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