arch/um/drivers/ubd_kern.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2018 Cambridge Greys Ltd
   4  * Copyright (C) 2015-2016 Anton Ivanov (aivanov@brocade.com)
   5  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
   6  */
   7
   8 /* 2001-09-28...2002-04-17
   9  * Partition stuff by James_McMechan@hotmail.com
  10  * old style ubd by setting UBD_SHIFT to 0
  11  * 2002-09-27...2002-10-18 massive tinkering for 2.5
  12  * partitions have changed in 2.5
  13  * 2003-01-29 more tinkering for 2.5.59-1
  14  * This should now address the sysfs problems and has
  15  * the symlink for devfs to allow for booting with
  16  * the common /dev/ubd/discX/... names rather than
  17  * only /dev/ubdN/discN this version also has lots of
  18  * clean ups preparing for ubd-many.
  19  * James McMechan
  20  */
  21
  22 #define UBD_SHIFT 4
  23
  24 #include <linux/module.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/blk-mq.h>
  28 #include <linux/ata.h>
  29 #include <linux/hdreg.h>
  30 #include <linux/cdrom.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/seq_file.h>
  33 #include <linux/ctype.h>
  34 #include <linux/slab.h>
  35 #include <linux/vmalloc.h>
  36 #include <linux/platform_device.h>
  37 #include <linux/scatterlist.h>
  38 #include <asm/tlbflush.h>
  39 #include <kern_util.h>
  40 #include "mconsole_kern.h"
  41 #include <init.h>
  42 #include <irq_kern.h>
  43 #include "ubd.h"
  44 #include <os.h>
  45 #include "cow.h"
  46
  47 /* Max request size is determined by sector mask - 32K */
  48 #define UBD_MAX_REQUEST (8 * sizeof(long))
  49
  50 struct io_desc {
  51         char *buffer;
  52         unsigned long length;
  53         unsigned long sector_mask;
  54         unsigned long long cow_offset;
  55         unsigned long bitmap_words[2];
  56 };
  57
  58 struct io_thread_req {
  59         struct request *req;
  60         int fds[2];
  61         unsigned long offsets[2];
  62         unsigned long long offset;
  63         int sectorsize;
  64         int error;
  65
  66         int desc_cnt;
  67         /* io_desc has to be the last element of the struct */
  68         struct io_desc io_desc[];
  69 };
  70
  71
  72 static struct io_thread_req * (*irq_req_buffer)[];
  73 static struct io_thread_req *irq_remainder;
  74 static int irq_remainder_size;
  75
  76 static struct io_thread_req * (*io_req_buffer)[];
  77 static struct io_thread_req *io_remainder;
  78 static int io_remainder_size;
  79
  80
  81
  82 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  83 {
  84         __u64 n;
  85         int bits, off;
  86
  87         bits = sizeof(data[0]) * 8;
  88         n = bit / bits;
  89         off = bit % bits;
  90         return (data[n] & (1 << off)) != 0;
  91 }
  92
  93 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  94 {
  95         __u64 n;
  96         int bits, off;
  97
  98         bits = sizeof(data[0]) * 8;
  99         n = bit / bits;
 100         off = bit % bits;
 101         data[n] |= (1 << off);
 102 }
 103 /*End stuff from ubd_user.h*/
 104
 105 #define DRIVER_NAME "uml-blkdev"
 106
 107 static DEFINE_MUTEX(ubd_lock);
 108 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
 109
 110 static int ubd_open(struct block_device *bdev, fmode_t mode);
 111 static void ubd_release(struct gendisk *disk, fmode_t mode);
 112 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
 113                      unsigned int cmd, unsigned long arg);
 114 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 115
 116 #define MAX_DEV (16)
 117
 118 static const struct block_device_operations ubd_blops = {
 119         .owner          = THIS_MODULE,
 120         .open           = ubd_open,
 121         .release        = ubd_release,
 122         .ioctl          = ubd_ioctl,
 123         .compat_ioctl   = blkdev_compat_ptr_ioctl,
 124         .getgeo         = ubd_getgeo,
 125 };
 126
 127 /* Protected by ubd_lock */
 128 static int fake_major = UBD_MAJOR;
 129 static struct gendisk *ubd_gendisk[MAX_DEV];
 130 static struct gendisk *fake_gendisk[MAX_DEV];
 131
 132 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 133 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 134                                          .cl = 1 })
 135 #else
 136 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 137                                          .cl = 1 })
 138 #endif
 139 static struct openflags global_openflags = OPEN_FLAGS;
 140
 141 struct cow {
 142         /* backing file name */
 143         char *file;
 144         /* backing file fd */
 145         int fd;
 146         unsigned long *bitmap;
 147         unsigned long bitmap_len;
 148         int bitmap_offset;
 149         int data_offset;
 150 };
 151
 152 #define MAX_SG 64
 153
 154 struct ubd {
 155         /* name (and fd, below) of the file opened for writing, either the
 156          * backing or the cow file. */
 157         char *file;
 158         int count;
 159         int fd;
 160         __u64 size;
 161         struct openflags boot_openflags;
 162         struct openflags openflags;
 163         unsigned shared:1;
 164         unsigned no_cow:1;
 165         unsigned no_trim:1;
 166         struct cow cow;
 167         struct platform_device pdev;
 168         struct request_queue *queue;
 169         struct blk_mq_tag_set tag_set;
 170         spinlock_t lock;
 171 };
 172
 173 #define DEFAULT_COW { \
 174         .file =                 NULL, \
 175         .fd =                   -1,     \
 176         .bitmap =               NULL, \
 177         .bitmap_offset =        0, \
 178         .data_offset =          0, \
 179 }
 180
 181 #define DEFAULT_UBD { \
 182         .file =                 NULL, \
 183         .count =                0, \
 184         .fd =                   -1, \
 185         .size =                 -1, \
 186         .boot_openflags =       OPEN_FLAGS, \
 187         .openflags =            OPEN_FLAGS, \
 188         .no_cow =               0, \
 189         .no_trim =              0, \
 190         .shared =               0, \
 191         .cow =                  DEFAULT_COW, \
 192         .lock =                 __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
 193 }
 194
 195 /* Protected by ubd_lock */
 196 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
 197
 198 /* Only changed by fake_ide_setup which is a setup */
 199 static int fake_ide = 0;
 200 static struct proc_dir_entry *proc_ide_root = NULL;
 201 static struct proc_dir_entry *proc_ide = NULL;
 202
 203 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
 204                                  const struct blk_mq_queue_data *bd);
 205
 206 static void make_proc_ide(void)
 207 {
 208         proc_ide_root = proc_mkdir("ide", NULL);
 209         proc_ide = proc_mkdir("ide0", proc_ide_root);
 210 }
 211
 212 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
 213 {
 214         seq_puts(m, "disk\n");
 215         return 0;
 216 }
 217
 218 static void make_ide_entries(const char *dev_name)
 219 {
 220         struct proc_dir_entry *dir, *ent;
 221         char name[64];
 222
 223         if(proc_ide_root == NULL) make_proc_ide();
 224
 225         dir = proc_mkdir(dev_name, proc_ide);
 226         if(!dir) return;
 227
 228         ent = proc_create_single("media", S_IRUGO, dir,
 229                         fake_ide_media_proc_show);
 230         if(!ent) return;
 231         snprintf(name, sizeof(name), "ide0/%s", dev_name);
 232         proc_symlink(dev_name, proc_ide_root, name);
 233 }
 234
 235 static int fake_ide_setup(char *str)
 236 {
 237         fake_ide = 1;
 238         return 1;
 239 }
 240
 241 __setup("fake_ide", fake_ide_setup);
 242
 243 __uml_help(fake_ide_setup,
 244 "fake_ide\n"
 245 "    Create ide0 entries that map onto ubd devices.\n\n"
 246 );
 247
 248 static int parse_unit(char **ptr)
 249 {
 250         char *str = *ptr, *end;
 251         int n = -1;
 252
 253         if(isdigit(*str)) {
 254                 n = simple_strtoul(str, &end, 0);
 255                 if(end == str)
 256                         return -1;
 257                 *ptr = end;
 258         }
 259         else if (('a' <= *str) && (*str <= 'z')) {
 260                 n = *str - 'a';
 261                 str++;
 262                 *ptr = str;
 263         }
 264         return n;
 265 }
 266
 267 /* If *index_out == -1 at exit, the passed option was a general one;
 268  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 269  * should not be freed on exit.
 270  */
 271 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 272 {
 273         struct ubd *ubd_dev;
 274         struct openflags flags = global_openflags;
 275         char *backing_file;
 276         int n, err = 0, i;
 277
 278         if(index_out) *index_out = -1;
 279         n = *str;
 280         if(n == '='){
 281                 char *end;
 282                 int major;
 283
 284                 str++;
 285                 if(!strcmp(str, "sync")){
 286                         global_openflags = of_sync(global_openflags);
 287                         return err;
 288                 }
 289
 290                 err = -EINVAL;
 291                 major = simple_strtoul(str, &end, 0);
 292                 if((*end != '\0') || (end == str)){
 293                         *error_out = "Didn't parse major number";
 294                         return err;
 295                 }
 296
 297                 mutex_lock(&ubd_lock);
 298                 if (fake_major != UBD_MAJOR) {
 299                         *error_out = "Can't assign a fake major twice";
 300                         goto out1;
 301                 }
 302
 303                 fake_major = major;
 304
 305                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 306                        major);
 307                 err = 0;
 308         out1:
 309                 mutex_unlock(&ubd_lock);
 310                 return err;
 311         }
 312
 313         n = parse_unit(&str);
 314         if(n < 0){
 315                 *error_out = "Couldn't parse device number";
 316                 return -EINVAL;
 317         }
 318         if(n >= MAX_DEV){
 319                 *error_out = "Device number out of range";
 320                 return 1;
 321         }
 322
 323         err = -EBUSY;
 324         mutex_lock(&ubd_lock);
 325
 326         ubd_dev = &ubd_devs[n];
 327         if(ubd_dev->file != NULL){
 328                 *error_out = "Device is already configured";
 329                 goto out;
 330         }
 331
 332         if (index_out)
 333                 *index_out = n;
 334
 335         err = -EINVAL;
 336         for (i = 0; i < sizeof("rscdt="); i++) {
 337                 switch (*str) {
 338                 case 'r':
 339                         flags.w = 0;
 340                         break;
 341                 case 's':
 342                         flags.s = 1;
 343                         break;
 344                 case 'd':
 345                         ubd_dev->no_cow = 1;
 346                         break;
 347                 case 'c':
 348                         ubd_dev->shared = 1;
 349                         break;
 350                 case 't':
 351                         ubd_dev->no_trim = 1;
 352                         break;
 353                 case '=':
 354                         str++;
 355                         goto break_loop;
 356                 default:
 357                         *error_out = "Expected '=' or flag letter "
 358                                 "(r, s, c, t or d)";
 359                         goto out;
 360                 }
 361                 str++;
 362         }
 363
 364         if (*str == '=')
 365                 *error_out = "Too many flags specified";
 366         else
 367                 *error_out = "Missing '='";
 368         goto out;
 369
 370 break_loop:
 371         backing_file = strchr(str, ',');
 372
 373         if (backing_file == NULL)
 374                 backing_file = strchr(str, ':');
 375
 376         if(backing_file != NULL){
 377                 if(ubd_dev->no_cow){
 378                         *error_out = "Can't specify both 'd' and a cow file";
 379                         goto out;
 380                 }
 381                 else {
 382                         *backing_file = '\0';
 383                         backing_file++;
 384                 }
 385         }
 386         err = 0;
 387         ubd_dev->file = str;
 388         ubd_dev->cow.file = backing_file;
 389         ubd_dev->boot_openflags = flags;
 390 out:
 391         mutex_unlock(&ubd_lock);
 392         return err;
 393 }
 394
 395 static int ubd_setup(char *str)
 396 {
 397         char *error;
 398         int err;
 399
 400         err = ubd_setup_common(str, NULL, &error);
 401         if(err)
 402                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 403                        "%s\n", str, error);
 404         return 1;
 405 }
 406
 407 __setup("ubd", ubd_setup);
 408 __uml_help(ubd_setup,
 409 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 410 "    This is used to associate a device with a file in the underlying\n"
 411 "    filesystem. When specifying two filenames, the first one is the\n"
 412 "    COW name and the second is the backing file name. As separator you can\n"
 413 "    use either a ':' or a ',': the first one allows writing things like;\n"
 414 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 415 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 416 "    When using only one filename, UML will detect whether to treat it like\n"
 417 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 418 "    flag:\n"
 419 "       ubd0d=BackingFile\n"
 420 "    Usually, there is a filesystem in the file, but \n"
 421 "    that's not required. Swap devices containing swap files can be\n"
 422 "    specified like this. Also, a file which doesn't contain a\n"
 423 "    filesystem can have its contents read in the virtual \n"
 424 "    machine by running 'dd' on the device. <n> must be in the range\n"
 425 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 426 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 427 "    an 's' will cause data to be written to disk on the host immediately.\n"
 428 "    'c' will cause the device to be treated as being shared between multiple\n"
 429 "    UMLs and file locking will be turned off - this is appropriate for a\n"
 430 "    cluster filesystem and inappropriate at almost all other times.\n\n"
 431 "    't' will disable trim/discard support on the device (enabled by default).\n\n"
 432 );
 433
 434 static int udb_setup(char *str)
 435 {
 436         printk("udb%s specified on command line is almost certainly a ubd -> "
 437                "udb TYPO\n", str);
 438         return 1;
 439 }
 440
 441 __setup("udb", udb_setup);
 442 __uml_help(udb_setup,
 443 "udb\n"
 444 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 445 "    to impossible to catch visually unless you specifically look for\n"
 446 "    them.  The only result of any option starting with 'udb' is an error\n"
 447 "    in the boot output.\n\n"
 448 );
 449
 450 /* Only changed by ubd_init, which is an initcall. */
 451 static int thread_fd = -1;
 452
 453 /* Function to read several request pointers at a time
 454 * handling fractional reads if (and as) needed
 455 */
 456
 457 static int bulk_req_safe_read(
 458         int fd,
 459         struct io_thread_req * (*request_buffer)[],
 460         struct io_thread_req **remainder,
 461         int *remainder_size,
 462         int max_recs
 463         )
 464 {
 465         int n = 0;
 466         int res = 0;
 467
 468         if (*remainder_size > 0) {
 469                 memmove(
 470                         (char *) request_buffer,
 471                         (char *) remainder, *remainder_size
 472                 );
 473                 n = *remainder_size;
 474         }
 475
 476         res = os_read_file(
 477                         fd,
 478                         ((char *) request_buffer) + *remainder_size,
 479                         sizeof(struct io_thread_req *)*max_recs
 480                                 - *remainder_size
 481                 );
 482         if (res > 0) {
 483                 n += res;
 484                 if ((n % sizeof(struct io_thread_req *)) > 0) {
 485                         /*
 486                         * Read somehow returned not a multiple of dword
 487                         * theoretically possible, but never observed in the
 488                         * wild, so read routine must be able to handle it
 489                         */
 490                         *remainder_size = n % sizeof(struct io_thread_req *);
 491                         WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
 492                         memmove(
 493                                 remainder,
 494                                 ((char *) request_buffer) +
 495                                         (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
 496                                 *remainder_size
 497                         );
 498                         n = n - *remainder_size;
 499                 }
 500         } else {
 501                 n = res;
 502         }
 503         return n;
 504 }
 505
 506 /* Called without dev->lock held, and only in interrupt context. */
 507 static void ubd_handler(void)
 508 {
 509         int n;
 510         int count;
 511
 512         while(1){
 513                 n = bulk_req_safe_read(
 514                         thread_fd,
 515                         irq_req_buffer,
 516                         &irq_remainder,
 517                         &irq_remainder_size,
 518                         UBD_REQ_BUFFER_SIZE
 519                 );
 520                 if (n < 0) {
 521                         if(n == -EAGAIN)
 522                                 break;
 523                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 524                                "err = %d\n", -n);
 525                         return;
 526                 }
 527                 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
 528                         struct io_thread_req *io_req = (*irq_req_buffer)[count];
 529
 530                         if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
 531                                 blk_queue_max_discard_sectors(io_req->req->q, 0);
 532                                 blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
 533                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, io_req->req->q);
 534                         }
 535                         blk_mq_end_request(io_req->req, io_req->error);
 536                         kfree(io_req);
 537                 }
 538         }
 539 }
 540
 541 static irqreturn_t ubd_intr(int irq, void *dev)
 542 {
 543         ubd_handler();
 544         return IRQ_HANDLED;
 545 }
 546
 547 /* Only changed by ubd_init, which is an initcall. */
 548 static int io_pid = -1;
 549
 550 static void kill_io_thread(void)
 551 {
 552         if(io_pid != -1)
 553                 os_kill_process(io_pid, 1);
 554 }
 555
 556 __uml_exitcall(kill_io_thread);
 557
 558 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 559 {
 560         char *file;
 561         int fd;
 562         int err;
 563
 564         __u32 version;
 565         __u32 align;
 566         char *backing_file;
 567         time64_t mtime;
 568         unsigned long long size;
 569         int sector_size;
 570         int bitmap_offset;
 571
 572         if (ubd_dev->file && ubd_dev->cow.file) {
 573                 file = ubd_dev->cow.file;
 574
 575                 goto out;
 576         }
 577
 578         fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
 579         if (fd < 0)
 580                 return fd;
 581
 582         err = read_cow_header(file_reader, &fd, &version, &backing_file, \
 583                 &mtime, &size, &sector_size, &align, &bitmap_offset);
 584         os_close_file(fd);
 585
 586         if(err == -EINVAL)
 587                 file = ubd_dev->file;
 588         else
 589                 file = backing_file;
 590
 591 out:
 592         return os_file_size(file, size_out);
 593 }
 594
 595 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
 596 {
 597         int err;
 598
 599         err = os_pread_file(fd, buf, len, offset);
 600         if (err < 0)
 601                 return err;
 602
 603         return 0;
 604 }
 605
 606 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
 607 {
 608         time64_t modtime;
 609         unsigned long long actual;
 610         int err;
 611
 612         err = os_file_modtime(file, &modtime);
 613         if (err < 0) {
 614                 printk(KERN_ERR "Failed to get modification time of backing "
 615                        "file \"%s\", err = %d\n", file, -err);
 616                 return err;
 617         }
 618
 619         err = os_file_size(file, &actual);
 620         if (err < 0) {
 621                 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
 622                        "err = %d\n", file, -err);
 623                 return err;
 624         }
 625
 626         if (actual != size) {
 627                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
 628                  * the typecast.*/
 629                 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
 630                        "vs backing file\n", (unsigned long long) size, actual);
 631                 return -EINVAL;
 632         }
 633         if (modtime != mtime) {
 634                 printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
 635                        "backing file\n", mtime, modtime);
 636                 return -EINVAL;
 637         }
 638         return 0;
 639 }
 640
 641 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
 642 {
 643         struct uml_stat buf1, buf2;
 644         int err;
 645
 646         if (from_cmdline == NULL)
 647                 return 0;
 648         if (!strcmp(from_cmdline, from_cow))
 649                 return 0;
 650
 651         err = os_stat_file(from_cmdline, &buf1);
 652         if (err < 0) {
 653                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
 654                        -err);
 655                 return 0;
 656         }
 657         err = os_stat_file(from_cow, &buf2);
 658         if (err < 0) {
 659                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
 660                        -err);
 661                 return 1;
 662         }
 663         if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
 664                 return 0;
 665
 666         printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
 667                "\"%s\" specified in COW header of \"%s\"\n",
 668                from_cmdline, from_cow, cow);
 669         return 1;
 670 }
 671
 672 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
 673                   char **backing_file_out, int *bitmap_offset_out,
 674                   unsigned long *bitmap_len_out, int *data_offset_out,
 675                   int *create_cow_out)
 676 {
 677         time64_t mtime;
 678         unsigned long long size;
 679         __u32 version, align;
 680         char *backing_file;
 681         int fd, err, sectorsize, asked_switch, mode = 0644;
 682
 683         fd = os_open_file(file, *openflags, mode);
 684         if (fd < 0) {
 685                 if ((fd == -ENOENT) && (create_cow_out != NULL))
 686                         *create_cow_out = 1;
 687                 if (!openflags->w ||
 688                     ((fd != -EROFS) && (fd != -EACCES)))
 689                         return fd;
 690                 openflags->w = 0;
 691                 fd = os_open_file(file, *openflags, mode);
 692                 if (fd < 0)
 693                         return fd;
 694         }
 695
 696         if (shared)
 697                 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
 698         else {
 699                 err = os_lock_file(fd, openflags->w);
 700                 if (err < 0) {
 701                         printk(KERN_ERR "Failed to lock '%s', err = %d\n",
 702                                file, -err);
 703                         goto out_close;
 704                 }
 705         }
 706
 707         /* Successful return case! */
 708         if (backing_file_out == NULL)
 709                 return fd;
 710
 711         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
 712                               &size, &sectorsize, &align, bitmap_offset_out);
 713         if (err && (*backing_file_out != NULL)) {
 714                 printk(KERN_ERR "Failed to read COW header from COW file "
 715                        "\"%s\", errno = %d\n", file, -err);
 716                 goto out_close;
 717         }
 718         if (err)
 719                 return fd;
 720
 721         asked_switch = path_requires_switch(*backing_file_out, backing_file,
 722                                             file);
 723
 724         /* Allow switching only if no mismatch. */
 725         if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
 726                                                    mtime)) {
 727                 printk(KERN_ERR "Switching backing file to '%s'\n",
 728                        *backing_file_out);
 729                 err = write_cow_header(file, fd, *backing_file_out,
 730                                        sectorsize, align, &size);
 731                 if (err) {
 732                         printk(KERN_ERR "Switch failed, errno = %d\n", -err);
 733                         goto out_close;
 734                 }
 735         } else {
 736                 *backing_file_out = backing_file;
 737                 err = backing_file_mismatch(*backing_file_out, size, mtime);
 738                 if (err)
 739                         goto out_close;
 740         }
 741
 742         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
 743                   bitmap_len_out, data_offset_out);
 744
 745         return fd;
 746  out_close:
 747         os_close_file(fd);
 748         return err;
 749 }
 750
 751 static int create_cow_file(char *cow_file, char *backing_file,
 752                     struct openflags flags,
 753                     int sectorsize, int alignment, int *bitmap_offset_out,
 754                     unsigned long *bitmap_len_out, int *data_offset_out)
 755 {
 756         int err, fd;
 757
 758         flags.c = 1;
 759         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
 760         if (fd < 0) {
 761                 err = fd;
 762                 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
 763                        cow_file, -err);
 764                 goto out;
 765         }
 766
 767         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
 768                             bitmap_offset_out, bitmap_len_out,
 769                             data_offset_out);
 770         if (!err)
 771                 return fd;
 772         os_close_file(fd);
 773  out:
 774         return err;
 775 }
 776
 777 static void ubd_close_dev(struct ubd *ubd_dev)
 778 {
 779         os_close_file(ubd_dev->fd);
 780         if(ubd_dev->cow.file == NULL)
 781                 return;
 782
 783         os_close_file(ubd_dev->cow.fd);
 784         vfree(ubd_dev->cow.bitmap);
 785         ubd_dev->cow.bitmap = NULL;
 786 }
 787
 788 static int ubd_open_dev(struct ubd *ubd_dev)
 789 {
 790         struct openflags flags;
 791         char **back_ptr;
 792         int err, create_cow, *create_ptr;
 793         int fd;
 794
 795         ubd_dev->openflags = ubd_dev->boot_openflags;
 796         create_cow = 0;
 797         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 798         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 799
 800         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 801                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 802                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 803                                 create_ptr);
 804
 805         if((fd == -ENOENT) && create_cow){
 806                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 807                                           ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
 808                                           &ubd_dev->cow.bitmap_offset,
 809                                           &ubd_dev->cow.bitmap_len,
 810                                           &ubd_dev->cow.data_offset);
 811                 if(fd >= 0){
 812                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 813                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 814                 }
 815         }
 816
 817         if(fd < 0){
 818                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 819                        -fd);
 820                 return fd;
 821         }
 822         ubd_dev->fd = fd;
 823
 824         if(ubd_dev->cow.file != NULL){
 825                 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
 826
 827                 err = -ENOMEM;
 828                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 829                 if(ubd_dev->cow.bitmap == NULL){
 830                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 831                         goto error;
 832                 }
 833                 flush_tlb_kernel_vm();
 834
 835                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 836                                       ubd_dev->cow.bitmap_offset,
 837                                       ubd_dev->cow.bitmap_len);
 838                 if(err < 0)
 839                         goto error;
 840
 841                 flags = ubd_dev->openflags;
 842                 flags.w = 0;
 843                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 844                                     NULL, NULL, NULL, NULL);
 845                 if(err < 0) goto error;
 846                 ubd_dev->cow.fd = err;
 847         }
 848         if (ubd_dev->no_trim == 0) {
 849                 ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE;
 850                 ubd_dev->queue->limits.discard_alignment = SECTOR_SIZE;
 851                 blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
 852                 blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
 853                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, ubd_dev->queue);
 854         }
 855         blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue);
 856         return 0;
 857  error:
 858         os_close_file(ubd_dev->fd);
 859         return err;
 860 }
 861
 862 static void ubd_device_release(struct device *dev)
 863 {
 864         struct ubd *ubd_dev = dev_get_drvdata(dev);
 865
 866         blk_cleanup_queue(ubd_dev->queue);
 867         blk_mq_free_tag_set(&ubd_dev->tag_set);
 868         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 869 }
 870
 871 static int ubd_disk_register(int major, u64 size, int unit,
 872                              struct gendisk **disk_out)
 873 {
 874         struct device *parent = NULL;
 875         struct gendisk *disk;
 876
 877         disk = alloc_disk(1 << UBD_SHIFT);
 878         if(disk == NULL)
 879                 return -ENOMEM;
 880
 881         disk->major = major;
 882         disk->first_minor = unit << UBD_SHIFT;
 883         disk->fops = &ubd_blops;
 884         set_capacity(disk, size / 512);
 885         if (major == UBD_MAJOR)
 886                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 887         else
 888                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 889
 890         /* sysfs register (not for ide fake devices) */
 891         if (major == UBD_MAJOR) {
 892                 ubd_devs[unit].pdev.id   = unit;
 893                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 894                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 895                 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
 896                 platform_device_register(&ubd_devs[unit].pdev);
 897                 parent = &ubd_devs[unit].pdev.dev;
 898         }
 899
 900         disk->private_data = &ubd_devs[unit];
 901         disk->queue = ubd_devs[unit].queue;
 902         device_add_disk(parent, disk, NULL);
 903
 904         *disk_out = disk;
 905         return 0;
 906 }
 907
 908 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
 909
 910 static const struct blk_mq_ops ubd_mq_ops = {
 911         .queue_rq = ubd_queue_rq,
 912 };
 913
 914 static int ubd_add(int n, char **error_out)
 915 {
 916         struct ubd *ubd_dev = &ubd_devs[n];
 917         int err = 0;
 918
 919         if(ubd_dev->file == NULL)
 920                 goto out;
 921
 922         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 923         if(err < 0){
 924                 *error_out = "Couldn't determine size of device's file";
 925                 goto out;
 926         }
 927
 928         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 929
 930         ubd_dev->tag_set.ops = &ubd_mq_ops;
 931         ubd_dev->tag_set.queue_depth = 64;
 932         ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
 933         ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 934         ubd_dev->tag_set.driver_data = ubd_dev;
 935         ubd_dev->tag_set.nr_hw_queues = 1;
 936
 937         err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
 938         if (err)
 939                 goto out;
 940
 941         ubd_dev->queue = blk_mq_init_queue(&ubd_dev->tag_set);
 942         if (IS_ERR(ubd_dev->queue)) {
 943                 err = PTR_ERR(ubd_dev->queue);
 944                 goto out_cleanup_tags;
 945         }
 946
 947         ubd_dev->queue->queuedata = ubd_dev;
 948         blk_queue_write_cache(ubd_dev->queue, true, false);
 949
 950         blk_queue_max_segments(ubd_dev->queue, MAX_SG);
 951         blk_queue_segment_boundary(ubd_dev->queue, PAGE_SIZE - 1);
 952         err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
 953         if(err){
 954                 *error_out = "Failed to register device";
 955                 goto out_cleanup_tags;
 956         }
 957
 958         if (fake_major != UBD_MAJOR)
 959                 ubd_disk_register(fake_major, ubd_dev->size, n,
 960                                   &fake_gendisk[n]);
 961
 962         /*
 963          * Perhaps this should also be under the "if (fake_major)" above
 964          * using the fake_disk->disk_name
 965          */
 966         if (fake_ide)
 967                 make_ide_entries(ubd_gendisk[n]->disk_name);
 968
 969         err = 0;
 970 out:
 971         return err;
 972
 973 out_cleanup_tags:
 974         blk_mq_free_tag_set(&ubd_dev->tag_set);
 975         if (!(IS_ERR(ubd_dev->queue)))
 976                 blk_cleanup_queue(ubd_dev->queue);
 977         goto out;
 978 }
 979
 980 static int ubd_config(char *str, char **error_out)
 981 {
 982         int n, ret;
 983
 984         /* This string is possibly broken up and stored, so it's only
 985          * freed if ubd_setup_common fails, or if only general options
 986          * were set.
 987          */
 988         str = kstrdup(str, GFP_KERNEL);
 989         if (str == NULL) {
 990                 *error_out = "Failed to allocate memory";
 991                 return -ENOMEM;
 992         }
 993
 994         ret = ubd_setup_common(str, &n, error_out);
 995         if (ret)
 996                 goto err_free;
 997
 998         if (n == -1) {
 999                 ret = 0;
1000                 goto err_free;
1001         }
1002
1003         mutex_lock(&ubd_lock);
1004         ret = ubd_add(n, error_out);
1005         if (ret)
1006                 ubd_devs[n].file = NULL;
1007         mutex_unlock(&ubd_lock);
1008
1009 out:
1010         return ret;
1011
1012 err_free:
1013         kfree(str);
1014         goto out;
1015 }
1016
1017 static int ubd_get_config(char *name, char *str, int size, char **error_out)
1018 {
1019         struct ubd *ubd_dev;
1020         int n, len = 0;
1021
1022         n = parse_unit(&name);
1023         if((n >= MAX_DEV) || (n < 0)){
1024                 *error_out = "ubd_get_config : device number out of range";
1025                 return -1;
1026         }
1027
1028         ubd_dev = &ubd_devs[n];
1029         mutex_lock(&ubd_lock);
1030
1031         if(ubd_dev->file == NULL){
1032                 CONFIG_CHUNK(str, size, len, "", 1);
1033                 goto out;
1034         }
1035
1036         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
1037
1038         if(ubd_dev->cow.file != NULL){
1039                 CONFIG_CHUNK(str, size, len, ",", 0);
1040                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
1041         }
1042         else CONFIG_CHUNK(str, size, len, "", 1);
1043
1044  out:
1045         mutex_unlock(&ubd_lock);
1046         return len;
1047 }
1048
1049 static int ubd_id(char **str, int *start_out, int *end_out)
1050 {
1051         int n;
1052
1053         n = parse_unit(str);
1054         *start_out = 0;
1055         *end_out = MAX_DEV - 1;
1056         return n;
1057 }
1058
1059 static int ubd_remove(int n, char **error_out)
1060 {
1061         struct gendisk *disk = ubd_gendisk[n];
1062         struct ubd *ubd_dev;
1063         int err = -ENODEV;
1064
1065         mutex_lock(&ubd_lock);
1066
1067         ubd_dev = &ubd_devs[n];
1068
1069         if(ubd_dev->file == NULL)
1070                 goto out;
1071
1072         /* you cannot remove a open disk */
1073         err = -EBUSY;
1074         if(ubd_dev->count > 0)
1075                 goto out;
1076
1077         ubd_gendisk[n] = NULL;
1078         if(disk != NULL){
1079                 del_gendisk(disk);
1080                 put_disk(disk);
1081         }
1082
1083         if(fake_gendisk[n] != NULL){
1084                 del_gendisk(fake_gendisk[n]);
1085                 put_disk(fake_gendisk[n]);
1086                 fake_gendisk[n] = NULL;
1087         }
1088
1089         err = 0;
1090         platform_device_unregister(&ubd_dev->pdev);
1091 out:
1092         mutex_unlock(&ubd_lock);
1093         return err;
1094 }
1095
1096 /* All these are called by mconsole in process context and without
1097  * ubd-specific locks.  The structure itself is const except for .list.
1098  */
1099 static struct mc_device ubd_mc = {
1100         .list           = LIST_HEAD_INIT(ubd_mc.list),
1101         .name           = "ubd",
1102         .config         = ubd_config,
1103         .get_config     = ubd_get_config,
1104         .id             = ubd_id,
1105         .remove         = ubd_remove,
1106 };
1107
1108 static int __init ubd_mc_init(void)
1109 {
1110         mconsole_register_dev(&ubd_mc);
1111         return 0;
1112 }
1113
1114 __initcall(ubd_mc_init);
1115
1116 static int __init ubd0_init(void)
1117 {
1118         struct ubd *ubd_dev = &ubd_devs[0];
1119
1120         mutex_lock(&ubd_lock);
1121         if(ubd_dev->file == NULL)
1122                 ubd_dev->file = "root_fs";
1123         mutex_unlock(&ubd_lock);
1124
1125         return 0;
1126 }
1127
1128 __initcall(ubd0_init);
1129
1130 /* Used in ubd_init, which is an initcall */
1131 static struct platform_driver ubd_driver = {
1132         .driver = {
1133                 .name  = DRIVER_NAME,
1134         },
1135 };
1136
1137 static int __init ubd_init(void)
1138 {
1139         char *error;
1140         int i, err;
1141
1142         if (register_blkdev(UBD_MAJOR, "ubd"))
1143                 return -1;
1144
1145         if (fake_major != UBD_MAJOR) {
1146                 char name[sizeof("ubd_nnn\0")];
1147
1148                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1149                 if (register_blkdev(fake_major, "ubd"))
1150                         return -1;
1151         }
1152
1153         irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1154                                        sizeof(struct io_thread_req *),
1155                                        GFP_KERNEL
1156                 );
1157         irq_remainder = 0;
1158
1159         if (irq_req_buffer == NULL) {
1160                 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1161                 return -1;
1162         }
1163         io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1164                                       sizeof(struct io_thread_req *),
1165                                       GFP_KERNEL
1166                 );
1167
1168         io_remainder = 0;
1169
1170         if (io_req_buffer == NULL) {
1171                 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1172                 return -1;
1173         }
1174         platform_driver_register(&ubd_driver);
1175         mutex_lock(&ubd_lock);
1176         for (i = 0; i < MAX_DEV; i++){
1177                 err = ubd_add(i, &error);
1178                 if(err)
1179                         printk(KERN_ERR "Failed to initialize ubd device %d :"
1180                                "%s\n", i, error);
1181         }
1182         mutex_unlock(&ubd_lock);
1183         return 0;
1184 }
1185
1186 late_initcall(ubd_init);
1187
1188 static int __init ubd_driver_init(void){
1189         unsigned long stack;
1190         int err;
1191
1192         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1193         if(global_openflags.s){
1194                 printk(KERN_INFO "ubd: Synchronous mode\n");
1195                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1196                  * enough. So use anyway the io thread. */
1197         }
1198         stack = alloc_stack(0, 0);
1199         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1200                                  &thread_fd);
1201         if(io_pid < 0){
1202                 printk(KERN_ERR
1203                        "ubd : Failed to start I/O thread (errno = %d) - "
1204                        "falling back to synchronous I/O\n", -io_pid);
1205                 io_pid = -1;
1206                 return 0;
1207         }
1208         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1209                              0, "ubd", ubd_devs);
1210         if(err != 0)
1211                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1212         return 0;
1213 }
1214
1215 device_initcall(ubd_driver_init);
1216
1217 static int ubd_open(struct block_device *bdev, fmode_t mode)
1218 {
1219         struct gendisk *disk = bdev->bd_disk;
1220         struct ubd *ubd_dev = disk->private_data;
1221         int err = 0;
1222
1223         mutex_lock(&ubd_mutex);
1224         if(ubd_dev->count == 0){
1225                 err = ubd_open_dev(ubd_dev);
1226                 if(err){
1227                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1228                                disk->disk_name, ubd_dev->file, -err);
1229                         goto out;
1230                 }
1231         }
1232         ubd_dev->count++;
1233         set_disk_ro(disk, !ubd_dev->openflags.w);
1234
1235         /* This should no more be needed. And it didn't work anyway to exclude
1236          * read-write remounting of filesystems.*/
1237         /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1238                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1239                 err = -EROFS;
1240         }*/
1241 out:
1242         mutex_unlock(&ubd_mutex);
1243         return err;
1244 }
1245
1246 static void ubd_release(struct gendisk *disk, fmode_t mode)
1247 {
1248         struct ubd *ubd_dev = disk->private_data;
1249
1250         mutex_lock(&ubd_mutex);
1251         if(--ubd_dev->count == 0)
1252                 ubd_close_dev(ubd_dev);
1253         mutex_unlock(&ubd_mutex);
1254 }
1255
1256 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1257                           __u64 *cow_offset, unsigned long *bitmap,
1258                           __u64 bitmap_offset, unsigned long *bitmap_words,
1259                           __u64 bitmap_len)
1260 {
1261         __u64 sector = io_offset >> SECTOR_SHIFT;
1262         int i, update_bitmap = 0;
1263
1264         for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1265                 if(cow_mask != NULL)
1266                         ubd_set_bit(i, (unsigned char *) cow_mask);
1267                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1268                         continue;
1269
1270                 update_bitmap = 1;
1271                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1272         }
1273
1274         if(!update_bitmap)
1275                 return;
1276
1277         *cow_offset = sector / (sizeof(unsigned long) * 8);
1278
1279         /* This takes care of the case where we're exactly at the end of the
1280          * device, and *cow_offset + 1 is off the end.  So, just back it up
1281          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1282          * for the original diagnosis.
1283          */
1284         if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1285                                          sizeof(unsigned long)) - 1))
1286                 (*cow_offset)--;
1287
1288         bitmap_words[0] = bitmap[*cow_offset];
1289         bitmap_words[1] = bitmap[*cow_offset + 1];
1290
1291         *cow_offset *= sizeof(unsigned long);
1292         *cow_offset += bitmap_offset;
1293 }
1294
1295 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1296                        unsigned long offset, unsigned long *bitmap,
1297                        __u64 bitmap_offset, __u64 bitmap_len)
1298 {
1299         __u64 sector = offset >> SECTOR_SHIFT;
1300         int i;
1301
1302         if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1303                 panic("Operation too long");
1304
1305         if (req_op(req->req) == REQ_OP_READ) {
1306                 for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1307                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1308                                 ubd_set_bit(i, (unsigned char *)
1309                                             &segment->sector_mask);
1310                 }
1311         } else {
1312                 cowify_bitmap(offset, segment->length, &segment->sector_mask,
1313                               &segment->cow_offset, bitmap, bitmap_offset,
1314                               segment->bitmap_words, bitmap_len);
1315         }
1316 }
1317
1318 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1319                         struct request *req)
1320 {
1321         struct bio_vec bvec;
1322         struct req_iterator iter;
1323         int i = 0;
1324         unsigned long byte_offset = io_req->offset;
1325         int op = req_op(req);
1326
1327         if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1328                 io_req->io_desc[0].buffer = NULL;
1329                 io_req->io_desc[0].length = blk_rq_bytes(req);
1330         } else {
1331                 rq_for_each_segment(bvec, req, iter) {
1332                         BUG_ON(i >= io_req->desc_cnt);
1333
1334                         io_req->io_desc[i].buffer =
1335                                 page_address(bvec.bv_page) + bvec.bv_offset;
1336                         io_req->io_desc[i].length = bvec.bv_len;
1337                         i++;
1338                 }
1339         }
1340
1341         if (dev->cow.file) {
1342                 for (i = 0; i < io_req->desc_cnt; i++) {
1343                         cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1344                                    dev->cow.bitmap, dev->cow.bitmap_offset,
1345                                    dev->cow.bitmap_len);
1346                         byte_offset += io_req->io_desc[i].length;
1347                 }
1348
1349         }
1350 }
1351
1352 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1353                                            int desc_cnt)
1354 {
1355         struct io_thread_req *io_req;
1356         int i;
1357
1358         io_req = kmalloc(sizeof(*io_req) +
1359                          (desc_cnt * sizeof(struct io_desc)),
1360                          GFP_ATOMIC);
1361         if (!io_req)
1362                 return NULL;
1363
1364         io_req->req = req;
1365         if (dev->cow.file)
1366                 io_req->fds[0] = dev->cow.fd;
1367         else
1368                 io_req->fds[0] = dev->fd;
1369         io_req->error = 0;
1370         io_req->sectorsize = SECTOR_SIZE;
1371         io_req->fds[1] = dev->fd;
1372         io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1373         io_req->offsets[0] = 0;
1374         io_req->offsets[1] = dev->cow.data_offset;
1375
1376         for (i = 0 ; i < desc_cnt; i++) {
1377                 io_req->io_desc[i].sector_mask = 0;
1378                 io_req->io_desc[i].cow_offset = -1;
1379         }
1380
1381         return io_req;
1382 }
1383
1384 static int ubd_submit_request(struct ubd *dev, struct request *req)
1385 {
1386         int segs = 0;
1387         struct io_thread_req *io_req;
1388         int ret;
1389         int op = req_op(req);
1390
1391         if (op == REQ_OP_FLUSH)
1392                 segs = 0;
1393         else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1394                 segs = 1;
1395         else
1396                 segs = blk_rq_nr_phys_segments(req);
1397
1398         io_req = ubd_alloc_req(dev, req, segs);
1399         if (!io_req)
1400                 return -ENOMEM;
1401
1402         io_req->desc_cnt = segs;
1403         if (segs)
1404                 ubd_map_req(dev, io_req, req);
1405
1406         ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1407         if (ret != sizeof(io_req)) {
1408                 if (ret != -EAGAIN)
1409                         pr_err("write to io thread failed: %d\n", -ret);
1410                 kfree(io_req);
1411         }
1412         return ret;
1413 }
1414
1415 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1416                                  const struct blk_mq_queue_data *bd)
1417 {
1418         struct ubd *ubd_dev = hctx->queue->queuedata;
1419         struct request *req = bd->rq;
1420         int ret = 0, res = BLK_STS_OK;
1421
1422         blk_mq_start_request(req);
1423
1424         spin_lock_irq(&ubd_dev->lock);
1425
1426         switch (req_op(req)) {
1427         case REQ_OP_FLUSH:
1428         case REQ_OP_READ:
1429         case REQ_OP_WRITE:
1430         case REQ_OP_DISCARD:
1431         case REQ_OP_WRITE_ZEROES:
1432                 ret = ubd_submit_request(ubd_dev, req);
1433                 break;
1434         default:
1435                 WARN_ON_ONCE(1);
1436                 res = BLK_STS_NOTSUPP;
1437         }
1438
1439         spin_unlock_irq(&ubd_dev->lock);
1440
1441         if (ret < 0) {
1442                 if (ret == -ENOMEM)
1443                         res = BLK_STS_RESOURCE;
1444                 else
1445                         res = BLK_STS_DEV_RESOURCE;
1446         }
1447
1448         return res;
1449 }
1450
1451 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1452 {
1453         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1454
1455         geo->heads = 128;
1456         geo->sectors = 32;
1457         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1458         return 0;
1459 }
1460
1461 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1462                      unsigned int cmd, unsigned long arg)
1463 {
1464         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1465         u16 ubd_id[ATA_ID_WORDS];
1466
1467         switch (cmd) {
1468                 struct cdrom_volctrl volume;
1469         case HDIO_GET_IDENTITY:
1470                 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1471                 ubd_id[ATA_ID_CYLS]     = ubd_dev->size / (128 * 32 * 512);
1472                 ubd_id[ATA_ID_HEADS]    = 128;
1473                 ubd_id[ATA_ID_SECTORS]  = 32;
1474                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1475                                  sizeof(ubd_id)))
1476                         return -EFAULT;
1477                 return 0;
1478
1479         case CDROMVOLREAD:
1480                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1481                         return -EFAULT;
1482                 volume.channel0 = 255;
1483                 volume.channel1 = 255;
1484                 volume.channel2 = 255;
1485                 volume.channel3 = 255;
1486                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1487                         return -EFAULT;
1488                 return 0;
1489         }
1490         return -EINVAL;
1491 }
1492
1493 static int map_error(int error_code)
1494 {
1495         switch (error_code) {
1496         case 0:
1497                 return BLK_STS_OK;
1498         case ENOSYS:
1499         case EOPNOTSUPP:
1500                 return BLK_STS_NOTSUPP;
1501         case ENOSPC:
1502                 return BLK_STS_NOSPC;
1503         }
1504         return BLK_STS_IOERR;
1505 }
1506
1507 /*
1508  * Everything from here onwards *IS NOT PART OF THE KERNEL*
1509  *
1510  * The following functions are part of UML hypervisor code.
1511  * All functions from here onwards are executed as a helper
1512  * thread and are not allowed to execute any kernel functions.
1513  *
1514  * Any communication must occur strictly via shared memory and IPC.
1515  *
1516  * Do not add printks, locks, kernel memory operations, etc - it
1517  * will result in unpredictable behaviour and/or crashes.
1518  */
1519
1520 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1521 {
1522         int n;
1523
1524         if (segment->cow_offset == -1)
1525                 return map_error(0);
1526
1527         n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1528                           sizeof(segment->bitmap_words), segment->cow_offset);
1529         if (n != sizeof(segment->bitmap_words))
1530                 return map_error(-n);
1531
1532         return map_error(0);
1533 }
1534
1535 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1536 {
1537         char *buf = NULL;
1538         unsigned long len;
1539         int n, nsectors, start, end, bit;
1540         __u64 off;
1541
1542         /* FLUSH is really a special case, we cannot "case" it with others */
1543
1544         if (req_op(req->req) == REQ_OP_FLUSH) {
1545                 /* fds[0] is always either the rw image or our cow file */
1546                 req->error = map_error(-os_sync_file(req->fds[0]));
1547                 return;
1548         }
1549
1550         nsectors = desc->length / req->sectorsize;
1551         start = 0;
1552         do {
1553                 bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1554                 end = start;
1555                 while((end < nsectors) &&
1556                       (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1557                         end++;
1558
1559                 off = req->offset + req->offsets[bit] +
1560                         start * req->sectorsize;
1561                 len = (end - start) * req->sectorsize;
1562                 if (desc->buffer != NULL)
1563                         buf = &desc->buffer[start * req->sectorsize];
1564
1565                 switch (req_op(req->req)) {
1566                 case REQ_OP_READ:
1567                         n = 0;
1568                         do {
1569                                 buf = &buf[n];
1570                                 len -= n;
1571                                 n = os_pread_file(req->fds[bit], buf, len, off);
1572                                 if (n < 0) {
1573                                         req->error = map_error(-n);
1574                                         return;
1575                                 }
1576                         } while((n < len) && (n != 0));
1577                         if (n < len) memset(&buf[n], 0, len - n);
1578                         break;
1579                 case REQ_OP_WRITE:
1580                         n = os_pwrite_file(req->fds[bit], buf, len, off);
1581                         if(n != len){
1582                                 req->error = map_error(-n);
1583                                 return;
1584                         }
1585                         break;
1586                 case REQ_OP_DISCARD:
1587                 case REQ_OP_WRITE_ZEROES:
1588                         n = os_falloc_punch(req->fds[bit], off, len);
1589                         if (n) {
1590                                 req->error = map_error(-n);
1591                                 return;
1592                         }
1593                         break;
1594                 default:
1595                         WARN_ON_ONCE(1);
1596                         req->error = BLK_STS_NOTSUPP;
1597                         return;
1598                 }
1599
1600                 start = end;
1601         } while(start < nsectors);
1602
1603         req->offset += len;
1604         req->error = update_bitmap(req, desc);
1605 }
1606
1607 /* Changed in start_io_thread, which is serialized by being called only
1608  * from ubd_init, which is an initcall.
1609  */
1610 int kernel_fd = -1;
1611
1612 /* Only changed by the io thread. XXX: currently unused. */
1613 static int io_count = 0;
1614
1615 int io_thread(void *arg)
1616 {
1617         int n, count, written, res;
1618
1619         os_fix_helper_signals();
1620
1621         while(1){
1622                 n = bulk_req_safe_read(
1623                         kernel_fd,
1624                         io_req_buffer,
1625                         &io_remainder,
1626                         &io_remainder_size,
1627                         UBD_REQ_BUFFER_SIZE
1628                 );
1629                 if (n <= 0) {
1630                         if (n == -EAGAIN)
1631                                 ubd_read_poll(-1);
1632
1633                         continue;
1634                 }
1635
1636                 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1637                         struct io_thread_req *req = (*io_req_buffer)[count];
1638                         int i;
1639
1640                         io_count++;
1641                         for (i = 0; !req->error && i < req->desc_cnt; i++)
1642                                 do_io(req, &(req->io_desc[i]));
1643
1644                 }
1645
1646                 written = 0;
1647
1648                 do {
1649                         res = os_write_file(kernel_fd,
1650                                             ((char *) io_req_buffer) + written,
1651                                             n - written);
1652                         if (res >= 0) {
1653                                 written += res;
1654                         }
1655                         if (written < n) {
1656                                 ubd_write_poll(-1);
1657                         }
1658                 } while (written < n);
1659         }
1660
1661         return 0;
1662 }