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