SAFE public projects git trees. - safe/jmp/linux-2.6/blob - arch/um/drivers/ubd_kern.c

   1 /*
   2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
   3  * Licensed under the GPL
   4  */
   5
   6 /* 2001-09-28...2002-04-17
   7  * Partition stuff by James_McMechan@hotmail.com
   8  * old style ubd by setting UBD_SHIFT to 0
   9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
  10  * partitions have changed in 2.5
  11  * 2003-01-29 more tinkering for 2.5.59-1
  12  * This should now address the sysfs problems and has
  13  * the symlink for devfs to allow for booting with
  14  * the common /dev/ubd/discX/... names rather than
  15  * only /dev/ubdN/discN this version also has lots of
  16  * clean ups preparing for ubd-many.
  17  * James McMechan
  18  */
  19
  20 #define MAJOR_NR UBD_MAJOR
  21 #define UBD_SHIFT 4
  22
  23 #include "linux/module.h"
  24 #include "linux/blkdev.h"
  25 #include "linux/hdreg.h"
  26 #include "linux/init.h"
  27 #include "linux/cdrom.h"
  28 #include "linux/proc_fs.h"
  29 #include "linux/ctype.h"
  30 #include "linux/capability.h"
  31 #include "linux/mm.h"
  32 #include "linux/vmalloc.h"
  33 #include "linux/blkpg.h"
  34 #include "linux/genhd.h"
  35 #include "linux/spinlock.h"
  36 #include "linux/platform_device.h"
  37 #include "asm/segment.h"
  38 #include "asm/uaccess.h"
  39 #include "asm/irq.h"
  40 #include "asm/types.h"
  41 #include "asm/tlbflush.h"
  42 #include "user_util.h"
  43 #include "mem_user.h"
  44 #include "kern_util.h"
  45 #include "kern.h"
  46 #include "mconsole_kern.h"
  47 #include "init.h"
  48 #include "irq_user.h"
  49 #include "irq_kern.h"
  50 #include "ubd_user.h"
  51 #include "os.h"
  52 #include "mem.h"
  53 #include "mem_kern.h"
  54 #include "cow.h"
  55
  56 enum ubd_req { UBD_READ, UBD_WRITE };
  57
  58 struct io_thread_req {
  59         struct request *req;
  60         enum ubd_req op;
  61         int fds[2];
  62         unsigned long offsets[2];
  63         unsigned long long offset;
  64         unsigned long length;
  65         char *buffer;
  66         int sectorsize;
  67         unsigned long sector_mask;
  68         unsigned long long cow_offset;
  69         unsigned long bitmap_words[2];
  70         int error;
  71 };
  72
  73 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
  74                          char **backing_file_out, int *bitmap_offset_out,
  75                          unsigned long *bitmap_len_out, int *data_offset_out,
  76                          int *create_cow_out);
  77 extern int create_cow_file(char *cow_file, char *backing_file,
  78                            struct openflags flags, int sectorsize,
  79                            int alignment, int *bitmap_offset_out,
  80                            unsigned long *bitmap_len_out,
  81                            int *data_offset_out);
  82 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
  83 extern void do_io(struct io_thread_req *req);
  84
  85 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  86 {
  87         __u64 n;
  88         int bits, off;
  89
  90         bits = sizeof(data[0]) * 8;
  91         n = bit / bits;
  92         off = bit % bits;
  93         return((data[n] & (1 << off)) != 0);
  94 }
  95
  96 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  97 {
  98         __u64 n;
  99         int bits, off;
 100
 101         bits = sizeof(data[0]) * 8;
 102         n = bit / bits;
 103         off = bit % bits;
 104         data[n] |= (1 << off);
 105 }
 106 /*End stuff from ubd_user.h*/
 107
 108 #define DRIVER_NAME "uml-blkdev"
 109
 110 static DEFINE_MUTEX(ubd_lock);
 111
 112 /* XXX - this made sense in 2.4 days, now it's only used as a boolean, and
 113  * probably it doesn't make sense even for that. */
 114 static int do_ubd;
 115
 116 static int ubd_open(struct inode * inode, struct file * filp);
 117 static int ubd_release(struct inode * inode, struct file * file);
 118 static int ubd_ioctl(struct inode * inode, struct file * file,
 119                      unsigned int cmd, unsigned long arg);
 120 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 121
 122 #define MAX_DEV (16)
 123
 124 static struct block_device_operations ubd_blops = {
 125         .owner          = THIS_MODULE,
 126         .open           = ubd_open,
 127         .release        = ubd_release,
 128         .ioctl          = ubd_ioctl,
 129         .getgeo         = ubd_getgeo,
 130 };
 131
 132 /* Protected by ubd_lock */
 133 static int fake_major = MAJOR_NR;
 134 static struct gendisk *ubd_gendisk[MAX_DEV];
 135 static struct gendisk *fake_gendisk[MAX_DEV];
 136
 137 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 138 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 139                                          .cl = 1 })
 140 #else
 141 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 142                                          .cl = 1 })
 143 #endif
 144 static struct openflags global_openflags = OPEN_FLAGS;
 145
 146 struct cow {
 147         /* backing file name */
 148         char *file;
 149         /* backing file fd */
 150         int fd;
 151         unsigned long *bitmap;
 152         unsigned long bitmap_len;
 153         int bitmap_offset;
 154         int data_offset;
 155 };
 156
 157 struct ubd {
 158         /* name (and fd, below) of the file opened for writing, either the
 159          * backing or the cow file. */
 160         char *file;
 161         int count;
 162         int fd;
 163         __u64 size;
 164         struct openflags boot_openflags;
 165         struct openflags openflags;
 166         unsigned shared:1;
 167         unsigned no_cow:1;
 168         struct cow cow;
 169         struct platform_device pdev;
 170         struct request_queue *queue;
 171         spinlock_t lock;
 172 };
 173
 174 #define DEFAULT_COW { \
 175         .file =                 NULL, \
 176         .fd =                   -1, \
 177         .bitmap =               NULL, \
 178         .bitmap_offset =        0, \
 179         .data_offset =          0, \
 180 }
 181
 182 #define DEFAULT_UBD { \
 183         .file =                 NULL, \
 184         .count =                0, \
 185         .fd =                   -1, \
 186         .size =                 -1, \
 187         .boot_openflags =       OPEN_FLAGS, \
 188         .openflags =            OPEN_FLAGS, \
 189         .no_cow =               0, \
 190         .shared =               0, \
 191         .cow =                  DEFAULT_COW, \
 192         .lock =                 SPIN_LOCK_UNLOCKED,     \
 193 }
 194
 195 /* Protected by ubd_lock */
 196 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
 197
 198 /* Only changed by fake_ide_setup which is a setup */
 199 static int fake_ide = 0;
 200 static struct proc_dir_entry *proc_ide_root = NULL;
 201 static struct proc_dir_entry *proc_ide = NULL;
 202
 203 static void make_proc_ide(void)
 204 {
 205         proc_ide_root = proc_mkdir("ide", NULL);
 206         proc_ide = proc_mkdir("ide0", proc_ide_root);
 207 }
 208
 209 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
 210                                int *eof, void *data)
 211 {
 212         int len;
 213
 214         strcpy(page, "disk\n");
 215         len = strlen("disk\n");
 216         len -= off;
 217         if (len < count){
 218                 *eof = 1;
 219                 if (len <= 0) return 0;
 220         }
 221         else len = count;
 222         *start = page + off;
 223         return len;
 224 }
 225
 226 static void make_ide_entries(char *dev_name)
 227 {
 228         struct proc_dir_entry *dir, *ent;
 229         char name[64];
 230
 231         if(proc_ide_root == NULL) make_proc_ide();
 232
 233         dir = proc_mkdir(dev_name, proc_ide);
 234         if(!dir) return;
 235
 236         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
 237         if(!ent) return;
 238         ent->nlink = 1;
 239         ent->data = NULL;
 240         ent->read_proc = proc_ide_read_media;
 241         ent->write_proc = NULL;
 242         sprintf(name,"ide0/%s", dev_name);
 243         proc_symlink(dev_name, proc_ide_root, name);
 244 }
 245
 246 static int fake_ide_setup(char *str)
 247 {
 248         fake_ide = 1;
 249         return(1);
 250 }
 251
 252 __setup("fake_ide", fake_ide_setup);
 253
 254 __uml_help(fake_ide_setup,
 255 "fake_ide\n"
 256 "    Create ide0 entries that map onto ubd devices.\n\n"
 257 );
 258
 259 static int parse_unit(char **ptr)
 260 {
 261         char *str = *ptr, *end;
 262         int n = -1;
 263
 264         if(isdigit(*str)) {
 265                 n = simple_strtoul(str, &end, 0);
 266                 if(end == str)
 267                         return(-1);
 268                 *ptr = end;
 269         }
 270         else if (('a' <= *str) && (*str <= 'z')) {
 271                 n = *str - 'a';
 272                 str++;
 273                 *ptr = str;
 274         }
 275         return(n);
 276 }
 277
 278 /* If *index_out == -1 at exit, the passed option was a general one;
 279  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 280  * should not be freed on exit.
 281  */
 282 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 283 {
 284         struct ubd *ubd_dev;
 285         struct openflags flags = global_openflags;
 286         char *backing_file;
 287         int n, err = 0, i;
 288
 289         if(index_out) *index_out = -1;
 290         n = *str;
 291         if(n == '='){
 292                 char *end;
 293                 int major;
 294
 295                 str++;
 296                 if(!strcmp(str, "sync")){
 297                         global_openflags = of_sync(global_openflags);
 298                         goto out1;
 299                 }
 300
 301                 err = -EINVAL;
 302                 major = simple_strtoul(str, &end, 0);
 303                 if((*end != '\0') || (end == str)){
 304                         *error_out = "Didn't parse major number";
 305                         goto out1;
 306                 }
 307
 308                 mutex_lock(&ubd_lock);
 309                 if(fake_major != MAJOR_NR){
 310                         *error_out = "Can't assign a fake major twice";
 311                         goto out1;
 312                 }
 313
 314                 fake_major = major;
 315
 316                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 317                        major);
 318                 err = 0;
 319         out1:
 320                 mutex_unlock(&ubd_lock);
 321                 return err;
 322         }
 323
 324         n = parse_unit(&str);
 325         if(n < 0){
 326                 *error_out = "Couldn't parse device number";
 327                 return -EINVAL;
 328         }
 329         if(n >= MAX_DEV){
 330                 *error_out = "Device number out of range";
 331                 return 1;
 332         }
 333
 334         err = -EBUSY;
 335         mutex_lock(&ubd_lock);
 336
 337         ubd_dev = &ubd_devs[n];
 338         if(ubd_dev->file != NULL){
 339                 *error_out = "Device is already configured";
 340                 goto out;
 341         }
 342
 343         if (index_out)
 344                 *index_out = n;
 345
 346         err = -EINVAL;
 347         for (i = 0; i < sizeof("rscd="); i++) {
 348                 switch (*str) {
 349                 case 'r':
 350                         flags.w = 0;
 351                         break;
 352                 case 's':
 353                         flags.s = 1;
 354                         break;
 355                 case 'd':
 356                         ubd_dev->no_cow = 1;
 357                         break;
 358                 case 'c':
 359                         ubd_dev->shared = 1;
 360                         break;
 361                 case '=':
 362                         str++;
 363                         goto break_loop;
 364                 default:
 365                         *error_out = "Expected '=' or flag letter "
 366                                 "(r, s, c, or d)";
 367                         goto out;
 368                 }
 369                 str++;
 370         }
 371
 372         if (*str == '=')
 373                 *error_out = "Too many flags specified";
 374         else
 375                 *error_out = "Missing '='";
 376         goto out;
 377
 378 break_loop:
 379         backing_file = strchr(str, ',');
 380
 381         if (backing_file == NULL)
 382                 backing_file = strchr(str, ':');
 383
 384         if(backing_file != NULL){
 385                 if(ubd_dev->no_cow){
 386                         *error_out = "Can't specify both 'd' and a cow file";
 387                         goto out;
 388                 }
 389                 else {
 390                         *backing_file = '\0';
 391                         backing_file++;
 392                 }
 393         }
 394         err = 0;
 395         ubd_dev->file = str;
 396         ubd_dev->cow.file = backing_file;
 397         ubd_dev->boot_openflags = flags;
 398 out:
 399         mutex_unlock(&ubd_lock);
 400         return err;
 401 }
 402
 403 static int ubd_setup(char *str)
 404 {
 405         char *error;
 406         int err;
 407
 408         err = ubd_setup_common(str, NULL, &error);
 409         if(err)
 410                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 411                        "%s\n", str, error);
 412         return 1;
 413 }
 414
 415 __setup("ubd", ubd_setup);
 416 __uml_help(ubd_setup,
 417 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 418 "    This is used to associate a device with a file in the underlying\n"
 419 "    filesystem. When specifying two filenames, the first one is the\n"
 420 "    COW name and the second is the backing file name. As separator you can\n"
 421 "    use either a ':' or a ',': the first one allows writing things like;\n"
 422 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 423 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 424 "    When using only one filename, UML will detect whether to treat it like\n"
 425 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 426 "    flag:\n"
 427 "       ubd0d=BackingFile\n"
 428 "    Usually, there is a filesystem in the file, but \n"
 429 "    that's not required. Swap devices containing swap files can be\n"
 430 "    specified like this. Also, a file which doesn't contain a\n"
 431 "    filesystem can have its contents read in the virtual \n"
 432 "    machine by running 'dd' on the device. <n> must be in the range\n"
 433 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 434 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 435 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
 436 );
 437
 438 static int udb_setup(char *str)
 439 {
 440         printk("udb%s specified on command line is almost certainly a ubd -> "
 441                "udb TYPO\n", str);
 442         return(1);
 443 }
 444
 445 __setup("udb", udb_setup);
 446 __uml_help(udb_setup,
 447 "udb\n"
 448 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 449 "    to impossible to catch visually unless you specifically look for\n"
 450 "    them.  The only result of any option starting with 'udb' is an error\n"
 451 "    in the boot output.\n\n"
 452 );
 453
 454 static int fakehd_set = 0;
 455 static int fakehd(char *str)
 456 {
 457         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
 458         fakehd_set = 1;
 459         return 1;
 460 }
 461
 462 __setup("fakehd", fakehd);
 463 __uml_help(fakehd,
 464 "fakehd\n"
 465 "    Change the ubd device name to \"hd\".\n\n"
 466 );
 467
 468 static void do_ubd_request(request_queue_t * q);
 469
 470 /* Only changed by ubd_init, which is an initcall. */
 471 int thread_fd = -1;
 472
 473 /* call ubd_finish if you need to serialize */
 474 static void __ubd_finish(struct request *req, int error)
 475 {
 476         int nsect;
 477
 478         if(error){
 479                 end_request(req, 0);
 480                 return;
 481         }
 482         nsect = req->current_nr_sectors;
 483         req->sector += nsect;
 484         req->buffer += nsect << 9;
 485         req->errors = 0;
 486         req->nr_sectors -= nsect;
 487         req->current_nr_sectors = 0;
 488         end_request(req, 1);
 489 }
 490
 491 /* Callable only from interrupt context - otherwise you need to do
 492  * spin_lock_irq()/spin_lock_irqsave() */
 493 static inline void ubd_finish(struct request *req, int error)
 494 {
 495         struct ubd *dev = req->rq_disk->private_data;
 496
 497         spin_lock(&dev->lock);
 498         __ubd_finish(req, error);
 499         spin_unlock(&dev->lock);
 500 }
 501
 502 /* XXX - move this inside ubd_intr. */
 503 /* Called without dev->lock held, and only in interrupt context. */
 504 static void ubd_handler(void)
 505 {
 506         struct io_thread_req req;
 507         struct request *rq;
 508         struct ubd *dev;
 509         int n;
 510
 511         do_ubd = 0;
 512         n = os_read_file(thread_fd, &req, sizeof(req));
 513         if(n != sizeof(req)){
 514                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
 515                        "err = %d\n", os_getpid(), -n);
 516                 return;
 517         }
 518
 519         rq = req.req;
 520         dev = rq->rq_disk->private_data;
 521
 522         ubd_finish(rq, req.error);
 523         reactivate_fd(thread_fd, UBD_IRQ);
 524         spin_lock(&dev->lock);
 525         do_ubd_request(dev->queue);
 526         spin_unlock(&dev->lock);
 527 }
 528
 529 static irqreturn_t ubd_intr(int irq, void *dev)
 530 {
 531         ubd_handler();
 532         return(IRQ_HANDLED);
 533 }
 534
 535 /* Only changed by ubd_init, which is an initcall. */
 536 static int io_pid = -1;
 537
 538 void kill_io_thread(void)
 539 {
 540         if(io_pid != -1)
 541                 os_kill_process(io_pid, 1);
 542 }
 543
 544 __uml_exitcall(kill_io_thread);
 545
 546 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 547 {
 548         char *file;
 549
 550         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
 551         return(os_file_size(file, size_out));
 552 }
 553
 554 static void ubd_close_dev(struct ubd *ubd_dev)
 555 {
 556         os_close_file(ubd_dev->fd);
 557         if(ubd_dev->cow.file == NULL)
 558                 return;
 559
 560         os_close_file(ubd_dev->cow.fd);
 561         vfree(ubd_dev->cow.bitmap);
 562         ubd_dev->cow.bitmap = NULL;
 563 }
 564
 565 static int ubd_open_dev(struct ubd *ubd_dev)
 566 {
 567         struct openflags flags;
 568         char **back_ptr;
 569         int err, create_cow, *create_ptr;
 570         int fd;
 571
 572         ubd_dev->openflags = ubd_dev->boot_openflags;
 573         create_cow = 0;
 574         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 575         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 576
 577         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 578                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 579                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 580                                 create_ptr);
 581
 582         if((fd == -ENOENT) && create_cow){
 583                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 584                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 585                                           &ubd_dev->cow.bitmap_offset,
 586                                           &ubd_dev->cow.bitmap_len,
 587                                           &ubd_dev->cow.data_offset);
 588                 if(fd >= 0){
 589                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 590                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 591                 }
 592         }
 593
 594         if(fd < 0){
 595                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 596                        -fd);
 597                 return fd;
 598         }
 599         ubd_dev->fd = fd;
 600
 601         if(ubd_dev->cow.file != NULL){
 602                 err = -ENOMEM;
 603                 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
 604                 if(ubd_dev->cow.bitmap == NULL){
 605                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 606                         goto error;
 607                 }
 608                 flush_tlb_kernel_vm();
 609
 610                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 611                                       ubd_dev->cow.bitmap_offset,
 612                                       ubd_dev->cow.bitmap_len);
 613                 if(err < 0)
 614                         goto error;
 615
 616                 flags = ubd_dev->openflags;
 617                 flags.w = 0;
 618                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 619                                     NULL, NULL, NULL, NULL);
 620                 if(err < 0) goto error;
 621                 ubd_dev->cow.fd = err;
 622         }
 623         return(0);
 624  error:
 625         os_close_file(ubd_dev->fd);
 626         return(err);
 627 }
 628
 629 static int ubd_disk_register(int major, u64 size, int unit,
 630                              struct gendisk **disk_out)
 631 {
 632         struct gendisk *disk;
 633
 634         disk = alloc_disk(1 << UBD_SHIFT);
 635         if(disk == NULL)
 636                 return(-ENOMEM);
 637
 638         disk->major = major;
 639         disk->first_minor = unit << UBD_SHIFT;
 640         disk->fops = &ubd_blops;
 641         set_capacity(disk, size / 512);
 642         if(major == MAJOR_NR)
 643                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 644         else
 645                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 646
 647         /* sysfs register (not for ide fake devices) */
 648         if (major == MAJOR_NR) {
 649                 ubd_devs[unit].pdev.id   = unit;
 650                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 651                 platform_device_register(&ubd_devs[unit].pdev);
 652                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 653         }
 654
 655         disk->private_data = &ubd_devs[unit];
 656         disk->queue = ubd_devs[unit].queue;
 657         add_disk(disk);
 658
 659         *disk_out = disk;
 660         return 0;
 661 }
 662
 663 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 664
 665 static int ubd_add(int n, char **error_out)
 666 {
 667         struct ubd *ubd_dev = &ubd_devs[n];
 668         int err = 0;
 669
 670         if(ubd_dev->file == NULL)
 671                 goto out;
 672
 673         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 674         if(err < 0){
 675                 *error_out = "Couldn't determine size of device's file";
 676                 goto out;
 677         }
 678
 679         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 680
 681         err = -ENOMEM;
 682         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 683         if (ubd_dev->queue == NULL) {
 684                 *error_out = "Failed to initialize device queue";
 685                 goto out;
 686         }
 687         ubd_dev->queue->queuedata = ubd_dev;
 688
 689         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
 690         if(err){
 691                 *error_out = "Failed to register device";
 692                 goto out_cleanup;
 693         }
 694
 695         if(fake_major != MAJOR_NR)
 696                 ubd_disk_register(fake_major, ubd_dev->size, n,
 697                                   &fake_gendisk[n]);
 698
 699         /* perhaps this should also be under the "if (fake_major)" above */
 700         /* using the fake_disk->disk_name and also the fakehd_set name */
 701         if (fake_ide)
 702                 make_ide_entries(ubd_gendisk[n]->disk_name);
 703
 704         err = 0;
 705 out:
 706         return err;
 707
 708 out_cleanup:
 709         blk_cleanup_queue(ubd_dev->queue);
 710         goto out;
 711 }
 712
 713 static int ubd_config(char *str, char **error_out)
 714 {
 715         int n, ret;
 716
 717         /* This string is possibly broken up and stored, so it's only
 718          * freed if ubd_setup_common fails, or if only general options
 719          * were set.
 720          */
 721         str = kstrdup(str, GFP_KERNEL);
 722         if (str == NULL) {
 723                 *error_out = "Failed to allocate memory";
 724                 return -ENOMEM;
 725         }
 726
 727         ret = ubd_setup_common(str, &n, error_out);
 728         if (ret)
 729                 goto err_free;
 730
 731         if (n == -1) {
 732                 ret = 0;
 733                 goto err_free;
 734         }
 735
 736         mutex_lock(&ubd_lock);
 737         ret = ubd_add(n, error_out);
 738         if (ret)
 739                 ubd_devs[n].file = NULL;
 740         mutex_unlock(&ubd_lock);
 741
 742 out:
 743         return ret;
 744
 745 err_free:
 746         kfree(str);
 747         goto out;
 748 }
 749
 750 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 751 {
 752         struct ubd *ubd_dev;
 753         int n, len = 0;
 754
 755         n = parse_unit(&name);
 756         if((n >= MAX_DEV) || (n < 0)){
 757                 *error_out = "ubd_get_config : device number out of range";
 758                 return(-1);
 759         }
 760
 761         ubd_dev = &ubd_devs[n];
 762         mutex_lock(&ubd_lock);
 763
 764         if(ubd_dev->file == NULL){
 765                 CONFIG_CHUNK(str, size, len, "", 1);
 766                 goto out;
 767         }
 768
 769         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 770
 771         if(ubd_dev->cow.file != NULL){
 772                 CONFIG_CHUNK(str, size, len, ",", 0);
 773                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 774         }
 775         else CONFIG_CHUNK(str, size, len, "", 1);
 776
 777  out:
 778         mutex_unlock(&ubd_lock);
 779         return(len);
 780 }
 781
 782 static int ubd_id(char **str, int *start_out, int *end_out)
 783 {
 784         int n;
 785
 786         n = parse_unit(str);
 787         *start_out = 0;
 788         *end_out = MAX_DEV - 1;
 789         return n;
 790 }
 791
 792 static int ubd_remove(int n, char **error_out)
 793 {
 794         struct ubd *ubd_dev;
 795         int err = -ENODEV;
 796
 797         mutex_lock(&ubd_lock);
 798
 799         if(ubd_gendisk[n] == NULL)
 800                 goto out;
 801
 802         ubd_dev = &ubd_devs[n];
 803
 804         if(ubd_dev->file == NULL)
 805                 goto out;
 806
 807         /* you cannot remove a open disk */
 808         err = -EBUSY;
 809         if(ubd_dev->count > 0)
 810                 goto out;
 811
 812         del_gendisk(ubd_gendisk[n]);
 813         put_disk(ubd_gendisk[n]);
 814         ubd_gendisk[n] = NULL;
 815
 816         if(fake_gendisk[n] != NULL){
 817                 del_gendisk(fake_gendisk[n]);
 818                 put_disk(fake_gendisk[n]);
 819                 fake_gendisk[n] = NULL;
 820         }
 821
 822         blk_cleanup_queue(ubd_dev->queue);
 823         platform_device_unregister(&ubd_dev->pdev);
 824         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 825         err = 0;
 826 out:
 827         mutex_unlock(&ubd_lock);
 828         return err;
 829 }
 830
 831 /* All these are called by mconsole in process context and without
 832  * ubd-specific locks.  The structure itself is const except for .list.
 833  */
 834 static struct mc_device ubd_mc = {
 835         .list           = LIST_HEAD_INIT(ubd_mc.list),
 836         .name           = "ubd",
 837         .config         = ubd_config,
 838         .get_config     = ubd_get_config,
 839         .id             = ubd_id,
 840         .remove         = ubd_remove,
 841 };
 842
 843 static int __init ubd_mc_init(void)
 844 {
 845         mconsole_register_dev(&ubd_mc);
 846         return 0;
 847 }
 848
 849 __initcall(ubd_mc_init);
 850
 851 static int __init ubd0_init(void)
 852 {
 853         struct ubd *ubd_dev = &ubd_devs[0];
 854
 855         mutex_lock(&ubd_lock);
 856         if(ubd_dev->file == NULL)
 857                 ubd_dev->file = "root_fs";
 858         mutex_unlock(&ubd_lock);
 859
 860         return(0);
 861 }
 862
 863 __initcall(ubd0_init);
 864
 865 /* Used in ubd_init, which is an initcall */
 866 static struct platform_driver ubd_driver = {
 867         .driver = {
 868                 .name  = DRIVER_NAME,
 869         },
 870 };
 871
 872 static int __init ubd_init(void)
 873 {
 874         char *error;
 875         int i, err;
 876
 877         if (register_blkdev(MAJOR_NR, "ubd"))
 878                 return -1;
 879
 880         if (fake_major != MAJOR_NR) {
 881                 char name[sizeof("ubd_nnn\0")];
 882
 883                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
 884                 if (register_blkdev(fake_major, "ubd"))
 885                         return -1;
 886         }
 887         platform_driver_register(&ubd_driver);
 888         mutex_lock(&ubd_lock);
 889         for (i = 0; i < MAX_DEV; i++){
 890                 err = ubd_add(i, &error);
 891                 if(err)
 892                         printk(KERN_ERR "Failed to initialize ubd device %d :"
 893                                "%s\n", i, error);
 894         }
 895         mutex_unlock(&ubd_lock);
 896         return 0;
 897 }
 898
 899 late_initcall(ubd_init);
 900
 901 static int __init ubd_driver_init(void){
 902         unsigned long stack;
 903         int err;
 904
 905         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
 906         if(global_openflags.s){
 907                 printk(KERN_INFO "ubd: Synchronous mode\n");
 908                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
 909                  * enough. So use anyway the io thread. */
 910         }
 911         stack = alloc_stack(0, 0);
 912         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
 913                                  &thread_fd);
 914         if(io_pid < 0){
 915                 printk(KERN_ERR
 916                        "ubd : Failed to start I/O thread (errno = %d) - "
 917                        "falling back to synchronous I/O\n", -io_pid);
 918                 io_pid = -1;
 919                 return(0);
 920         }
 921         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
 922                              IRQF_DISABLED, "ubd", ubd_devs);
 923         if(err != 0)
 924                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
 925         return 0;
 926 }
 927
 928 device_initcall(ubd_driver_init);
 929
 930 static int ubd_open(struct inode *inode, struct file *filp)
 931 {
 932         struct gendisk *disk = inode->i_bdev->bd_disk;
 933         struct ubd *ubd_dev = disk->private_data;
 934         int err = 0;
 935
 936         if(ubd_dev->count == 0){
 937                 err = ubd_open_dev(ubd_dev);
 938                 if(err){
 939                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
 940                                disk->disk_name, ubd_dev->file, -err);
 941                         goto out;
 942                 }
 943         }
 944         ubd_dev->count++;
 945         set_disk_ro(disk, !ubd_dev->openflags.w);
 946
 947         /* This should no more be needed. And it didn't work anyway to exclude
 948          * read-write remounting of filesystems.*/
 949         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
 950                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
 951                 err = -EROFS;
 952         }*/
 953  out:
 954         return(err);
 955 }
 956
 957 static int ubd_release(struct inode * inode, struct file * file)
 958 {
 959         struct gendisk *disk = inode->i_bdev->bd_disk;
 960         struct ubd *ubd_dev = disk->private_data;
 961
 962         if(--ubd_dev->count == 0)
 963                 ubd_close_dev(ubd_dev);
 964         return(0);
 965 }
 966
 967 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
 968                           __u64 *cow_offset, unsigned long *bitmap,
 969                           __u64 bitmap_offset, unsigned long *bitmap_words,
 970                           __u64 bitmap_len)
 971 {
 972         __u64 sector = io_offset >> 9;
 973         int i, update_bitmap = 0;
 974
 975         for(i = 0; i < length >> 9; i++){
 976                 if(cow_mask != NULL)
 977                         ubd_set_bit(i, (unsigned char *) cow_mask);
 978                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
 979                         continue;
 980
 981                 update_bitmap = 1;
 982                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
 983         }
 984
 985         if(!update_bitmap)
 986                 return;
 987
 988         *cow_offset = sector / (sizeof(unsigned long) * 8);
 989
 990         /* This takes care of the case where we're exactly at the end of the
 991          * device, and *cow_offset + 1 is off the end.  So, just back it up
 992          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
 993          * for the original diagnosis.
 994          */
 995         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
 996                            sizeof(unsigned long) - 1))
 997                 (*cow_offset)--;
 998
 999         bitmap_words[0] = bitmap[*cow_offset];
1000         bitmap_words[1] = bitmap[*cow_offset + 1];
1001
1002         *cow_offset *= sizeof(unsigned long);
1003         *cow_offset += bitmap_offset;
1004 }
1005
1006 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1007                        __u64 bitmap_offset, __u64 bitmap_len)
1008 {
1009         __u64 sector = req->offset >> 9;
1010         int i;
1011
1012         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1013                 panic("Operation too long");
1014
1015         if(req->op == UBD_READ) {
1016                 for(i = 0; i < req->length >> 9; i++){
1017                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1018                                 ubd_set_bit(i, (unsigned char *)
1019                                             &req->sector_mask);
1020                 }
1021         }
1022         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1023                            &req->cow_offset, bitmap, bitmap_offset,
1024                            req->bitmap_words, bitmap_len);
1025 }
1026
1027 /* Called with dev->lock held */
1028 static int prepare_request(struct request *req, struct io_thread_req *io_req)
1029 {
1030         struct gendisk *disk = req->rq_disk;
1031         struct ubd *ubd_dev = disk->private_data;
1032         __u64 offset;
1033         int len;
1034
1035         /* This should be impossible now */
1036         if((rq_data_dir(req) == WRITE) && !ubd_dev->openflags.w){
1037                 printk("Write attempted on readonly ubd device %s\n",
1038                        disk->disk_name);
1039                 end_request(req, 0);
1040                 return(1);
1041         }
1042
1043         offset = ((__u64) req->sector) << 9;
1044         len = req->current_nr_sectors << 9;
1045
1046         io_req->req = req;
1047         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd : ubd_dev->fd;
1048         io_req->fds[1] = ubd_dev->fd;
1049         io_req->cow_offset = -1;
1050         io_req->offset = offset;
1051         io_req->length = len;
1052         io_req->error = 0;
1053         io_req->sector_mask = 0;
1054
1055         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1056         io_req->offsets[0] = 0;
1057         io_req->offsets[1] = ubd_dev->cow.data_offset;
1058         io_req->buffer = req->buffer;
1059         io_req->sectorsize = 1 << 9;
1060
1061         if(ubd_dev->cow.file != NULL)
1062                 cowify_req(io_req, ubd_dev->cow.bitmap, ubd_dev->cow.bitmap_offset,
1063                            ubd_dev->cow.bitmap_len);
1064
1065         return(0);
1066 }
1067
1068 /* Called with dev->lock held */
1069 static void do_ubd_request(request_queue_t *q)
1070 {
1071         struct io_thread_req io_req;
1072         struct request *req;
1073         int err, n;
1074
1075         if(thread_fd == -1){
1076                 while((req = elv_next_request(q)) != NULL){
1077                         err = prepare_request(req, &io_req);
1078                         if(!err){
1079                                 do_io(&io_req);
1080                                 __ubd_finish(req, io_req.error);
1081                         }
1082                 }
1083         }
1084         else {
1085                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1086                         return;
1087                 err = prepare_request(req, &io_req);
1088                 if(!err){
1089                         do_ubd = 1;
1090                         n = os_write_file(thread_fd, (char *) &io_req,
1091                                          sizeof(io_req));
1092                         if(n != sizeof(io_req))
1093                                 printk("write to io thread failed, "
1094                                        "errno = %d\n", -n);
1095                 }
1096         }
1097 }
1098
1099 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1100 {
1101         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1102
1103         geo->heads = 128;
1104         geo->sectors = 32;
1105         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1106         return 0;
1107 }
1108
1109 static int ubd_ioctl(struct inode * inode, struct file * file,
1110                      unsigned int cmd, unsigned long arg)
1111 {
1112         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1113         struct hd_driveid ubd_id = {
1114                 .cyls           = 0,
1115                 .heads          = 128,
1116                 .sectors        = 32,
1117         };
1118
1119         switch (cmd) {
1120                 struct cdrom_volctrl volume;
1121         case HDIO_GET_IDENTITY:
1122                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1123                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1124                                  sizeof(ubd_id)))
1125                         return(-EFAULT);
1126                 return(0);
1127
1128         case CDROMVOLREAD:
1129                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1130                         return(-EFAULT);
1131                 volume.channel0 = 255;
1132                 volume.channel1 = 255;
1133                 volume.channel2 = 255;
1134                 volume.channel3 = 255;
1135                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1136                         return(-EFAULT);
1137                 return(0);
1138         }
1139         return(-EINVAL);
1140 }
1141
1142 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1143 {
1144         struct uml_stat buf1, buf2;
1145         int err;
1146
1147         if(from_cmdline == NULL)
1148                 return 0;
1149         if(!strcmp(from_cmdline, from_cow))
1150                 return 0;
1151
1152         err = os_stat_file(from_cmdline, &buf1);
1153         if(err < 0){
1154                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1155                 return 0;
1156         }
1157         err = os_stat_file(from_cow, &buf2);
1158         if(err < 0){
1159                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1160                 return 1;
1161         }
1162         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1163                 return 0;
1164
1165         printk("Backing file mismatch - \"%s\" requested,\n"
1166                "\"%s\" specified in COW header of \"%s\"\n",
1167                from_cmdline, from_cow, cow);
1168         return 1;
1169 }
1170
1171 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1172 {
1173         unsigned long modtime;
1174         unsigned long long actual;
1175         int err;
1176
1177         err = os_file_modtime(file, &modtime);
1178         if(err < 0){
1179                 printk("Failed to get modification time of backing file "
1180                        "\"%s\", err = %d\n", file, -err);
1181                 return(err);
1182         }
1183
1184         err = os_file_size(file, &actual);
1185         if(err < 0){
1186                 printk("Failed to get size of backing file \"%s\", "
1187                        "err = %d\n", file, -err);
1188                 return(err);
1189         }
1190
1191         if(actual != size){
1192                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1193                  * the typecast.*/
1194                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1195                        "file\n", (unsigned long long) size, actual);
1196                 return(-EINVAL);
1197         }
1198         if(modtime != mtime){
1199                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1200                        "file\n", mtime, modtime);
1201                 return(-EINVAL);
1202         }
1203         return(0);
1204 }
1205
1206 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1207 {
1208         int err;
1209
1210         err = os_seek_file(fd, offset);
1211         if(err < 0)
1212                 return(err);
1213
1214         err = os_read_file(fd, buf, len);
1215         if(err < 0)
1216                 return(err);
1217
1218         return(0);
1219 }
1220
1221 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1222                   char **backing_file_out, int *bitmap_offset_out,
1223                   unsigned long *bitmap_len_out, int *data_offset_out,
1224                   int *create_cow_out)
1225 {
1226         time_t mtime;
1227         unsigned long long size;
1228         __u32 version, align;
1229         char *backing_file;
1230         int fd, err, sectorsize, asked_switch, mode = 0644;
1231
1232         fd = os_open_file(file, *openflags, mode);
1233         if (fd < 0) {
1234                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1235                         *create_cow_out = 1;
1236                 if (!openflags->w ||
1237                    ((fd != -EROFS) && (fd != -EACCES)))
1238                         return fd;
1239                 openflags->w = 0;
1240                 fd = os_open_file(file, *openflags, mode);
1241                 if (fd < 0)
1242                         return fd;
1243         }
1244
1245         if(shared)
1246                 printk("Not locking \"%s\" on the host\n", file);
1247         else {
1248                 err = os_lock_file(fd, openflags->w);
1249                 if(err < 0){
1250                         printk("Failed to lock '%s', err = %d\n", file, -err);
1251                         goto out_close;
1252                 }
1253         }
1254
1255         /* Successful return case! */
1256         if(backing_file_out == NULL)
1257                 return(fd);
1258
1259         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1260                               &size, &sectorsize, &align, bitmap_offset_out);
1261         if(err && (*backing_file_out != NULL)){
1262                 printk("Failed to read COW header from COW file \"%s\", "
1263                        "errno = %d\n", file, -err);
1264                 goto out_close;
1265         }
1266         if(err)
1267                 return(fd);
1268
1269         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1270
1271         /* Allow switching only if no mismatch. */
1272         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1273                 printk("Switching backing file to '%s'\n", *backing_file_out);
1274                 err = write_cow_header(file, fd, *backing_file_out,
1275                                        sectorsize, align, &size);
1276                 if (err) {
1277                         printk("Switch failed, errno = %d\n", -err);
1278                         goto out_close;
1279                 }
1280         } else {
1281                 *backing_file_out = backing_file;
1282                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1283                 if (err)
1284                         goto out_close;
1285         }
1286
1287         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1288                   bitmap_len_out, data_offset_out);
1289
1290         return fd;
1291  out_close:
1292         os_close_file(fd);
1293         return err;
1294 }
1295
1296 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1297                     int sectorsize, int alignment, int *bitmap_offset_out,
1298                     unsigned long *bitmap_len_out, int *data_offset_out)
1299 {
1300         int err, fd;
1301
1302         flags.c = 1;
1303         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1304         if(fd < 0){
1305                 err = fd;
1306                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1307                        -err);
1308                 goto out;
1309         }
1310
1311         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1312                             bitmap_offset_out, bitmap_len_out,
1313                             data_offset_out);
1314         if(!err)
1315                 return(fd);
1316         os_close_file(fd);
1317  out:
1318         return(err);
1319 }
1320
1321 static int update_bitmap(struct io_thread_req *req)
1322 {
1323         int n;
1324
1325         if(req->cow_offset == -1)
1326                 return(0);
1327
1328         n = os_seek_file(req->fds[1], req->cow_offset);
1329         if(n < 0){
1330                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1331                 return(1);
1332         }
1333
1334         n = os_write_file(req->fds[1], &req->bitmap_words,
1335                           sizeof(req->bitmap_words));
1336         if(n != sizeof(req->bitmap_words)){
1337                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1338                        req->fds[1]);
1339                 return(1);
1340         }
1341
1342         return(0);
1343 }
1344
1345 void do_io(struct io_thread_req *req)
1346 {
1347         char *buf;
1348         unsigned long len;
1349         int n, nsectors, start, end, bit;
1350         int err;
1351         __u64 off;
1352
1353         nsectors = req->length / req->sectorsize;
1354         start = 0;
1355         do {
1356                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1357                 end = start;
1358                 while((end < nsectors) &&
1359                       (ubd_test_bit(end, (unsigned char *)
1360                                     &req->sector_mask) == bit))
1361                         end++;
1362
1363                 off = req->offset + req->offsets[bit] +
1364                         start * req->sectorsize;
1365                 len = (end - start) * req->sectorsize;
1366                 buf = &req->buffer[start * req->sectorsize];
1367
1368                 err = os_seek_file(req->fds[bit], off);
1369                 if(err < 0){
1370                         printk("do_io - lseek failed : err = %d\n", -err);
1371                         req->error = 1;
1372                         return;
1373                 }
1374                 if(req->op == UBD_READ){
1375                         n = 0;
1376                         do {
1377                                 buf = &buf[n];
1378                                 len -= n;
1379                                 n = os_read_file(req->fds[bit], buf, len);
1380                                 if (n < 0) {
1381                                         printk("do_io - read failed, err = %d "
1382                                                "fd = %d\n", -n, req->fds[bit]);
1383                                         req->error = 1;
1384                                         return;
1385                                 }
1386                         } while((n < len) && (n != 0));
1387                         if (n < len) memset(&buf[n], 0, len - n);
1388                 } else {
1389                         n = os_write_file(req->fds[bit], buf, len);
1390                         if(n != len){
1391                                 printk("do_io - write failed err = %d "
1392                                        "fd = %d\n", -n, req->fds[bit]);
1393                                 req->error = 1;
1394                                 return;
1395                         }
1396                 }
1397
1398                 start = end;
1399         } while(start < nsectors);
1400
1401         req->error = update_bitmap(req);
1402 }
1403
1404 /* Changed in start_io_thread, which is serialized by being called only
1405  * from ubd_init, which is an initcall.
1406  */
1407 int kernel_fd = -1;
1408
1409 /* Only changed by the io thread. XXX: currently unused. */
1410 static int io_count = 0;
1411
1412 int io_thread(void *arg)
1413 {
1414         struct io_thread_req req;
1415         int n;
1416
1417         ignore_sigwinch_sig();
1418         while(1){
1419                 n = os_read_file(kernel_fd, &req, sizeof(req));
1420                 if(n != sizeof(req)){
1421                         if(n < 0)
1422                                 printk("io_thread - read failed, fd = %d, "
1423                                        "err = %d\n", kernel_fd, -n);
1424                         else {
1425                                 printk("io_thread - short read, fd = %d, "
1426                                        "length = %d\n", kernel_fd, n);
1427                         }
1428                         continue;
1429                 }
1430                 io_count++;
1431                 do_io(&req);
1432                 n = os_write_file(kernel_fd, &req, sizeof(req));
1433                 if(n != sizeof(req))
1434                         printk("io_thread - write failed, fd = %d, err = %d\n",
1435                                kernel_fd, -n);
1436         }
1437
1438         return 0;
1439 }