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 "mem_user.h"
  43 #include "kern_util.h"
  44 #include "kern.h"
  45 #include "mconsole_kern.h"
  46 #include "init.h"
  47 #include "irq_user.h"
  48 #include "irq_kern.h"
  49 #include "ubd_user.h"
  50 #include "os.h"
  51 #include "mem.h"
  52 #include "mem_kern.h"
  53 #include "cow.h"
  54
  55 enum ubd_req { UBD_READ, UBD_WRITE };
  56
  57 struct io_thread_req {
  58         struct request *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 static DEFINE_MUTEX(ubd_lock);
 110
 111 static int ubd_open(struct inode * inode, struct file * filp);
 112 static int ubd_release(struct inode * inode, struct file * file);
 113 static int ubd_ioctl(struct inode * inode, struct file * file,
 114                      unsigned int cmd, unsigned long arg);
 115 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 116
 117 #define MAX_DEV (16)
 118
 119 static struct block_device_operations ubd_blops = {
 120         .owner          = THIS_MODULE,
 121         .open           = ubd_open,
 122         .release        = ubd_release,
 123         .ioctl          = ubd_ioctl,
 124         .getgeo         = ubd_getgeo,
 125 };
 126
 127 /* Protected by ubd_lock */
 128 static int fake_major = MAJOR_NR;
 129 static struct gendisk *ubd_gendisk[MAX_DEV];
 130 static struct gendisk *fake_gendisk[MAX_DEV];
 131
 132 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 133 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 134                                          .cl = 1 })
 135 #else
 136 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 137                                          .cl = 1 })
 138 #endif
 139 static struct openflags global_openflags = OPEN_FLAGS;
 140
 141 struct cow {
 142         /* backing file name */
 143         char *file;
 144         /* backing file fd */
 145         int fd;
 146         unsigned long *bitmap;
 147         unsigned long bitmap_len;
 148         int bitmap_offset;
 149         int data_offset;
 150 };
 151
 152 #define MAX_SG 64
 153
 154 struct ubd {
 155         struct list_head restart;
 156         /* name (and fd, below) of the file opened for writing, either the
 157          * backing or the cow file. */
 158         char *file;
 159         int count;
 160         int fd;
 161         __u64 size;
 162         struct openflags boot_openflags;
 163         struct openflags openflags;
 164         unsigned shared:1;
 165         unsigned no_cow:1;
 166         struct cow cow;
 167         struct platform_device pdev;
 168         struct request_queue *queue;
 169         spinlock_t lock;
 170         struct scatterlist sg[MAX_SG];
 171         struct request *request;
 172         int start_sg, end_sg;
 173 };
 174
 175 #define DEFAULT_COW { \
 176         .file =                 NULL, \
 177         .fd =                   -1, \
 178         .bitmap =               NULL, \
 179         .bitmap_offset =        0, \
 180         .data_offset =          0, \
 181 }
 182
 183 #define DEFAULT_UBD { \
 184         .file =                 NULL, \
 185         .count =                0, \
 186         .fd =                   -1, \
 187         .size =                 -1, \
 188         .boot_openflags =       OPEN_FLAGS, \
 189         .openflags =            OPEN_FLAGS, \
 190         .no_cow =               0, \
 191         .shared =               0, \
 192         .cow =                  DEFAULT_COW, \
 193         .lock =                 SPIN_LOCK_UNLOCKED,     \
 194         .request =              NULL, \
 195         .start_sg =             0, \
 196         .end_sg =               0, \
 197 }
 198
 199 /* Protected by ubd_lock */
 200 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
 201
 202 /* Only changed by fake_ide_setup which is a setup */
 203 static int fake_ide = 0;
 204 static struct proc_dir_entry *proc_ide_root = NULL;
 205 static struct proc_dir_entry *proc_ide = NULL;
 206
 207 static void make_proc_ide(void)
 208 {
 209         proc_ide_root = proc_mkdir("ide", NULL);
 210         proc_ide = proc_mkdir("ide0", proc_ide_root);
 211 }
 212
 213 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
 214                                int *eof, void *data)
 215 {
 216         int len;
 217
 218         strcpy(page, "disk\n");
 219         len = strlen("disk\n");
 220         len -= off;
 221         if (len < count){
 222                 *eof = 1;
 223                 if (len <= 0) return 0;
 224         }
 225         else len = count;
 226         *start = page + off;
 227         return len;
 228 }
 229
 230 static void make_ide_entries(char *dev_name)
 231 {
 232         struct proc_dir_entry *dir, *ent;
 233         char name[64];
 234
 235         if(proc_ide_root == NULL) make_proc_ide();
 236
 237         dir = proc_mkdir(dev_name, proc_ide);
 238         if(!dir) return;
 239
 240         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
 241         if(!ent) return;
 242         ent->data = NULL;
 243         ent->read_proc = proc_ide_read_media;
 244         ent->write_proc = NULL;
 245         sprintf(name,"ide0/%s", dev_name);
 246         proc_symlink(dev_name, proc_ide_root, name);
 247 }
 248
 249 static int fake_ide_setup(char *str)
 250 {
 251         fake_ide = 1;
 252         return(1);
 253 }
 254
 255 __setup("fake_ide", fake_ide_setup);
 256
 257 __uml_help(fake_ide_setup,
 258 "fake_ide\n"
 259 "    Create ide0 entries that map onto ubd devices.\n\n"
 260 );
 261
 262 static int parse_unit(char **ptr)
 263 {
 264         char *str = *ptr, *end;
 265         int n = -1;
 266
 267         if(isdigit(*str)) {
 268                 n = simple_strtoul(str, &end, 0);
 269                 if(end == str)
 270                         return(-1);
 271                 *ptr = end;
 272         }
 273         else if (('a' <= *str) && (*str <= 'z')) {
 274                 n = *str - 'a';
 275                 str++;
 276                 *ptr = str;
 277         }
 278         return(n);
 279 }
 280
 281 /* If *index_out == -1 at exit, the passed option was a general one;
 282  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 283  * should not be freed on exit.
 284  */
 285 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 286 {
 287         struct ubd *ubd_dev;
 288         struct openflags flags = global_openflags;
 289         char *backing_file;
 290         int n, err = 0, i;
 291
 292         if(index_out) *index_out = -1;
 293         n = *str;
 294         if(n == '='){
 295                 char *end;
 296                 int major;
 297
 298                 str++;
 299                 if(!strcmp(str, "sync")){
 300                         global_openflags = of_sync(global_openflags);
 301                         goto out1;
 302                 }
 303
 304                 err = -EINVAL;
 305                 major = simple_strtoul(str, &end, 0);
 306                 if((*end != '\0') || (end == str)){
 307                         *error_out = "Didn't parse major number";
 308                         goto out1;
 309                 }
 310
 311                 mutex_lock(&ubd_lock);
 312                 if(fake_major != MAJOR_NR){
 313                         *error_out = "Can't assign a fake major twice";
 314                         goto out1;
 315                 }
 316
 317                 fake_major = major;
 318
 319                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 320                        major);
 321                 err = 0;
 322         out1:
 323                 mutex_unlock(&ubd_lock);
 324                 return err;
 325         }
 326
 327         n = parse_unit(&str);
 328         if(n < 0){
 329                 *error_out = "Couldn't parse device number";
 330                 return -EINVAL;
 331         }
 332         if(n >= MAX_DEV){
 333                 *error_out = "Device number out of range";
 334                 return 1;
 335         }
 336
 337         err = -EBUSY;
 338         mutex_lock(&ubd_lock);
 339
 340         ubd_dev = &ubd_devs[n];
 341         if(ubd_dev->file != NULL){
 342                 *error_out = "Device is already configured";
 343                 goto out;
 344         }
 345
 346         if (index_out)
 347                 *index_out = n;
 348
 349         err = -EINVAL;
 350         for (i = 0; i < sizeof("rscd="); i++) {
 351                 switch (*str) {
 352                 case 'r':
 353                         flags.w = 0;
 354                         break;
 355                 case 's':
 356                         flags.s = 1;
 357                         break;
 358                 case 'd':
 359                         ubd_dev->no_cow = 1;
 360                         break;
 361                 case 'c':
 362                         ubd_dev->shared = 1;
 363                         break;
 364                 case '=':
 365                         str++;
 366                         goto break_loop;
 367                 default:
 368                         *error_out = "Expected '=' or flag letter "
 369                                 "(r, s, c, or d)";
 370                         goto out;
 371                 }
 372                 str++;
 373         }
 374
 375         if (*str == '=')
 376                 *error_out = "Too many flags specified";
 377         else
 378                 *error_out = "Missing '='";
 379         goto out;
 380
 381 break_loop:
 382         backing_file = strchr(str, ',');
 383
 384         if (backing_file == NULL)
 385                 backing_file = strchr(str, ':');
 386
 387         if(backing_file != NULL){
 388                 if(ubd_dev->no_cow){
 389                         *error_out = "Can't specify both 'd' and a cow file";
 390                         goto out;
 391                 }
 392                 else {
 393                         *backing_file = '\0';
 394                         backing_file++;
 395                 }
 396         }
 397         err = 0;
 398         ubd_dev->file = str;
 399         ubd_dev->cow.file = backing_file;
 400         ubd_dev->boot_openflags = flags;
 401 out:
 402         mutex_unlock(&ubd_lock);
 403         return err;
 404 }
 405
 406 static int ubd_setup(char *str)
 407 {
 408         char *error;
 409         int err;
 410
 411         err = ubd_setup_common(str, NULL, &error);
 412         if(err)
 413                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 414                        "%s\n", str, error);
 415         return 1;
 416 }
 417
 418 __setup("ubd", ubd_setup);
 419 __uml_help(ubd_setup,
 420 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 421 "    This is used to associate a device with a file in the underlying\n"
 422 "    filesystem. When specifying two filenames, the first one is the\n"
 423 "    COW name and the second is the backing file name. As separator you can\n"
 424 "    use either a ':' or a ',': the first one allows writing things like;\n"
 425 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 426 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 427 "    When using only one filename, UML will detect whether to treat it like\n"
 428 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 429 "    flag:\n"
 430 "       ubd0d=BackingFile\n"
 431 "    Usually, there is a filesystem in the file, but \n"
 432 "    that's not required. Swap devices containing swap files can be\n"
 433 "    specified like this. Also, a file which doesn't contain a\n"
 434 "    filesystem can have its contents read in the virtual \n"
 435 "    machine by running 'dd' on the device. <n> must be in the range\n"
 436 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 437 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 438 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
 439 );
 440
 441 static int udb_setup(char *str)
 442 {
 443         printk("udb%s specified on command line is almost certainly a ubd -> "
 444                "udb TYPO\n", str);
 445         return(1);
 446 }
 447
 448 __setup("udb", udb_setup);
 449 __uml_help(udb_setup,
 450 "udb\n"
 451 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 452 "    to impossible to catch visually unless you specifically look for\n"
 453 "    them.  The only result of any option starting with 'udb' is an error\n"
 454 "    in the boot output.\n\n"
 455 );
 456
 457 static int fakehd_set = 0;
 458 static int fakehd(char *str)
 459 {
 460         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
 461         fakehd_set = 1;
 462         return 1;
 463 }
 464
 465 __setup("fakehd", fakehd);
 466 __uml_help(fakehd,
 467 "fakehd\n"
 468 "    Change the ubd device name to \"hd\".\n\n"
 469 );
 470
 471 static void do_ubd_request(request_queue_t * q);
 472
 473 /* Only changed by ubd_init, which is an initcall. */
 474 int thread_fd = -1;
 475
 476 static void ubd_end_request(struct request *req, int bytes, int uptodate)
 477 {
 478         if (!end_that_request_first(req, uptodate, bytes >> 9)) {
 479                 struct ubd *dev = req->rq_disk->private_data;
 480                 unsigned long flags;
 481
 482                 add_disk_randomness(req->rq_disk);
 483                 spin_lock_irqsave(&dev->lock, flags);
 484                 end_that_request_last(req, uptodate);
 485                 spin_unlock_irqrestore(&dev->lock, flags);
 486         }
 487 }
 488
 489 /* Callable only from interrupt context - otherwise you need to do
 490  * spin_lock_irq()/spin_lock_irqsave() */
 491 static inline void ubd_finish(struct request *req, int bytes)
 492 {
 493         if(bytes < 0){
 494                 ubd_end_request(req, 0, 0);
 495                 return;
 496         }
 497         ubd_end_request(req, bytes, 1);
 498 }
 499
 500 static LIST_HEAD(restart);
 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 *ubd;
 509         struct list_head *list, *next_ele;
 510         unsigned long flags;
 511         int n;
 512
 513         while(1){
 514                 n = os_read_file_k(thread_fd, &req, sizeof(req));
 515                 if(n != sizeof(req)){
 516                         if(n == -EAGAIN)
 517                                 break;
 518                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 519                                "err = %d\n", -n);
 520                         return;
 521                 }
 522
 523                 rq = req.req;
 524                 rq->nr_sectors -= req.length >> 9;
 525                 if(rq->nr_sectors == 0)
 526                         ubd_finish(rq, rq->hard_nr_sectors << 9);
 527         }
 528         reactivate_fd(thread_fd, UBD_IRQ);
 529
 530         list_for_each_safe(list, next_ele, &restart){
 531                 ubd = container_of(list, struct ubd, restart);
 532                 list_del_init(&ubd->restart);
 533                 spin_lock_irqsave(&ubd->lock, flags);
 534                 do_ubd_request(ubd->queue);
 535                 spin_unlock_irqrestore(&ubd->lock, flags);
 536         }
 537 }
 538
 539 static irqreturn_t ubd_intr(int irq, void *dev)
 540 {
 541         ubd_handler();
 542         return(IRQ_HANDLED);
 543 }
 544
 545 /* Only changed by ubd_init, which is an initcall. */
 546 static int io_pid = -1;
 547
 548 void kill_io_thread(void)
 549 {
 550         if(io_pid != -1)
 551                 os_kill_process(io_pid, 1);
 552 }
 553
 554 __uml_exitcall(kill_io_thread);
 555
 556 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 557 {
 558         char *file;
 559
 560         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
 561         return(os_file_size(file, size_out));
 562 }
 563
 564 static void ubd_close_dev(struct ubd *ubd_dev)
 565 {
 566         os_close_file(ubd_dev->fd);
 567         if(ubd_dev->cow.file == NULL)
 568                 return;
 569
 570         os_close_file(ubd_dev->cow.fd);
 571         vfree(ubd_dev->cow.bitmap);
 572         ubd_dev->cow.bitmap = NULL;
 573 }
 574
 575 static int ubd_open_dev(struct ubd *ubd_dev)
 576 {
 577         struct openflags flags;
 578         char **back_ptr;
 579         int err, create_cow, *create_ptr;
 580         int fd;
 581
 582         ubd_dev->openflags = ubd_dev->boot_openflags;
 583         create_cow = 0;
 584         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 585         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 586
 587         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 588                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 589                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 590                                 create_ptr);
 591
 592         if((fd == -ENOENT) && create_cow){
 593                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 594                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 595                                           &ubd_dev->cow.bitmap_offset,
 596                                           &ubd_dev->cow.bitmap_len,
 597                                           &ubd_dev->cow.data_offset);
 598                 if(fd >= 0){
 599                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 600                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 601                 }
 602         }
 603
 604         if(fd < 0){
 605                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 606                        -fd);
 607                 return fd;
 608         }
 609         ubd_dev->fd = fd;
 610
 611         if(ubd_dev->cow.file != NULL){
 612                 err = -ENOMEM;
 613                 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
 614                 if(ubd_dev->cow.bitmap == NULL){
 615                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 616                         goto error;
 617                 }
 618                 flush_tlb_kernel_vm();
 619
 620                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 621                                       ubd_dev->cow.bitmap_offset,
 622                                       ubd_dev->cow.bitmap_len);
 623                 if(err < 0)
 624                         goto error;
 625
 626                 flags = ubd_dev->openflags;
 627                 flags.w = 0;
 628                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 629                                     NULL, NULL, NULL, NULL);
 630                 if(err < 0) goto error;
 631                 ubd_dev->cow.fd = err;
 632         }
 633         return(0);
 634  error:
 635         os_close_file(ubd_dev->fd);
 636         return(err);
 637 }
 638
 639 static void ubd_device_release(struct device *dev)
 640 {
 641         struct ubd *ubd_dev = dev->driver_data;
 642
 643         blk_cleanup_queue(ubd_dev->queue);
 644         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 645 }
 646
 647 static int ubd_disk_register(int major, u64 size, int unit,
 648                              struct gendisk **disk_out)
 649 {
 650         struct gendisk *disk;
 651
 652         disk = alloc_disk(1 << UBD_SHIFT);
 653         if(disk == NULL)
 654                 return(-ENOMEM);
 655
 656         disk->major = major;
 657         disk->first_minor = unit << UBD_SHIFT;
 658         disk->fops = &ubd_blops;
 659         set_capacity(disk, size / 512);
 660         if(major == MAJOR_NR)
 661                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 662         else
 663                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 664
 665         /* sysfs register (not for ide fake devices) */
 666         if (major == MAJOR_NR) {
 667                 ubd_devs[unit].pdev.id   = unit;
 668                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 669                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 670                 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
 671                 platform_device_register(&ubd_devs[unit].pdev);
 672                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 673         }
 674
 675         disk->private_data = &ubd_devs[unit];
 676         disk->queue = ubd_devs[unit].queue;
 677         add_disk(disk);
 678
 679         *disk_out = disk;
 680         return 0;
 681 }
 682
 683 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 684
 685 static int ubd_add(int n, char **error_out)
 686 {
 687         struct ubd *ubd_dev = &ubd_devs[n];
 688         int err = 0;
 689
 690         if(ubd_dev->file == NULL)
 691                 goto out;
 692
 693         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 694         if(err < 0){
 695                 *error_out = "Couldn't determine size of device's file";
 696                 goto out;
 697         }
 698
 699         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 700
 701         INIT_LIST_HEAD(&ubd_dev->restart);
 702
 703         err = -ENOMEM;
 704         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 705         if (ubd_dev->queue == NULL) {
 706                 *error_out = "Failed to initialize device queue";
 707                 goto out;
 708         }
 709         ubd_dev->queue->queuedata = ubd_dev;
 710
 711         blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG);
 712         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
 713         if(err){
 714                 *error_out = "Failed to register device";
 715                 goto out_cleanup;
 716         }
 717
 718         if(fake_major != MAJOR_NR)
 719                 ubd_disk_register(fake_major, ubd_dev->size, n,
 720                                   &fake_gendisk[n]);
 721
 722         /* perhaps this should also be under the "if (fake_major)" above */
 723         /* using the fake_disk->disk_name and also the fakehd_set name */
 724         if (fake_ide)
 725                 make_ide_entries(ubd_gendisk[n]->disk_name);
 726
 727         err = 0;
 728 out:
 729         return err;
 730
 731 out_cleanup:
 732         blk_cleanup_queue(ubd_dev->queue);
 733         goto out;
 734 }
 735
 736 static int ubd_config(char *str, char **error_out)
 737 {
 738         int n, ret;
 739
 740         /* This string is possibly broken up and stored, so it's only
 741          * freed if ubd_setup_common fails, or if only general options
 742          * were set.
 743          */
 744         str = kstrdup(str, GFP_KERNEL);
 745         if (str == NULL) {
 746                 *error_out = "Failed to allocate memory";
 747                 return -ENOMEM;
 748         }
 749
 750         ret = ubd_setup_common(str, &n, error_out);
 751         if (ret)
 752                 goto err_free;
 753
 754         if (n == -1) {
 755                 ret = 0;
 756                 goto err_free;
 757         }
 758
 759         mutex_lock(&ubd_lock);
 760         ret = ubd_add(n, error_out);
 761         if (ret)
 762                 ubd_devs[n].file = NULL;
 763         mutex_unlock(&ubd_lock);
 764
 765 out:
 766         return ret;
 767
 768 err_free:
 769         kfree(str);
 770         goto out;
 771 }
 772
 773 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 774 {
 775         struct ubd *ubd_dev;
 776         int n, len = 0;
 777
 778         n = parse_unit(&name);
 779         if((n >= MAX_DEV) || (n < 0)){
 780                 *error_out = "ubd_get_config : device number out of range";
 781                 return(-1);
 782         }
 783
 784         ubd_dev = &ubd_devs[n];
 785         mutex_lock(&ubd_lock);
 786
 787         if(ubd_dev->file == NULL){
 788                 CONFIG_CHUNK(str, size, len, "", 1);
 789                 goto out;
 790         }
 791
 792         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 793
 794         if(ubd_dev->cow.file != NULL){
 795                 CONFIG_CHUNK(str, size, len, ",", 0);
 796                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 797         }
 798         else CONFIG_CHUNK(str, size, len, "", 1);
 799
 800  out:
 801         mutex_unlock(&ubd_lock);
 802         return(len);
 803 }
 804
 805 static int ubd_id(char **str, int *start_out, int *end_out)
 806 {
 807         int n;
 808
 809         n = parse_unit(str);
 810         *start_out = 0;
 811         *end_out = MAX_DEV - 1;
 812         return n;
 813 }
 814
 815 static int ubd_remove(int n, char **error_out)
 816 {
 817         struct gendisk *disk = ubd_gendisk[n];
 818         struct ubd *ubd_dev;
 819         int err = -ENODEV;
 820
 821         mutex_lock(&ubd_lock);
 822
 823         ubd_dev = &ubd_devs[n];
 824
 825         if(ubd_dev->file == NULL)
 826                 goto out;
 827
 828         /* you cannot remove a open disk */
 829         err = -EBUSY;
 830         if(ubd_dev->count > 0)
 831                 goto out;
 832
 833         ubd_gendisk[n] = NULL;
 834         if(disk != NULL){
 835                 del_gendisk(disk);
 836                 put_disk(disk);
 837         }
 838
 839         if(fake_gendisk[n] != NULL){
 840                 del_gendisk(fake_gendisk[n]);
 841                 put_disk(fake_gendisk[n]);
 842                 fake_gendisk[n] = NULL;
 843         }
 844
 845         err = 0;
 846         platform_device_unregister(&ubd_dev->pdev);
 847 out:
 848         mutex_unlock(&ubd_lock);
 849         return err;
 850 }
 851
 852 /* All these are called by mconsole in process context and without
 853  * ubd-specific locks.  The structure itself is const except for .list.
 854  */
 855 static struct mc_device ubd_mc = {
 856         .list           = LIST_HEAD_INIT(ubd_mc.list),
 857         .name           = "ubd",
 858         .config         = ubd_config,
 859         .get_config     = ubd_get_config,
 860         .id             = ubd_id,
 861         .remove         = ubd_remove,
 862 };
 863
 864 static int __init ubd_mc_init(void)
 865 {
 866         mconsole_register_dev(&ubd_mc);
 867         return 0;
 868 }
 869
 870 __initcall(ubd_mc_init);
 871
 872 static int __init ubd0_init(void)
 873 {
 874         struct ubd *ubd_dev = &ubd_devs[0];
 875
 876         mutex_lock(&ubd_lock);
 877         if(ubd_dev->file == NULL)
 878                 ubd_dev->file = "root_fs";
 879         mutex_unlock(&ubd_lock);
 880
 881         return(0);
 882 }
 883
 884 __initcall(ubd0_init);
 885
 886 /* Used in ubd_init, which is an initcall */
 887 static struct platform_driver ubd_driver = {
 888         .driver = {
 889                 .name  = DRIVER_NAME,
 890         },
 891 };
 892
 893 static int __init ubd_init(void)
 894 {
 895         char *error;
 896         int i, err;
 897
 898         if (register_blkdev(MAJOR_NR, "ubd"))
 899                 return -1;
 900
 901         if (fake_major != MAJOR_NR) {
 902                 char name[sizeof("ubd_nnn\0")];
 903
 904                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
 905                 if (register_blkdev(fake_major, "ubd"))
 906                         return -1;
 907         }
 908         platform_driver_register(&ubd_driver);
 909         mutex_lock(&ubd_lock);
 910         for (i = 0; i < MAX_DEV; i++){
 911                 err = ubd_add(i, &error);
 912                 if(err)
 913                         printk(KERN_ERR "Failed to initialize ubd device %d :"
 914                                "%s\n", i, error);
 915         }
 916         mutex_unlock(&ubd_lock);
 917         return 0;
 918 }
 919
 920 late_initcall(ubd_init);
 921
 922 static int __init ubd_driver_init(void){
 923         unsigned long stack;
 924         int err;
 925
 926         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
 927         if(global_openflags.s){
 928                 printk(KERN_INFO "ubd: Synchronous mode\n");
 929                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
 930                  * enough. So use anyway the io thread. */
 931         }
 932         stack = alloc_stack(0, 0);
 933         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
 934                                  &thread_fd);
 935         if(io_pid < 0){
 936                 printk(KERN_ERR
 937                        "ubd : Failed to start I/O thread (errno = %d) - "
 938                        "falling back to synchronous I/O\n", -io_pid);
 939                 io_pid = -1;
 940                 return(0);
 941         }
 942         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
 943                              IRQF_DISABLED, "ubd", ubd_devs);
 944         if(err != 0)
 945                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
 946         return 0;
 947 }
 948
 949 device_initcall(ubd_driver_init);
 950
 951 static int ubd_open(struct inode *inode, struct file *filp)
 952 {
 953         struct gendisk *disk = inode->i_bdev->bd_disk;
 954         struct ubd *ubd_dev = disk->private_data;
 955         int err = 0;
 956
 957         if(ubd_dev->count == 0){
 958                 err = ubd_open_dev(ubd_dev);
 959                 if(err){
 960                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
 961                                disk->disk_name, ubd_dev->file, -err);
 962                         goto out;
 963                 }
 964         }
 965         ubd_dev->count++;
 966         set_disk_ro(disk, !ubd_dev->openflags.w);
 967
 968         /* This should no more be needed. And it didn't work anyway to exclude
 969          * read-write remounting of filesystems.*/
 970         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
 971                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
 972                 err = -EROFS;
 973         }*/
 974  out:
 975         return(err);
 976 }
 977
 978 static int ubd_release(struct inode * inode, struct file * file)
 979 {
 980         struct gendisk *disk = inode->i_bdev->bd_disk;
 981         struct ubd *ubd_dev = disk->private_data;
 982
 983         if(--ubd_dev->count == 0)
 984                 ubd_close_dev(ubd_dev);
 985         return(0);
 986 }
 987
 988 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
 989                           __u64 *cow_offset, unsigned long *bitmap,
 990                           __u64 bitmap_offset, unsigned long *bitmap_words,
 991                           __u64 bitmap_len)
 992 {
 993         __u64 sector = io_offset >> 9;
 994         int i, update_bitmap = 0;
 995
 996         for(i = 0; i < length >> 9; i++){
 997                 if(cow_mask != NULL)
 998                         ubd_set_bit(i, (unsigned char *) cow_mask);
 999                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1000                         continue;
1001
1002                 update_bitmap = 1;
1003                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1004         }
1005
1006         if(!update_bitmap)
1007                 return;
1008
1009         *cow_offset = sector / (sizeof(unsigned long) * 8);
1010
1011         /* This takes care of the case where we're exactly at the end of the
1012          * device, and *cow_offset + 1 is off the end.  So, just back it up
1013          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1014          * for the original diagnosis.
1015          */
1016         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
1017                            sizeof(unsigned long) - 1))
1018                 (*cow_offset)--;
1019
1020         bitmap_words[0] = bitmap[*cow_offset];
1021         bitmap_words[1] = bitmap[*cow_offset + 1];
1022
1023         *cow_offset *= sizeof(unsigned long);
1024         *cow_offset += bitmap_offset;
1025 }
1026
1027 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1028                        __u64 bitmap_offset, __u64 bitmap_len)
1029 {
1030         __u64 sector = req->offset >> 9;
1031         int i;
1032
1033         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1034                 panic("Operation too long");
1035
1036         if(req->op == UBD_READ) {
1037                 for(i = 0; i < req->length >> 9; i++){
1038                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1039                                 ubd_set_bit(i, (unsigned char *)
1040                                             &req->sector_mask);
1041                 }
1042         }
1043         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1044                            &req->cow_offset, bitmap, bitmap_offset,
1045                            req->bitmap_words, bitmap_len);
1046 }
1047
1048 /* Called with dev->lock held */
1049 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1050                             unsigned long long offset, int page_offset,
1051                             int len, struct page *page)
1052 {
1053         struct gendisk *disk = req->rq_disk;
1054         struct ubd *ubd_dev = disk->private_data;
1055
1056         io_req->req = req;
1057         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1058                 ubd_dev->fd;
1059         io_req->fds[1] = ubd_dev->fd;
1060         io_req->cow_offset = -1;
1061         io_req->offset = offset;
1062         io_req->length = len;
1063         io_req->error = 0;
1064         io_req->sector_mask = 0;
1065
1066         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1067         io_req->offsets[0] = 0;
1068         io_req->offsets[1] = ubd_dev->cow.data_offset;
1069         io_req->buffer = page_address(page) + page_offset;
1070         io_req->sectorsize = 1 << 9;
1071
1072         if(ubd_dev->cow.file != NULL)
1073                 cowify_req(io_req, ubd_dev->cow.bitmap,
1074                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1075
1076 }
1077
1078 /* Called with dev->lock held */
1079 static void do_ubd_request(request_queue_t *q)
1080 {
1081         struct io_thread_req io_req;
1082         struct request *req;
1083         int n;
1084
1085         while(1){
1086                 struct ubd *dev = q->queuedata;
1087                 if(dev->end_sg == 0){
1088                         struct request *req = elv_next_request(q);
1089                         if(req == NULL)
1090                                 return;
1091
1092                         dev->request = req;
1093                         blkdev_dequeue_request(req);
1094                         dev->start_sg = 0;
1095                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1096                 }
1097
1098                 req = dev->request;
1099                 while(dev->start_sg < dev->end_sg){
1100                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1101
1102                         prepare_request(req, &io_req,
1103                                         (unsigned long long) req->sector << 9,
1104                                         sg->offset, sg->length, sg->page);
1105
1106                         n = os_write_file_k(thread_fd, (char *) &io_req,
1107                                             sizeof(io_req));
1108                         if(n != sizeof(io_req)){
1109                                 if(n != -EAGAIN)
1110                                         printk("write to io thread failed, "
1111                                                "errno = %d\n", -n);
1112                                 else if(list_empty(&dev->restart))
1113                                         list_add(&dev->restart, &restart);
1114                                 return;
1115                         }
1116
1117                         req->sector += sg->length >> 9;
1118                         dev->start_sg++;
1119                 }
1120                 dev->end_sg = 0;
1121                 dev->request = NULL;
1122         }
1123 }
1124
1125 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1126 {
1127         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1128
1129         geo->heads = 128;
1130         geo->sectors = 32;
1131         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1132         return 0;
1133 }
1134
1135 static int ubd_ioctl(struct inode * inode, struct file * file,
1136                      unsigned int cmd, unsigned long arg)
1137 {
1138         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1139         struct hd_driveid ubd_id = {
1140                 .cyls           = 0,
1141                 .heads          = 128,
1142                 .sectors        = 32,
1143         };
1144
1145         switch (cmd) {
1146                 struct cdrom_volctrl volume;
1147         case HDIO_GET_IDENTITY:
1148                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1149                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1150                                  sizeof(ubd_id)))
1151                         return(-EFAULT);
1152                 return(0);
1153
1154         case CDROMVOLREAD:
1155                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1156                         return(-EFAULT);
1157                 volume.channel0 = 255;
1158                 volume.channel1 = 255;
1159                 volume.channel2 = 255;
1160                 volume.channel3 = 255;
1161                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1162                         return(-EFAULT);
1163                 return(0);
1164         }
1165         return(-EINVAL);
1166 }
1167
1168 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1169 {
1170         struct uml_stat buf1, buf2;
1171         int err;
1172
1173         if(from_cmdline == NULL)
1174                 return 0;
1175         if(!strcmp(from_cmdline, from_cow))
1176                 return 0;
1177
1178         err = os_stat_file(from_cmdline, &buf1);
1179         if(err < 0){
1180                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1181                 return 0;
1182         }
1183         err = os_stat_file(from_cow, &buf2);
1184         if(err < 0){
1185                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1186                 return 1;
1187         }
1188         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1189                 return 0;
1190
1191         printk("Backing file mismatch - \"%s\" requested,\n"
1192                "\"%s\" specified in COW header of \"%s\"\n",
1193                from_cmdline, from_cow, cow);
1194         return 1;
1195 }
1196
1197 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1198 {
1199         unsigned long modtime;
1200         unsigned long long actual;
1201         int err;
1202
1203         err = os_file_modtime(file, &modtime);
1204         if(err < 0){
1205                 printk("Failed to get modification time of backing file "
1206                        "\"%s\", err = %d\n", file, -err);
1207                 return(err);
1208         }
1209
1210         err = os_file_size(file, &actual);
1211         if(err < 0){
1212                 printk("Failed to get size of backing file \"%s\", "
1213                        "err = %d\n", file, -err);
1214                 return(err);
1215         }
1216
1217         if(actual != size){
1218                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1219                  * the typecast.*/
1220                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1221                        "file\n", (unsigned long long) size, actual);
1222                 return(-EINVAL);
1223         }
1224         if(modtime != mtime){
1225                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1226                        "file\n", mtime, modtime);
1227                 return(-EINVAL);
1228         }
1229         return(0);
1230 }
1231
1232 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1233 {
1234         int err;
1235
1236         err = os_seek_file(fd, offset);
1237         if(err < 0)
1238                 return(err);
1239
1240         err = os_read_file(fd, buf, len);
1241         if(err < 0)
1242                 return(err);
1243
1244         return(0);
1245 }
1246
1247 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1248                   char **backing_file_out, int *bitmap_offset_out,
1249                   unsigned long *bitmap_len_out, int *data_offset_out,
1250                   int *create_cow_out)
1251 {
1252         time_t mtime;
1253         unsigned long long size;
1254         __u32 version, align;
1255         char *backing_file;
1256         int fd, err, sectorsize, asked_switch, mode = 0644;
1257
1258         fd = os_open_file(file, *openflags, mode);
1259         if (fd < 0) {
1260                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1261                         *create_cow_out = 1;
1262                 if (!openflags->w ||
1263                    ((fd != -EROFS) && (fd != -EACCES)))
1264                         return fd;
1265                 openflags->w = 0;
1266                 fd = os_open_file(file, *openflags, mode);
1267                 if (fd < 0)
1268                         return fd;
1269         }
1270
1271         if(shared)
1272                 printk("Not locking \"%s\" on the host\n", file);
1273         else {
1274                 err = os_lock_file(fd, openflags->w);
1275                 if(err < 0){
1276                         printk("Failed to lock '%s', err = %d\n", file, -err);
1277                         goto out_close;
1278                 }
1279         }
1280
1281         /* Successful return case! */
1282         if(backing_file_out == NULL)
1283                 return(fd);
1284
1285         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1286                               &size, &sectorsize, &align, bitmap_offset_out);
1287         if(err && (*backing_file_out != NULL)){
1288                 printk("Failed to read COW header from COW file \"%s\", "
1289                        "errno = %d\n", file, -err);
1290                 goto out_close;
1291         }
1292         if(err)
1293                 return(fd);
1294
1295         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1296
1297         /* Allow switching only if no mismatch. */
1298         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1299                 printk("Switching backing file to '%s'\n", *backing_file_out);
1300                 err = write_cow_header(file, fd, *backing_file_out,
1301                                        sectorsize, align, &size);
1302                 if (err) {
1303                         printk("Switch failed, errno = %d\n", -err);
1304                         goto out_close;
1305                 }
1306         } else {
1307                 *backing_file_out = backing_file;
1308                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1309                 if (err)
1310                         goto out_close;
1311         }
1312
1313         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1314                   bitmap_len_out, data_offset_out);
1315
1316         return fd;
1317  out_close:
1318         os_close_file(fd);
1319         return err;
1320 }
1321
1322 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1323                     int sectorsize, int alignment, int *bitmap_offset_out,
1324                     unsigned long *bitmap_len_out, int *data_offset_out)
1325 {
1326         int err, fd;
1327
1328         flags.c = 1;
1329         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1330         if(fd < 0){
1331                 err = fd;
1332                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1333                        -err);
1334                 goto out;
1335         }
1336
1337         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1338                             bitmap_offset_out, bitmap_len_out,
1339                             data_offset_out);
1340         if(!err)
1341                 return(fd);
1342         os_close_file(fd);
1343  out:
1344         return(err);
1345 }
1346
1347 static int update_bitmap(struct io_thread_req *req)
1348 {
1349         int n;
1350
1351         if(req->cow_offset == -1)
1352                 return(0);
1353
1354         n = os_seek_file(req->fds[1], req->cow_offset);
1355         if(n < 0){
1356                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1357                 return(1);
1358         }
1359
1360         n = os_write_file_k(req->fds[1], &req->bitmap_words,
1361                             sizeof(req->bitmap_words));
1362         if(n != sizeof(req->bitmap_words)){
1363                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1364                        req->fds[1]);
1365                 return(1);
1366         }
1367
1368         return(0);
1369 }
1370
1371 void do_io(struct io_thread_req *req)
1372 {
1373         char *buf;
1374         unsigned long len;
1375         int n, nsectors, start, end, bit;
1376         int err;
1377         __u64 off;
1378
1379         nsectors = req->length / req->sectorsize;
1380         start = 0;
1381         do {
1382                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1383                 end = start;
1384                 while((end < nsectors) &&
1385                       (ubd_test_bit(end, (unsigned char *)
1386                                     &req->sector_mask) == bit))
1387                         end++;
1388
1389                 off = req->offset + req->offsets[bit] +
1390                         start * req->sectorsize;
1391                 len = (end - start) * req->sectorsize;
1392                 buf = &req->buffer[start * req->sectorsize];
1393
1394                 err = os_seek_file(req->fds[bit], off);
1395                 if(err < 0){
1396                         printk("do_io - lseek failed : err = %d\n", -err);
1397                         req->error = 1;
1398                         return;
1399                 }
1400                 if(req->op == UBD_READ){
1401                         n = 0;
1402                         do {
1403                                 buf = &buf[n];
1404                                 len -= n;
1405                                 n = os_read_file_k(req->fds[bit], buf, len);
1406                                 if (n < 0) {
1407                                         printk("do_io - read failed, err = %d "
1408                                                "fd = %d\n", -n, req->fds[bit]);
1409                                         req->error = 1;
1410                                         return;
1411                                 }
1412                         } while((n < len) && (n != 0));
1413                         if (n < len) memset(&buf[n], 0, len - n);
1414                 } else {
1415                         n = os_write_file_k(req->fds[bit], buf, len);
1416                         if(n != len){
1417                                 printk("do_io - write failed err = %d "
1418                                        "fd = %d\n", -n, req->fds[bit]);
1419                                 req->error = 1;
1420                                 return;
1421                         }
1422                 }
1423
1424                 start = end;
1425         } while(start < nsectors);
1426
1427         req->error = update_bitmap(req);
1428 }
1429
1430 /* Changed in start_io_thread, which is serialized by being called only
1431  * from ubd_init, which is an initcall.
1432  */
1433 int kernel_fd = -1;
1434
1435 /* Only changed by the io thread. XXX: currently unused. */
1436 static int io_count = 0;
1437
1438 int io_thread(void *arg)
1439 {
1440         struct io_thread_req req;
1441         int n;
1442
1443         ignore_sigwinch_sig();
1444         while(1){
1445                 n = os_read_file_k(kernel_fd, &req, sizeof(req));
1446                 if(n != sizeof(req)){
1447                         if(n < 0)
1448                                 printk("io_thread - read failed, fd = %d, "
1449                                        "err = %d\n", kernel_fd, -n);
1450                         else {
1451                                 printk("io_thread - short read, fd = %d, "
1452                                        "length = %d\n", kernel_fd, n);
1453                         }
1454                         continue;
1455                 }
1456                 io_count++;
1457                 do_io(&req);
1458                 n = os_write_file_k(kernel_fd, &req, sizeof(req));
1459                 if(n != sizeof(req))
1460                         printk("io_thread - write failed, fd = %d, err = %d\n",
1461                                kernel_fd, -n);
1462         }
1463
1464         return 0;
1465 }