", (unsigned long long) size, actual);
1168                 return(-EINVAL);
1169         }
1170         if(modtime != mtime){
1171                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1172                        "file\n", mtime, modtime);
1173                 return(-EINVAL);
1174         }
1175         return(0);
1176 }
1177
1178 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1179 {
1180         int err;
1181
1182         err = os_seek_file(fd, offset);
1183         if(err < 0)
1184                 return(err);
1185
1186         err = os_read_file(fd, buf, len);
1187         if(err < 0)
1188                 return(err);
1189
1190         return(0);
1191 }
1192
1193 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1194                   char **backing_file_out, int *bitmap_offset_out,
1195                   unsigned long *bitmap_len_out, int *data_offset_out,
1196                   int *create_cow_out)
1197 {
1198         time_t mtime;
1199         unsigned long long size;
1200         __u32 version, align;
1201         char *backing_file;
1202         int fd, err, sectorsize, asked_switch, mode = 0644;
1203
1204         fd = os_open_file(file, *openflags, mode);
1205         if (fd < 0) {
1206                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1207                         *create_cow_out = 1;
1208                 if (!openflags->w ||
1209                    ((fd != -EROFS) && (fd != -EACCES)))
1210                         return fd;
1211                 openflags->w = 0;
1212                 fd = os_open_file(file, *openflags, mode);
1213                 if (fd < 0)
1214                         return fd;
1215         }
1216
1217         if(shared)
1218                 printk("Not locking \"%s\" on the host\n", file);
1219         else {
1220                 err = os_lock_file(fd, openflags->w);
1221                 if(err < 0){
1222                         printk("Failed to lock '%s', err = %d\n", file, -err);
1223                         goto out_close;
1224                 }
1225         }
1226
1227         /* Successful return case! */
1228         if(backing_file_out == NULL)
1229                 return(fd);
1230
1231         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1232                               &size, &sectorsize, &align, bitmap_offset_out);
1233         if(err && (*backing_file_out != NULL)){
1234                 printk("Failed to read COW header from COW file \"%s\", "
1235                        "errno = %d\n", file, -err);
1236                 goto out_close;
1237         }
1238         if(err)
1239                 return(fd);
1240
1241         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1242
1243         /* Allow switching only if no mismatch. */
1244         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1245                 printk("Switching backing file to '%s'\n", *backing_file_out);
1246                 err = write_cow_header(file, fd, *backing_file_out,
1247                                        sectorsize, align, &size);
1248                 if (err) {
1249                         printk("Switch failed, errno = %d\n", -err);
1250                         goto out_close;
1251                 }
1252         } else {
1253                 *backing_file_out = backing_file;
1254                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1255                 if (err)
1256                         goto out_close;
1257         }
1258
1259         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1260                   bitmap_len_out, data_offset_out);
1261
1262         return fd;
1263  out_close:
1264         os_close_file(fd);
1265         return err;
1266 }
1267
1268 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1269                     int sectorsize, int alignment, int *bitmap_offset_out,
1270                     unsigned long *bitmap_len_out, int *data_offset_out)
1271 {
1272         int err, fd;
1273
1274         flags.c = 1;
1275         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1276         if(fd < 0){
1277                 err = fd;
1278                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1279                        -err);
1280                 goto out;
1281         }
1282
1283         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1284                             bitmap_offset_out, bitmap_len_out,
1285                             data_offset_out);
1286         if(!err)
1287                 return(fd);
1288         os_close_file(fd);
1289  out:
1290         return(err);
1291 }
1292
1293 static int update_bitmap(struct io_thread_req *req)
1294 {
1295         int n;
1296
1297         if(req->cow_offset == -1)
1298                 return(0);
1299
1300         n = os_seek_file(req->fds[1], req->cow_offset);
1301         if(n < 0){
1302                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1303                 return(1);
1304         }
1305
1306         n = os_write_file(req->fds[1], &req->bitmap_words,
1307                           sizeof(req->bitmap_words));
1308         if(n != sizeof(req->bitmap_words)){
1309                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1310                        req->fds[1]);
1311                 return(1);
1312         }
1313
1314         return(0);
1315 }
1316
1317 void do_io(struct io_thread_req *req)
1318 {
1319         char *buf;
1320         unsigned long len;
1321         int n, nsectors, start, end, bit;
1322         int err;
1323         __u64 off;
1324
1325         nsectors = req->length / req->sectorsize;
1326         start = 0;
1327         do {
1328                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1329                 end = start;
1330                 while((end < nsectors) &&
1331                       (ubd_test_bit(end, (unsigned char *)
1332                                     &req->sector_mask) == bit))
1333                         end++;
1334
1335                 off = req->offset + req->offsets[bit] +
1336                         start * req->sectorsize;
1337                 len = (end - start) * req->sectorsize;
1338                 buf = &req->buffer[start * req->sectorsize];
1339
1340                 err = os_seek_file(req->fds[bit], off);
1341                 if(err < 0){
1342                         printk("do_io - lseek failed : err = %d\n", -err);
1343                         req->error = 1;
1344                         return;
1345                 }
1346                 if(req->op == UBD_READ){
1347                         n = 0;
1348                         do {
1349                                 buf = &buf[n];
1350                                 len -= n;
1351                                 n = os_read_file(req->fds[bit], buf, len);
1352                                 if (n < 0) {
1353                                         printk("do_io - read failed, err = %d "
1354                                                "fd = %d\n", -n, req->fds[bit]);
1355                                         req->error = 1;
1356                                         return;
1357                                 }
1358                         } while((n < len) && (n != 0));
1359                         if (n < len) memset(&buf[n], 0, len - n);
1360                 } else {
1361                         n = os_write_file(req->fds[bit], buf, len);
1362                         if(n != len){
1363                                 printk("do_io - write failed err = %d "
1364                                        "fd = %d\n", -n, req->fds[bit]);
1365                                 req->error = 1;
1366                                 return;
1367                         }
1368                 }
1369
1370                 start = end;
1371         } while(start < nsectors);
1372
1373         req->error = update_bitmap(req);
1374 }
1375
1376 /* Changed in start_io_thread, which is serialized by being called only
1377  * from ubd_init, which is an initcall.
1378  */
1379 int kernel_fd = -1;
1380
1381 /* Only changed by the io thread */
1382 int io_count = 0;
1383
1384 int io_thread(void *arg)
1385 {
1386         struct io_thread_req req;
1387         int n;
1388
1389         ignore_sigwinch_sig();
1390         while(1){
1391                 n = os_read_file(kernel_fd, &req, sizeof(req));
1392                 if(n != sizeof(req)){
1393                         if(n < 0)
1394                                 printk("io_thread - read failed, fd = %d, "
1395                                        "err = %d\n", kernel_fd, -n);
1396                         else {
1397                                 printk("io_thread - short read, fd = %d, "
1398                                        "length = %d\n", kernel_fd, n);
1399                         }
1400                         continue;
1401                 }
1402                 io_count++;
1403                 do_io(&req);
1404                 n = os_write_file(kernel_fd, &req, sizeof(req));
1405                 if(n != sizeof(req))
1406                         printk("io_thread - write failed, fd = %d, err = %d\n",
1407                                kernel_fd, -n);
1408         }
1409
1410         return 0;
1411 }