drivers/block/ll_rw_blk.c

   1 /*
   2  *  linux/drivers/block/ll_rw_blk.c
   3  *
   4  * Copyright (C) 1991, 1992 Linus Torvalds
   5  * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
   6  */
   7
   8 /*
   9  * This handles all read/write requests to block devices
  10  */
  11 #include <linux/sched.h>
  12 #include <linux/kernel.h>
  13 #include <linux/kernel_stat.h>
  14 #include <linux/errno.h>
  15 #include <linux/string.h>
  16 #include <linux/config.h>
  17 #include <linux/locks.h>
  18 #include <linux/mm.h>
  19
  20 #include <asm/system.h>
  21 #include <asm/io.h>
  22 #include <linux/blk.h>
  23
  24 /*
  25  * The request-struct contains all necessary data
  26  * to load a nr of sectors into memory
  27  */
  28 static struct request all_requests[NR_REQUEST];
  29
  30 /*
  31  * used to wait on when there are no free requests
  32  */
  33 struct wait_queue * wait_for_request = NULL;
  34
  35 /* This specifies how many sectors to read ahead on the disk.  */
  36
  37 int read_ahead[MAX_BLKDEV] = {0, };
  38
  39 /* blk_dev_struct is:
  40  *      do_request-address
  41  *      next-request
  42  */
  43 struct blk_dev_struct blk_dev[MAX_BLKDEV] = {
  44         { NULL, NULL },         /* 0 no_dev */
  45         { NULL, NULL },         /* 1 dev mem */
  46         { NULL, NULL },         /* 2 dev fd */
  47         { NULL, NULL },         /* 3 dev ide0 or hd */
  48         { NULL, NULL },         /* 4 dev ttyx */
  49         { NULL, NULL },         /* 5 dev tty */
  50         { NULL, NULL },         /* 6 dev lp */
  51         { NULL, NULL },         /* 7 dev pipes */
  52         { NULL, NULL },         /* 8 dev sd */
  53         { NULL, NULL },         /* 9 dev st */
  54         { NULL, NULL },         /* 10 */
  55         { NULL, NULL },         /* 11 */
  56         { NULL, NULL },         /* 12 */
  57         { NULL, NULL },         /* 13 */
  58         { NULL, NULL },         /* 14 */
  59         { NULL, NULL },         /* 15 */
  60         { NULL, NULL },         /* 16 */
  61         { NULL, NULL },         /* 17 */
  62         { NULL, NULL },         /* 18 */
  63         { NULL, NULL },         /* 19 */
  64         { NULL, NULL },         /* 20 */
  65         { NULL, NULL },         /* 21 */
  66         { NULL, NULL }          /* 22 dev ide1 */
  67 };
  68
  69 /*
  70  * blk_size contains the size of all block-devices in units of 1024 byte
  71  * sectors:
  72  *
  73  * blk_size[MAJOR][MINOR]
  74  *
  75  * if (!blk_size[MAJOR]) then no minor size checking is done.
  76  */
  77 int * blk_size[MAX_BLKDEV] = { NULL, NULL, };
  78
  79 /*
  80  * blksize_size contains the size of all block-devices:
  81  *
  82  * blksize_size[MAJOR][MINOR]
  83  *
  84  * if (!blksize_size[MAJOR]) then 1024 bytes is assumed.
  85  */
  86 int * blksize_size[MAX_BLKDEV] = { NULL, NULL, };
  87
  88 /*
  89  * hardsect_size contains the size of the hardware sector of a device.
  90  *
  91  * hardsect_size[MAJOR][MINOR]
  92  *
  93  * if (!hardsect_size[MAJOR])
  94  *              then 512 bytes is assumed.
  95  * else
  96  *              sector_size is hardsect_size[MAJOR][MINOR]
  97  * This is currently set by some scsi device and read by the msdos fs driver
  98  * This might be a some uses later.
  99  */
 100 int * hardsect_size[MAX_BLKDEV] = { NULL, NULL, };
 101
 102 /*
 103  * "plug" the device if there are no outstanding requests: this will
 104  * force the transfer to start only after we have put all the requests
 105  * on the list.
 106  */
 107 static void plug_device(struct blk_dev_struct * dev, struct request * plug)
 108 {
 109         unsigned long flags;
 110
 111         plug->rq_status = RQ_INACTIVE;
 112         plug->cmd = -1;
 113         plug->next = NULL;
 114         save_flags(flags);
 115         cli();
 116         if (!dev->current_request)
 117                 dev->current_request = plug;
 118         restore_flags(flags);
 119 }
 120
 121 /*
 122  * remove the plug and let it rip..
 123  */
 124 static void unplug_device(struct blk_dev_struct * dev)
 125 {
 126         struct request * req;
 127         unsigned long flags;
 128
 129         save_flags(flags);
 130         cli();
 131         req = dev->current_request;
 132         if (req && req->rq_status == RQ_INACTIVE && req->cmd == -1) {
 133                 dev->current_request = req->next;
 134                 (dev->request_fn)();
 135         }
 136         restore_flags(flags);
 137 }
 138
 139 /*
 140  * look for a free request in the first N entries.
 141  * NOTE: interrupts must be disabled on the way in, and will still
 142  *       be disabled on the way out.
 143  */
 144 static inline struct request * get_request(int n, kdev_t dev)
 145 {
 146         static struct request *prev_found = NULL, *prev_limit = NULL;
 147         register struct request *req, *limit;
 148
 149         if (n <= 0)
 150                 panic("get_request(%d): impossible!\n", n);
 151
 152         limit = all_requests + n;
 153         if (limit != prev_limit) {
 154                 prev_limit = limit;
 155                 prev_found = all_requests;
 156         }
 157         req = prev_found;
 158         for (;;) {
 159                 req = ((req > all_requests) ? req : limit) - 1;
 160                 if (req->rq_status == RQ_INACTIVE)
 161                         break;
 162                 if (req == prev_found)
 163                         return NULL;
 164         }
 165         prev_found = req;
 166         req->rq_status = RQ_ACTIVE;
 167         req->rq_dev = dev;
 168         return req;
 169 }
 170
 171 /*
 172  * wait until a free request in the first N entries is available.
 173  */
 174 static struct request * __get_request_wait(int n, kdev_t dev)
 175 {
 176         register struct request *req;
 177         struct wait_queue wait = { current, NULL };
 178
 179         add_wait_queue(&wait_for_request, &wait);
 180         for (;;) {
 181                 unplug_device(MAJOR(dev)+blk_dev);
 182                 current->state = TASK_UNINTERRUPTIBLE;
 183                 cli();
 184                 req = get_request(n, dev);
 185                 sti();
 186                 if (req)
 187                         break;
 188                 schedule();
 189         }
 190         remove_wait_queue(&wait_for_request, &wait);
 191         current->state = TASK_RUNNING;
 192         return req;
 193 }
 194
 195 static inline struct request * get_request_wait(int n, kdev_t dev)
 196 {
 197         register struct request *req;
 198
 199         cli();
 200         req = get_request(n, dev);
 201         sti();
 202         if (req)
 203                 return req;
 204         return __get_request_wait(n, dev);
 205 }
 206
 207 /* RO fail safe mechanism */
 208
 209 static long ro_bits[MAX_BLKDEV][8];
 210
 211 int is_read_only(kdev_t dev)
 212 {
 213         int minor,major;
 214
 215         major = MAJOR(dev);
 216         minor = MINOR(dev);
 217         if (major < 0 || major >= MAX_BLKDEV) return 0;
 218         return ro_bits[major][minor >> 5] & (1 << (minor & 31));
 219 }
 220
 221 void set_device_ro(kdev_t dev,int flag)
 222 {
 223         int minor,major;
 224
 225         major = MAJOR(dev);
 226         minor = MINOR(dev);
 227         if (major < 0 || major >= MAX_BLKDEV) return;
 228         if (flag) ro_bits[major][minor >> 5] |= 1 << (minor & 31);
 229         else ro_bits[major][minor >> 5] &= ~(1 << (minor & 31));
 230 }
 231
 232 /*
 233  * add-request adds a request to the linked list.
 234  * It disables interrupts so that it can muck with the
 235  * request-lists in peace.
 236  */
 237 static void add_request(struct blk_dev_struct * dev, struct request * req)
 238 {
 239         struct request * tmp;
 240         short            disk_index;
 241
 242         switch (MAJOR(req->rq_dev)) {
 243                 case SCSI_DISK_MAJOR:
 244                         disk_index = (MINOR(req->rq_dev) & 0x0070) >> 4;
 245                         if (disk_index < 4)
 246                                 kstat.dk_drive[disk_index]++;
 247                         break;
 248                 case IDE0_MAJOR:        /* same as HD_MAJOR */
 249                 case XT_DISK_MAJOR:
 250                         disk_index = (MINOR(req->rq_dev) & 0x0040) >> 6;
 251                         kstat.dk_drive[disk_index]++;
 252                         break;
 253                 case IDE1_MAJOR:
 254                         disk_index = ((MINOR(req->rq_dev) & 0x0040) >> 6) + 2;
 255                         kstat.dk_drive[disk_index]++;
 256                 default:
 257                         break;
 258         }
 259
 260         req->next = NULL;
 261         cli();
 262         if (req->bh)
 263                 mark_buffer_clean(req->bh);
 264         if (!(tmp = dev->current_request)) {
 265                 dev->current_request = req;
 266                 (dev->request_fn)();
 267                 sti();
 268                 return;
 269         }
 270         for ( ; tmp->next ; tmp = tmp->next) {
 271                 if ((IN_ORDER(tmp,req) ||
 272                     !IN_ORDER(tmp,tmp->next)) &&
 273                     IN_ORDER(req,tmp->next))
 274                         break;
 275         }
 276         req->next = tmp->next;
 277         tmp->next = req;
 278
 279 /* for SCSI devices, call request_fn unconditionally */
 280         if (scsi_major(MAJOR(req->rq_dev)))
 281                 (dev->request_fn)();
 282
 283         sti();
 284 }
 285
 286 static void make_request(int major,int rw, struct buffer_head * bh)
 287 {
 288         unsigned int sector, count;
 289         struct request * req;
 290         int rw_ahead, max_req;
 291
 292 /* WRITEA/READA is special case - it is not really needed, so if the */
 293 /* buffer is locked, we just forget about it, else it's a normal read */
 294         rw_ahead = (rw == READA || rw == WRITEA);
 295         if (rw_ahead) {
 296                 if (bh->b_lock)
 297                         return;
 298                 if (rw == READA)
 299                         rw = READ;
 300                 else
 301                         rw = WRITE;
 302         }
 303         if (rw!=READ && rw!=WRITE) {
 304                 printk("Bad block dev command, must be R/W/RA/WA\n");
 305                 return;
 306         }
 307         count = bh->b_size >> 9;
 308         sector = bh->b_blocknr * count;
 309         if (blk_size[major])
 310                 if (blk_size[major][MINOR(bh->b_dev)] < (sector + count)>>1) {
 311                         bh->b_dirt = bh->b_uptodate = 0;
 312                         bh->b_req = 0;
 313                         printk("attempt to access beyond end of device\n");
 314                         return;
 315                 }
 316         /* Uhhuh.. Nasty dead-lock possible here.. */
 317         if (bh->b_lock)
 318                 return;
 319         /* Maybe the above fixes it, and maybe it doesn't boot. Life is interesting */
 320         lock_buffer(bh);
 321         if ((rw == WRITE && !bh->b_dirt) || (rw == READ && bh->b_uptodate)) {
 322                 unlock_buffer(bh);
 323                 return;
 324         }
 325
 326 /* we don't allow the write-requests to fill up the queue completely:
 327  * we want some room for reads: they take precedence. The last third
 328  * of the requests are only for reads.
 329  */
 330         max_req = (rw == READ) ? NR_REQUEST : ((NR_REQUEST*2)/3);
 331
 332 /* look for a free request. */
 333         cli();
 334
 335 /* The scsi disk drivers and the IDE driver completely remove the request
 336  * from the queue when they start processing an entry.  For this reason
 337  * it is safe to continue to add links to the top entry for those devices.
 338  */
 339         if ((   major == IDE0_MAJOR     /* same as HD_MAJOR */
 340              || major == IDE1_MAJOR
 341              || major == FLOPPY_MAJOR
 342              || major == SCSI_DISK_MAJOR
 343              || major == SCSI_CDROM_MAJOR
 344              || major == IDE2_MAJOR
 345              || major == IDE3_MAJOR)
 346             && (req = blk_dev[major].current_request))
 347         {
 348 #ifdef CONFIG_BLK_DEV_HD
 349                 if (major == HD_MAJOR || major == FLOPPY_MAJOR)
 350 #else
 351                 if (major == FLOPPY_MAJOR)
 352 #endif CONFIG_BLK_DEV_HD
 353                         req = req->next;
 354                 while (req) {
 355                         if (req->rq_dev == bh->b_dev &&
 356                             !req->sem &&
 357                             req->cmd == rw &&
 358                             req->sector + req->nr_sectors == sector &&
 359                             req->nr_sectors < 244)
 360                         {
 361                                 req->bhtail->b_reqnext = bh;
 362                                 req->bhtail = bh;
 363                                 req->nr_sectors += count;
 364                                 mark_buffer_clean(bh);
 365                                 sti();
 366                                 return;
 367                         }
 368
 369                         if (req->rq_dev == bh->b_dev &&
 370                             !req->sem &&
 371                             req->cmd == rw &&
 372                             req->sector - count == sector &&
 373                             req->nr_sectors < 244)
 374                         {
 375                                 req->nr_sectors += count;
 376                                 bh->b_reqnext = req->bh;
 377                                 req->buffer = bh->b_data;
 378                                 req->current_nr_sectors = count;
 379                                 req->sector = sector;
 380                                 mark_buffer_clean(bh);
 381                                 req->bh = bh;
 382                                 sti();
 383                                 return;
 384                         }
 385
 386                         req = req->next;
 387                 }
 388         }
 389
 390 /* find an unused request. */
 391         req = get_request(max_req, bh->b_dev);
 392         sti();
 393
 394 /* if no request available: if rw_ahead, forget it; otherwise try again blocking.. */
 395         if (!req) {
 396                 if (rw_ahead) {
 397                         unlock_buffer(bh);
 398                         return;
 399                 }
 400                 req = __get_request_wait(max_req, bh->b_dev);
 401         }
 402
 403 /* fill up the request-info, and add it to the queue */
 404         req->cmd = rw;
 405         req->errors = 0;
 406         req->sector = sector;
 407         req->nr_sectors = count;
 408         req->current_nr_sectors = count;
 409         req->buffer = bh->b_data;
 410         req->sem = NULL;
 411         req->bh = bh;
 412         req->bhtail = bh;
 413         req->next = NULL;
 414         add_request(major+blk_dev,req);
 415 }
 416
 417 void ll_rw_page(int rw, kdev_t dev, unsigned long page, char * buffer)
 418 {
 419         struct request * req;
 420         unsigned int major = MAJOR(dev);
 421         unsigned long sector = page * (PAGE_SIZE / 512);
 422         struct semaphore sem = MUTEX_LOCKED;
 423
 424         if (major >= MAX_BLKDEV || !(blk_dev[major].request_fn)) {
 425                 printk("Trying to read nonexistent block-device %s (%ld)\n",
 426                        kdevname(dev), sector);
 427                 return;
 428         }
 429         if (rw!=READ && rw!=WRITE)
 430                 panic("Bad block dev command, must be R/W");
 431         if (rw == WRITE && is_read_only(dev)) {
 432                 printk("Can't page to read-only device %s\n",
 433                        kdevname(dev));
 434                 return;
 435         }
 436         req = get_request_wait(NR_REQUEST, dev);
 437 /* fill up the request-info, and add it to the queue */
 438         req->cmd = rw;
 439         req->errors = 0;
 440         req->sector = sector;
 441         req->nr_sectors = PAGE_SIZE / 512;
 442         req->current_nr_sectors = PAGE_SIZE / 512;
 443         req->buffer = buffer;
 444         req->sem = &sem;
 445         req->bh = NULL;
 446         req->next = NULL;
 447         add_request(major+blk_dev,req);
 448         down(&sem);
 449 }
 450
 451 /* This function can be used to request a number of buffers from a block
 452    device. Currently the only restriction is that all buffers must belong to
 453    the same device */
 454
 455 void ll_rw_block(int rw, int nr, struct buffer_head * bh[])
 456 {
 457         unsigned int major;
 458         struct request plug;
 459         int correct_size;
 460         struct blk_dev_struct * dev;
 461         int i;
 462
 463         /* Make sure that the first block contains something reasonable */
 464         while (!*bh) {
 465                 bh++;
 466                 if (--nr <= 0)
 467                         return;
 468         };
 469
 470         dev = NULL;
 471         if ((major = MAJOR(bh[0]->b_dev)) < MAX_BLKDEV)
 472                 dev = blk_dev + major;
 473         if (!dev || !dev->request_fn) {
 474                 printk(
 475         "ll_rw_block: Trying to read nonexistent block-device %s (%ld)\n",
 476                 kdevname(bh[0]->b_dev), bh[0]->b_blocknr);
 477                 goto sorry;
 478         }
 479
 480         /* Determine correct block size for this device.  */
 481         correct_size = BLOCK_SIZE;
 482         if (blksize_size[major]) {
 483                 i = blksize_size[major][MINOR(bh[0]->b_dev)];
 484                 if (i)
 485                         correct_size = i;
 486         }
 487
 488         /* Verify requested block sizes.  */
 489         for (i = 0; i < nr; i++) {
 490                 if (bh[i] && bh[i]->b_size != correct_size) {
 491                         printk(
 492                         "ll_rw_block: only %d-char blocks implemented (%lu)\n",
 493                                correct_size, bh[i]->b_size);
 494                         goto sorry;
 495                 }
 496         }
 497
 498         if ((rw == WRITE || rw == WRITEA) && is_read_only(bh[0]->b_dev)) {
 499                 printk("Can't write to read-only device %s\n",
 500                        kdevname(bh[0]->b_dev));
 501                 goto sorry;
 502         }
 503
 504         /* If there are no pending requests for this device, then we insert
 505            a dummy request for that device.  This will prevent the request
 506            from starting until we have shoved all of the blocks into the
 507            queue, and then we let it rip.  */
 508
 509         if (nr > 1)
 510                 plug_device(dev, &plug);
 511         for (i = 0; i < nr; i++) {
 512                 if (bh[i]) {
 513                         bh[i]->b_req = 1;
 514                         make_request(major, rw, bh[i]);
 515                         if (rw == READ || rw == READA)
 516                                 kstat.pgpgin++;
 517                         else
 518                                 kstat.pgpgout++;
 519                 }
 520         }
 521         unplug_device(dev);
 522         return;
 523
 524       sorry:
 525         for (i = 0; i < nr; i++) {
 526                 if (bh[i])
 527                         bh[i]->b_dirt = bh[i]->b_uptodate = 0;
 528         }
 529         return;
 530 }
 531
 532 void ll_rw_swap_file(int rw, kdev_t dev, unsigned int *b, int nb, char *buf)
 533 {
 534         int i, j;
 535         int buffersize;
 536         struct request * req[8];
 537         unsigned int major = MAJOR(dev);
 538         struct semaphore sem = MUTEX_LOCKED;
 539
 540         if (major >= MAX_BLKDEV || !(blk_dev[major].request_fn)) {
 541                 printk("ll_rw_swap_file: trying to swap nonexistent block-device\n");
 542                 return;
 543         }
 544
 545         if (rw != READ && rw != WRITE) {
 546                 printk("ll_rw_swap: bad block dev command, must be R/W");
 547                 return;
 548         }
 549         if (rw == WRITE && is_read_only(dev)) {
 550                 printk("Can't swap to read-only device %s\n",
 551                        kdevname(dev));
 552                 return;
 553         }
 554
 555         buffersize = PAGE_SIZE / nb;
 556
 557         for (j=0, i=0; i<nb;)
 558         {
 559                 for (; j < 8 && i < nb; j++, i++, buf += buffersize)
 560                 {
 561                         if (j == 0) {
 562                                 req[j] = get_request_wait(NR_REQUEST, dev);
 563                         } else {
 564                                 cli();
 565                                 req[j] = get_request(NR_REQUEST, dev);
 566                                 sti();
 567                                 if (req[j] == NULL)
 568                                         break;
 569                         }
 570                         req[j]->cmd = rw;
 571                         req[j]->errors = 0;
 572                         req[j]->sector = (b[i] * buffersize) >> 9;
 573                         req[j]->nr_sectors = buffersize >> 9;
 574                         req[j]->current_nr_sectors = buffersize >> 9;
 575                         req[j]->buffer = buf;
 576                         req[j]->sem = &sem;
 577                         req[j]->bh = NULL;
 578                         req[j]->next = NULL;
 579                         add_request(major+blk_dev,req[j]);
 580                 }
 581                 while (j > 0) {
 582                         j--;
 583                         down(&sem);
 584                 }
 585         }
 586 }
 587
 588 int blk_dev_init(void)
 589 {
 590         struct request * req;
 591
 592         req = all_requests + NR_REQUEST;
 593         while (--req >= all_requests) {
 594                 req->rq_status = RQ_INACTIVE;
 595                 req->next = NULL;
 596         }
 597         memset(ro_bits,0,sizeof(ro_bits));
 598 #ifdef CONFIG_BLK_DEV_IDE
 599         ide_init();             /* this MUST preceed hd_init */
 600 #endif
 601 #ifdef CONFIG_BLK_DEV_HD
 602         hd_init();
 603 #endif
 604 #ifdef CONFIG_BLK_DEV_XD
 605         xd_init();
 606 #endif
 607 #ifdef CONFIG_BLK_DEV_FD
 608         floppy_init();
 609 #else
 610         outb_p(0xc, 0x3f2);
 611 #endif
 612 #ifdef CONFIG_CDU31A
 613         cdu31a_init();
 614 #endif CONFIG_CDU31A
 615 #ifdef CONFIG_MCD
 616         mcd_init();
 617 #endif CONFIG_MCD
 618 #ifdef CONFIG_MCDX
 619         mcdx_init();
 620 #endif CONFIG_MCDX
 621 #ifdef CONFIG_SBPCD
 622         sbpcd_init();
 623 #endif CONFIG_SBPCD
 624 #ifdef CONFIG_AZTCD
 625         aztcd_init();
 626 #endif CONFIG_AZTCD
 627 #ifdef CONFIG_CDU535
 628         sony535_init();
 629 #endif CONFIG_CDU535
 630 #ifdef CONFIG_GSCD
 631         gscd_init();
 632 #endif CONFIG_GSCD
 633 #ifdef CONFIG_CM206
 634         cm206_init();
 635 #endif
 636 #ifdef CONFIG_OPTCD
 637         optcd_init();
 638 #endif CONFIG_OPTCD
 639 #ifdef CONFIG_SJCD
 640         sjcd_init();
 641 #endif CONFIG_SJCD
 642         return 0;
 643 }