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  */
  12
  13 #define MAJOR_NR UBD_MAJOR
  14 #define UBD_SHIFT 4
  15
  16 #include "linux/config.h"
  17 #include "linux/module.h"
  18 #include "linux/blkdev.h"
  19 #include "linux/hdreg.h"
  20 #include "linux/init.h"
  21 #include "linux/devfs_fs_kernel.h"
  22 #include "linux/cdrom.h"
  23 #include "linux/proc_fs.h"
  24 #include "linux/ctype.h"
  25 #include "linux/capability.h"
  26 #include "linux/mm.h"
  27 #include "linux/vmalloc.h"
  28 #include "linux/blkpg.h"
  29 #include "linux/genhd.h"
  30 #include "linux/spinlock.h"
  31 #include "asm/segment.h"
  32 #include "asm/uaccess.h"
  33 #include "asm/irq.h"
  34 #include "asm/types.h"
  35 #include "asm/tlbflush.h"
  36 #include "user_util.h"
  37 #include "mem_user.h"
  38 #include "kern_util.h"
  39 #include "kern.h"
  40 #include "mconsole_kern.h"
  41 #include "init.h"
  42 #include "irq_user.h"
  43 #include "ubd_user.h"
  44 #include "2_5compat.h"
  45 #include "os.h"
  46
  47 static spinlock_t ubd_io_lock = SPIN_LOCK_UNLOCKED;
  48 static spinlock_t ubd_lock = SPIN_LOCK_UNLOCKED;
  49
  50 static void (*do_ubd)(void);
  51
  52 static int ubd_open(struct inode * inode, struct file * filp);
  53 static int ubd_release(struct inode * inode, struct file * file);
  54 static int ubd_ioctl(struct inode * inode, struct file * file,
  55                      unsigned int cmd, unsigned long arg);
  56
  57 #define MAX_DEV (8)
  58
  59 static struct block_device_operations ubd_blops = {
  60         .owner          = THIS_MODULE,
  61         .open           = ubd_open,
  62         .release        = ubd_release,
  63         .ioctl          = ubd_ioctl,
  64 };
  65
  66 /* Protected by the queue_lock */
  67 static request_queue_t *ubd_queue;
  68
  69 /* Protected by ubd_lock */
  70 static int fake_major = 0;
  71
  72 static struct gendisk *ubd_gendisk[MAX_DEV];
  73 static struct gendisk *fake_gendisk[MAX_DEV];
  74
  75 #ifdef CONFIG_BLK_DEV_UBD_SYNC
  76 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
  77                                          .cl = 1 })
  78 #else
  79 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
  80                                          .cl = 1 })
  81 #endif
  82
  83 /* Not protected - changed only in ubd_setup_common and then only to
  84  * to enable O_SYNC.
  85  */
  86 static struct openflags global_openflags = OPEN_FLAGS;
  87
  88 struct cow {
  89         char *file;
  90         int fd;
  91         unsigned long *bitmap;
  92         unsigned long bitmap_len;
  93         int bitmap_offset;
  94         int data_offset;
  95 };
  96
  97 struct ubd {
  98         char *file;
  99         int is_dir;
 100         int count;
 101         int fd;
 102         __u64 size;
 103         struct openflags boot_openflags;
 104         struct openflags openflags;
 105         struct cow cow;
 106 };
 107
 108 #define DEFAULT_COW { \
 109         .file =                 NULL, \
 110         .fd =                   -1, \
 111         .bitmap =               NULL, \
 112         .bitmap_offset =        0, \
 113         .data_offset =          0, \
 114 }
 115
 116 #define DEFAULT_UBD { \
 117         .file =                 NULL, \
 118         .is_dir =               0, \
 119         .count =                0, \
 120         .fd =                   -1, \
 121         .size =                 -1, \
 122         .boot_openflags =       OPEN_FLAGS, \
 123         .openflags =            OPEN_FLAGS, \
 124         .cow =                  DEFAULT_COW, \
 125 }
 126
 127 struct ubd ubd_dev[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
 128
 129 static int ubd0_init(void)
 130 {
 131         if(ubd_dev[0].file == NULL)
 132                 ubd_dev[0].file = "root_fs";
 133         return(0);
 134 }
 135
 136 __initcall(ubd0_init);
 137
 138 /* Only changed by fake_ide_setup which is a setup */
 139 static int fake_ide = 0;
 140 static struct proc_dir_entry *proc_ide_root = NULL;
 141 static struct proc_dir_entry *proc_ide = NULL;
 142
 143 static void make_proc_ide(void)
 144 {
 145         proc_ide_root = proc_mkdir("ide", 0);
 146         proc_ide = proc_mkdir("ide0", proc_ide_root);
 147 }
 148
 149 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
 150                                int *eof, void *data)
 151 {
 152         int len;
 153
 154         strcpy(page, "disk\n");
 155         len = strlen("disk\n");
 156         len -= off;
 157         if (len < count){
 158                 *eof = 1;
 159                 if (len <= 0) return 0;
 160         }
 161         else len = count;
 162         *start = page + off;
 163         return len;
 164 }
 165
 166 static void make_ide_entries(char *dev_name)
 167 {
 168         struct proc_dir_entry *dir, *ent;
 169         char name[64];
 170
 171         if(proc_ide_root == NULL) make_proc_ide();
 172
 173         dir = proc_mkdir(dev_name, proc_ide);
 174         if(!dir) return;
 175
 176         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
 177         if(!ent) return;
 178         ent->nlink = 1;
 179         ent->data = NULL;
 180         ent->read_proc = proc_ide_read_media;
 181         ent->write_proc = NULL;
 182         sprintf(name,"ide0/%s", dev_name);
 183         proc_symlink(dev_name, proc_ide_root, name);
 184 }
 185
 186 static int fake_ide_setup(char *str)
 187 {
 188         fake_ide = 1;
 189         return(1);
 190 }
 191
 192 __setup("fake_ide", fake_ide_setup);
 193
 194 __uml_help(fake_ide_setup,
 195 "fake_ide\n"
 196 "    Create ide0 entries that map onto ubd devices.\n\n"
 197 );
 198
 199 static int ubd_setup_common(char *str, int *index_out)
 200 {
 201         struct openflags flags = global_openflags;
 202         char *backing_file;
 203         int n, err;
 204
 205         if(index_out) *index_out = -1;
 206         n = *str++;
 207         if(n == '='){
 208                 static int fake_major_allowed = 1;
 209                 char *end;
 210                 int major;
 211
 212                 if(!strcmp(str, "sync")){
 213                         global_openflags.s = 1;
 214                         return(0);
 215                 }
 216                 major = simple_strtoul(str, &end, 0);
 217                 if((*end != '\0') || (end == str)){
 218                         printk(KERN_ERR
 219                                "ubd_setup : didn't parse major number\n");
 220                         return(1);
 221                 }
 222
 223                 if(!fake_major_allowed){
 224                         printk(KERN_ERR "Can't assign a fake major twice\n");
 225                         return(1);
 226                 }
 227
 228                 err = 1;
 229                 spin_lock(&ubd_lock);
 230                 if(!fake_major_allowed){
 231                         printk(KERN_ERR "Can't assign a fake major twice\n");
 232                         goto out1;
 233                 }
 234
 235                 fake_major = major;
 236                 fake_major_allowed = 0;
 237
 238                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 239                        major);
 240                 err = 0;
 241         out1:
 242                 spin_unlock(&ubd_lock);
 243                 return(err);
 244         }
 245
 246         if(n < '0'){
 247                 printk(KERN_ERR "ubd_setup : index out of range\n"); }
 248
 249         if((n >= '0') && (n <= '9')) n -= '0';
 250         else if((n >= 'a') && (n <= 'z')) n -= 'a';
 251         else {
 252                 printk(KERN_ERR "ubd_setup : device syntax invalid\n");
 253                 return(1);
 254         }
 255         if(n >= MAX_DEV){
 256                 printk(KERN_ERR "ubd_setup : index out of range "
 257                        "(%d devices)\n", MAX_DEV);
 258                 return(1);
 259         }
 260
 261         err = 1;
 262         spin_lock(&ubd_lock);
 263
 264         if(ubd_dev[n].file != NULL){
 265                 printk(KERN_ERR "ubd_setup : device already configured\n");
 266                 goto out2;
 267         }
 268
 269         if(index_out) *index_out = n;
 270
 271         if (*str == 'r'){
 272                 flags.w = 0;
 273                 str++;
 274         }
 275         if (*str == 's'){
 276                 flags.s = 1;
 277                 str++;
 278         }
 279         if(*str++ != '='){
 280                 printk(KERN_ERR "ubd_setup : Expected '='\n");
 281                 goto out2;
 282         }
 283
 284         err = 0;
 285         backing_file = strchr(str, ',');
 286         if(backing_file){
 287                 *backing_file = '\0';
 288                 backing_file++;
 289         }
 290         ubd_dev[n].file = str;
 291         if(ubd_is_dir(ubd_dev[n].file))
 292                 ubd_dev[n].is_dir = 1;
 293         ubd_dev[n].cow.file = backing_file;
 294         ubd_dev[n].boot_openflags = flags;
 295  out2:
 296         spin_unlock(&ubd_lock);
 297         return(err);
 298 }
 299
 300 static int ubd_setup(char *str)
 301 {
 302         ubd_setup_common(str, NULL);
 303         return(1);
 304 }
 305
 306 __setup("ubd", ubd_setup);
 307 __uml_help(ubd_setup,
 308 "ubd<n>=<filename>\n"
 309 "    This is used to associate a device with a file in the underlying\n"
 310 "    filesystem. Usually, there is a filesystem in the file, but \n"
 311 "    that's not required. Swap devices containing swap files can be\n"
 312 "    specified like this. Also, a file which doesn't contain a\n"
 313 "    filesystem can have its contents read in the virtual \n"
 314 "    machine by running dd on the device. n must be in the range\n"
 315 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 316 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 317 "    an 's' (has to be _after_ 'r', if there is one) will cause data\n"
 318 "    to be written to disk on the host immediately.\n\n"
 319 );
 320
 321 static int fakehd_set = 0;
 322 static int fakehd(char *str)
 323 {
 324         printk(KERN_INFO
 325                "fakehd : Changing ubd name to \"hd\".\n");
 326         fakehd_set = 1;
 327         return 1;
 328 }
 329
 330 __setup("fakehd", fakehd);
 331 __uml_help(fakehd,
 332 "fakehd\n"
 333 "    Change the ubd device name to \"hd\".\n\n"
 334 );
 335
 336 static void do_ubd_request(request_queue_t * q);
 337
 338 /* Only changed by ubd_init, which is an initcall. */
 339 int thread_fd = -1;
 340
 341 /* Changed by ubd_handler, which is serialized because interrupts only
 342  * happen on CPU 0.
 343  */
 344 int intr_count = 0;
 345
 346 static void ubd_finish(struct request *req, int error)
 347 {
 348         int nsect;
 349
 350         if(error){
 351                 spin_lock(&ubd_io_lock);
 352                 end_request(req, 0);
 353                 spin_unlock(&ubd_io_lock);
 354                 return;
 355         }
 356         nsect = req->current_nr_sectors;
 357         req->sector += nsect;
 358         req->buffer += nsect << 9;
 359         req->errors = 0;
 360         req->nr_sectors -= nsect;
 361         req->current_nr_sectors = 0;
 362         spin_lock(&ubd_io_lock);
 363         end_request(req, 1);
 364         spin_unlock(&ubd_io_lock);
 365 }
 366
 367 static void ubd_handler(void)
 368 {
 369         struct io_thread_req req;
 370         struct request *rq = elv_next_request(ubd_queue);
 371         int n;
 372
 373         do_ubd = NULL;
 374         intr_count++;
 375         n = read_ubd_fs(thread_fd, &req, sizeof(req));
 376         if(n != sizeof(req)){
 377                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
 378                        "errno = %d\n", os_getpid(), -n);
 379                 spin_lock(&ubd_io_lock);
 380                 end_request(rq, 0);
 381                 spin_unlock(&ubd_io_lock);
 382                 return;
 383         }
 384
 385         if((req.offset != ((__u64) (rq->sector)) << 9) ||
 386            (req.length != (rq->current_nr_sectors) << 9))
 387                 panic("I/O op mismatch");
 388
 389         ubd_finish(rq, req.error);
 390         reactivate_fd(thread_fd, UBD_IRQ);
 391         do_ubd_request(ubd_queue);
 392 }
 393
 394 static void ubd_intr(int irq, void *dev, struct pt_regs *unused)
 395 {
 396         ubd_handler();
 397 }
 398
 399 /* Only changed by ubd_init, which is an initcall. */
 400 static int io_pid = -1;
 401
 402 void kill_io_thread(void)
 403 {
 404         if(io_pid != -1)
 405                 os_kill_process(io_pid, 1);
 406 }
 407
 408 __uml_exitcall(kill_io_thread);
 409
 410 static int ubd_file_size(struct ubd *dev, __u64 *size_out)
 411 {
 412         char *file;
 413
 414         file = dev->cow.file ? dev->cow.file : dev->file;
 415         return(os_file_size(file, size_out));
 416 }
 417
 418 static void ubd_close(struct ubd *dev)
 419 {
 420         os_close_file(dev->fd);
 421         if(dev->cow.file == NULL)
 422                 return;
 423
 424         os_close_file(dev->cow.fd);
 425         vfree(dev->cow.bitmap);
 426         dev->cow.bitmap = NULL;
 427 }
 428
 429 static int ubd_open_dev(struct ubd *dev)
 430 {
 431         struct openflags flags;
 432         int err, n, create_cow, *create_ptr;
 433
 434         create_cow = 0;
 435         create_ptr = (dev->cow.file != NULL) ? &create_cow : NULL;
 436         dev->fd = open_ubd_file(dev->file, &dev->openflags, &dev->cow.file,
 437                                 &dev->cow.bitmap_offset, &dev->cow.bitmap_len,
 438                                 &dev->cow.data_offset, create_ptr);
 439
 440         if((dev->fd == -ENOENT) && create_cow){
 441                 n = dev - ubd_dev;
 442                 dev->fd = create_cow_file(dev->file, dev->cow.file,
 443                                           dev->openflags, 1 << 9,
 444                                           &dev->cow.bitmap_offset,
 445                                           &dev->cow.bitmap_len,
 446                                           &dev->cow.data_offset);
 447                 if(dev->fd >= 0){
 448                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 449                                "\"%s\"\n", dev->file, dev->cow.file);
 450                 }
 451         }
 452
 453         if(dev->fd < 0) return(dev->fd);
 454
 455         if(dev->cow.file != NULL){
 456                 err = -ENOMEM;
 457                 dev->cow.bitmap = (void *) vmalloc(dev->cow.bitmap_len);
 458                 if(dev->cow.bitmap == NULL) goto error;
 459                 flush_tlb_kernel_vm();
 460
 461                 err = read_cow_bitmap(dev->fd, dev->cow.bitmap,
 462                                       dev->cow.bitmap_offset,
 463                                       dev->cow.bitmap_len);
 464                 if(err) goto error;
 465
 466                 flags = dev->openflags;
 467                 flags.w = 0;
 468                 err = open_ubd_file(dev->cow.file, &flags, NULL, NULL, NULL,
 469                                     NULL, NULL);
 470                 if(err < 0) goto error;
 471                 dev->cow.fd = err;
 472         }
 473         return(0);
 474  error:
 475         os_close_file(dev->fd);
 476         return(err);
 477 }
 478
 479 static int ubd_new_disk(int major, u64 size, int unit,
 480                         struct gendisk **disk_out)
 481
 482 {
 483         struct gendisk *disk;
 484
 485         disk = alloc_disk(1 << UBD_SHIFT);
 486         if (!disk)
 487                 return -ENOMEM;
 488
 489         disk->major = major;
 490         disk->first_minor = unit << UBD_SHIFT;
 491         disk->fops = &ubd_blops;
 492         set_capacity(disk, size / 512);
 493         sprintf(disk->disk_name, "ubd");
 494         sprintf(disk->devfs_name, "ubd/disc%d", unit);
 495
 496         disk->private_data = &ubd_dev[unit];
 497         disk->queue = ubd_queue;
 498         add_disk(disk);
 499
 500         *disk_out = disk;
 501         return 0;
 502 }
 503
 504 static int ubd_add(int n)
 505 {
 506         struct ubd *dev = &ubd_dev[n];
 507         int err;
 508
 509         if(dev->is_dir)
 510                 return(-EISDIR);
 511
 512         if (!dev->file)
 513                 return(-ENODEV);
 514
 515         if (ubd_open_dev(dev))
 516                 return(-ENODEV);
 517
 518         err = ubd_file_size(dev, &dev->size);
 519         if(err)
 520                 return(err);
 521
 522         err = ubd_new_disk(MAJOR_NR, dev->size, n, &ubd_gendisk[n]);
 523         if(err)
 524                 return(err);
 525
 526         if(fake_major)
 527                 ubd_new_disk(fake_major, dev->size, n,
 528                              &fake_gendisk[n]);
 529
 530         /* perhaps this should also be under the "if (fake_major)" above */
 531         /* using the fake_disk->disk_name and also the fakehd_set name */
 532         if (fake_ide)
 533                 make_ide_entries(ubd_gendisk[n]->disk_name);
 534
 535         ubd_close(dev);
 536         return 0;
 537 }
 538
 539 static int ubd_config(char *str)
 540 {
 541         int n, err;
 542
 543         str = uml_strdup(str);
 544         if(str == NULL){
 545                 printk(KERN_ERR "ubd_config failed to strdup string\n");
 546                 return(1);
 547         }
 548         err = ubd_setup_common(str, &n);
 549         if(err){
 550                 kfree(str);
 551                 return(-1);
 552         }
 553         if(n == -1) return(0);
 554
 555         spin_lock(&ubd_lock);
 556         err = ubd_add(n);
 557         if(err)
 558                 ubd_dev[n].file = NULL;
 559         spin_unlock(&ubd_lock);
 560
 561         return(err);
 562 }
 563
 564 static int ubd_get_config(char *dev, char *str, int size, char **error_out)
 565 {
 566         struct ubd *ubd;
 567         char *end;
 568         int major, n = 0;
 569
 570         major = simple_strtoul(dev, &end, 0);
 571         if((*end != '\0') || (end == dev)){
 572                 *error_out = "ubd_get_config : didn't parse major number";
 573                 return(-1);
 574         }
 575
 576         if((major >= MAX_DEV) || (major < 0)){
 577                 *error_out = "ubd_get_config : major number out of range";
 578                 return(-1);
 579         }
 580
 581         ubd = &ubd_dev[major];
 582         spin_lock(&ubd_lock);
 583
 584         if(ubd->file == NULL){
 585                 CONFIG_CHUNK(str, size, n, "", 1);
 586                 goto out;
 587         }
 588
 589         CONFIG_CHUNK(str, size, n, ubd->file, 0);
 590
 591         if(ubd->cow.file != NULL){
 592                 CONFIG_CHUNK(str, size, n, ",", 0);
 593                 CONFIG_CHUNK(str, size, n, ubd->cow.file, 1);
 594         }
 595         else CONFIG_CHUNK(str, size, n, "", 1);
 596
 597  out:
 598         spin_unlock(&ubd_lock);
 599         return(n);
 600 }
 601
 602 static int ubd_remove(char *str)
 603 {
 604         struct ubd *dev;
 605         int n, err = -ENODEV;
 606
 607         if(!isdigit(*str))
 608                 return(err);    /* it should be a number 0-7/a-h */
 609
 610         n = *str - '0';
 611         if(n >= MAX_DEV)
 612                 return(err);
 613
 614         dev = &ubd_dev[n];
 615         if(dev->count > 0)
 616                 return(-EBUSY); /* you cannot remove a open disk */
 617
 618         err = 0;
 619         spin_lock(&ubd_lock);
 620
 621         if(ubd_gendisk[n] == NULL)
 622                 goto out;
 623
 624         del_gendisk(ubd_gendisk[n]);
 625         put_disk(ubd_gendisk[n]);
 626         ubd_gendisk[n] = NULL;
 627
 628         if(fake_gendisk[n] != NULL){
 629                 del_gendisk(fake_gendisk[n]);
 630                 put_disk(fake_gendisk[n]);
 631                 fake_gendisk[n] = NULL;
 632         }
 633
 634         *dev = ((struct ubd) DEFAULT_UBD);
 635         err = 0;
 636  out:
 637         spin_unlock(&ubd_lock);
 638         return(err);
 639 }
 640
 641 static struct mc_device ubd_mc = {
 642         .name           = "ubd",
 643         .config         = ubd_config,
 644         .get_config     = ubd_get_config,
 645         .remove         = ubd_remove,
 646 };
 647
 648 static int ubd_mc_init(void)
 649 {
 650         mconsole_register_dev(&ubd_mc);
 651         return 0;
 652 }
 653
 654 __initcall(ubd_mc_init);
 655
 656 int ubd_init(void)
 657 {
 658         int i;
 659
 660         devfs_mk_dir("ubd");
 661         if (register_blkdev(MAJOR_NR, "ubd"))
 662                 return -1;
 663
 664         ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
 665         if (!ubd_queue) {
 666                 unregister_blkdev(MAJOR_NR, "ubd");
 667                 return -1;
 668         }
 669
 670         elevator_init(ubd_queue, &elevator_noop);
 671
 672         if (fake_major != 0) {
 673                 char name[sizeof("ubd_nnn\0")];
 674
 675                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
 676                 devfs_mk_dir(name);
 677                 if (register_blkdev(fake_major, "ubd"))
 678                         return -1;
 679         }
 680         for (i = 0; i < MAX_DEV; i++)
 681                 ubd_add(i);
 682         return 0;
 683 }
 684
 685 late_initcall(ubd_init);
 686
 687 int ubd_driver_init(void){
 688         unsigned long stack;
 689         int err;
 690
 691         if(global_openflags.s){
 692                 printk(KERN_INFO "ubd : Synchronous mode\n");
 693                 return(0);
 694         }
 695         stack = alloc_stack(0, 0);
 696         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
 697                                  &thread_fd);
 698         if(io_pid < 0){
 699                 printk(KERN_ERR
 700                        "ubd : Failed to start I/O thread (errno = %d) - "
 701                        "falling back to synchronous I/O\n", -io_pid);
 702                 return(0);
 703         }
 704         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
 705                              SA_INTERRUPT, "ubd", ubd_dev);
 706         if(err != 0) printk(KERN_ERR
 707                             "um_request_irq failed - errno = %d\n", -err);
 708         return(err);
 709 }
 710
 711 device_initcall(ubd_driver_init);
 712
 713 static int ubd_open(struct inode *inode, struct file *filp)
 714 {
 715         struct gendisk *disk = inode->i_bdev->bd_disk;
 716         struct ubd *dev = disk->private_data;
 717         int err = -EISDIR;
 718
 719         if(dev->is_dir == 1)
 720                 goto out;
 721
 722         err = 0;
 723         if(dev->count == 0){
 724                 dev->openflags = dev->boot_openflags;
 725
 726                 err = ubd_open_dev(dev);
 727                 if(err){
 728                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
 729                                disk->disk_name, dev->file, -err);
 730                         goto out;
 731                 }
 732         }
 733         dev->count++;
 734         if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
 735                 if(--dev->count == 0) ubd_close(dev);
 736                 err = -EROFS;
 737         }
 738  out:
 739         return(err);
 740 }
 741
 742 static int ubd_release(struct inode * inode, struct file * file)
 743 {
 744         struct gendisk *disk = inode->i_bdev->bd_disk;
 745         struct ubd *dev = disk->private_data;
 746
 747         if(--dev->count == 0)
 748                 ubd_close(dev);
 749         return(0);
 750 }
 751
 752 void cowify_req(struct io_thread_req *req, struct ubd *dev)
 753 {
 754         int i, update_bitmap, sector = req->offset >> 9;
 755
 756         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
 757                 panic("Operation too long");
 758         if(req->op == UBD_READ) {
 759                 for(i = 0; i < req->length >> 9; i++){
 760                         if(ubd_test_bit(sector + i, (unsigned char *)
 761                                         dev->cow.bitmap)){
 762                                 ubd_set_bit(i, (unsigned char *)
 763                                             &req->sector_mask);
 764                         }
 765                 }
 766         }
 767         else {
 768                 update_bitmap = 0;
 769                 for(i = 0; i < req->length >> 9; i++){
 770                         ubd_set_bit(i, (unsigned char *)
 771                                     &req->sector_mask);
 772                         if(!ubd_test_bit(sector + i, (unsigned char *)
 773                                          dev->cow.bitmap))
 774                                 update_bitmap = 1;
 775                         ubd_set_bit(sector + i, (unsigned char *)
 776                                     dev->cow.bitmap);
 777                 }
 778                 if(update_bitmap){
 779                         req->cow_offset = sector / (sizeof(unsigned long) * 8);
 780                         req->bitmap_words[0] =
 781                                 dev->cow.bitmap[req->cow_offset];
 782                         req->bitmap_words[1] =
 783                                 dev->cow.bitmap[req->cow_offset + 1];
 784                         req->cow_offset *= sizeof(unsigned long);
 785                         req->cow_offset += dev->cow.bitmap_offset;
 786                 }
 787         }
 788 }
 789
 790 static int prepare_request(struct request *req, struct io_thread_req *io_req)
 791 {
 792         struct gendisk *disk = req->rq_disk;
 793         struct ubd *dev = disk->private_data;
 794         __u64 block;
 795         int nsect;
 796
 797         if(req->rq_status == RQ_INACTIVE) return(1);
 798
 799         if(dev->is_dir){
 800                 strcpy(req->buffer, "HOSTFS:");
 801                 strcat(req->buffer, dev->file);
 802                 spin_lock(&ubd_io_lock);
 803                 end_request(req, 1);
 804                 spin_unlock(&ubd_io_lock);
 805                 return(1);
 806         }
 807
 808         if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
 809                 printk("Write attempted on readonly ubd device %s\n",
 810                        disk->disk_name);
 811                 spin_lock(&ubd_io_lock);
 812                 end_request(req, 0);
 813                 spin_unlock(&ubd_io_lock);
 814                 return(1);
 815         }
 816
 817         block = req->sector;
 818         nsect = req->current_nr_sectors;
 819
 820         io_req->op = rq_data_dir(req) == READ ? UBD_READ : UBD_WRITE;
 821         io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
 822         io_req->fds[1] = dev->fd;
 823         io_req->offsets[0] = 0;
 824         io_req->offsets[1] = dev->cow.data_offset;
 825         io_req->offset = ((__u64) block) << 9;
 826         io_req->length = nsect << 9;
 827         io_req->buffer = req->buffer;
 828         io_req->sectorsize = 1 << 9;
 829         io_req->sector_mask = 0;
 830         io_req->cow_offset = -1;
 831         io_req->error = 0;
 832
 833         if(dev->cow.file != NULL) cowify_req(io_req, dev);
 834         return(0);
 835 }
 836
 837 static void do_ubd_request(request_queue_t *q)
 838 {
 839         struct io_thread_req io_req;
 840         struct request *req;
 841         int err, n;
 842
 843         if(thread_fd == -1){
 844                 while(!list_empty(&q->queue_head)){
 845                         req = elv_next_request(q);
 846                         err = prepare_request(req, &io_req);
 847                         if(!err){
 848                                 do_io(&io_req);
 849                                 ubd_finish(req, io_req.error);
 850                         }
 851                 }
 852         }
 853         else {
 854                 if(do_ubd || list_empty(&q->queue_head)) return;
 855                 req = elv_next_request(q);
 856                 err = prepare_request(req, &io_req);
 857                 if(!err){
 858                         do_ubd = ubd_handler;
 859                         n = write_ubd_fs(thread_fd, (char *) &io_req,
 860                                          sizeof(io_req));
 861                         if(n != sizeof(io_req))
 862                                 printk("write to io thread failed, "
 863                                        "errno = %d\n", -n);
 864                 }
 865         }
 866 }
 867
 868 static int ubd_ioctl(struct inode * inode, struct file * file,
 869                      unsigned int cmd, unsigned long arg)
 870 {
 871         struct hd_geometry *loc = (struct hd_geometry *) arg;
 872         struct ubd *dev = inode->i_bdev->bd_disk->private_data;
 873         int err;
 874         struct hd_driveid ubd_id = {
 875                 .cyls           = 0,
 876                 .heads          = 128,
 877                 .sectors        = 32,
 878         };
 879
 880         switch (cmd) {
 881                 struct hd_geometry g;
 882                 struct cdrom_volctrl volume;
 883         case HDIO_GETGEO:
 884                 if(!loc) return(-EINVAL);
 885                 g.heads = 128;
 886                 g.sectors = 32;
 887                 g.cylinders = dev->size / (128 * 32 * 512);
 888                 g.start = 2;
 889                 return(copy_to_user(loc, &g, sizeof(g)) ? -EFAULT : 0);
 890
 891         case HDIO_SET_UNMASKINTR:
 892                 if(!capable(CAP_SYS_ADMIN)) return(-EACCES);
 893                 if((arg > 1) || (inode->i_bdev->bd_contains != inode->i_bdev))
 894                         return(-EINVAL);
 895                 return(0);
 896
 897         case HDIO_GET_UNMASKINTR:
 898                 if(!arg)  return(-EINVAL);
 899                 err = verify_area(VERIFY_WRITE, (long *) arg, sizeof(long));
 900                 if(err)
 901                         return(err);
 902                 return(0);
 903
 904         case HDIO_GET_MULTCOUNT:
 905                 if(!arg)  return(-EINVAL);
 906                 err = verify_area(VERIFY_WRITE, (long *) arg, sizeof(long));
 907                 if(err)
 908                         return(err);
 909                 return(0);
 910
 911         case HDIO_SET_MULTCOUNT:
 912                 if(!capable(CAP_SYS_ADMIN)) return(-EACCES);
 913                 if(inode->i_bdev->bd_contains != inode->i_bdev)
 914                         return(-EINVAL);
 915                 return(0);
 916
 917         case HDIO_GET_IDENTITY:
 918                 ubd_id.cyls = dev->size / (128 * 32 * 512);
 919                 if(copy_to_user((char *) arg, (char *) &ubd_id,
 920                                  sizeof(ubd_id)))
 921                         return(-EFAULT);
 922                 return(0);
 923
 924         case CDROMVOLREAD:
 925                 if(copy_from_user(&volume, (char *) arg, sizeof(volume)))
 926                         return(-EFAULT);
 927                 volume.channel0 = 255;
 928                 volume.channel1 = 255;
 929                 volume.channel2 = 255;
 930                 volume.channel3 = 255;
 931                 if(copy_to_user((char *) arg, &volume, sizeof(volume)))
 932                         return(-EFAULT);
 933                 return(0);
 934         }
 935         return(-EINVAL);
 936 }
 937
 938 /*
 939  * Overrides for Emacs so that we follow Linus's tabbing style.
 940  * Emacs will notice this stuff at the end of the file and automatically
 941  * adjust the settings for this buffer only.  This must remain at the end
 942  * of the file.
 943  * ---------------------------------------------------------------------------
 944  * Local variables:
 945  * c-file-style: "linux"
 946  * End:
 947  */