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

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