root/fs/buffer.c

/* */

DEFINITIONS

1. __wait_on_buffer
2. sync_buffers
3. sync_dev
4. fsync_dev
5. sys_sync
6. file_fsync
7. sys_fsync
8. sys_fdatasync
9. invalidate_buffers
10. remove_from_hash_queue
11. remove_from_lru_list
12. remove_from_free_list
13. remove_from_queues
14. put_last_lru
15. put_last_free
16. insert_into_queues
17. find_buffer
18. get_hash_table
19. set_blocksize
20. refill_freelist
21. getblk
22. set_writetime
23. refile_buffer
24. __brelse
25. __bforget
26. bread
27. breada
28. put_unused_buffer_head
29. get_more_buffer_heads
30. recover_reusable_buffer_heads
31. get_unused_buffer_head
32. create_buffers
33. brw_page
34. mark_buffer_uptodate
35. unlock_buffer
36. generic_readpage
37. grow_buffers
38. try_to_free_buffer
39. age_buffer
40. maybe_shrink_lav_buffers
41. shrink_specific_buffers
42. show_buffers
43. try_to_reassign
44. reassign_cluster
45. try_to_generate_cluster
46. generate_cluster
47. buffer_init
48. wakeup_bdflush
49. sync_old_buffers
50. sys_bdflush
51. bdflush

   1 /*
   2  *  linux/fs/buffer.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6 
   7 /*
   8  *  'buffer.c' implements the buffer-cache functions. Race-conditions have
   9  * been avoided by NEVER letting an interrupt change a buffer (except for the
  10  * data, of course), but instead letting the caller do it.
  11  */
  12 
  13 /*
  14  * NOTE! There is one discordant note here: checking floppies for
  15  * disk change. This is where it fits best, I think, as it should
  16  * invalidate changed floppy-disk-caches.
  17  */
  18  
  19 /* Some bdflush() changes for the dynamic ramdisk - Paul Gortmaker, 12/94 */
  20 
  21 #include <linux/sched.h>
  22 #include <linux/kernel.h>
  23 #include <linux/major.h>
  24 #include <linux/string.h>
  25 #include <linux/locks.h>
  26 #include <linux/errno.h>
  27 #include <linux/malloc.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/swap.h>
  30 #include <linux/swapctl.h>
  31 #include <linux/smp.h>
  32 #include <linux/smp_lock.h>
  33 
  34 #include <asm/system.h>
  35 #include <asm/segment.h>
  36 #include <asm/io.h>
  37 
  38 #define NR_SIZES 5
  39 static char buffersize_index[17] =
  40 {-1,  0,  1, -1,  2, -1, -1, -1, 3, -1, -1, -1, -1, -1, -1, -1, 4};
  41 static short int bufferindex_size[NR_SIZES] = {512, 1024, 2048, 4096, 8192};
  42 
  43 #define BUFSIZE_INDEX(X) ((int) buffersize_index[(X)>>9])
  44 #define MAX_BUF_PER_PAGE (PAGE_SIZE / 512)
  45 
  46 static int grow_buffers(int pri, int size);
  47 static int shrink_specific_buffers(unsigned int priority, int size);
  48 static int maybe_shrink_lav_buffers(int);
  49 
  50 static int nr_hash = 0;  /* Size of hash table */
  51 static struct buffer_head ** hash_table;
  52 struct buffer_head ** buffer_pages;
  53 static struct buffer_head * lru_list[NR_LIST] = {NULL, };
  54 /* next_to_age is an array of pointers into the lru lists, used to
  55    cycle through the buffers aging their contents when deciding which
  56    buffers to discard when more memory is needed */
  57 static struct buffer_head * next_to_age[NR_LIST] = {NULL, };
  58 static struct buffer_head * free_list[NR_SIZES] = {NULL, };
  59 
  60 static struct buffer_head * unused_list = NULL;
  61 struct buffer_head * reuse_list = NULL;
  62 static struct wait_queue * buffer_wait = NULL;
  63 
  64 int nr_buffers = 0;
  65 int nr_buffers_type[NR_LIST] = {0,};
  66 int nr_buffers_size[NR_SIZES] = {0,};
  67 int nr_buffers_st[NR_SIZES][NR_LIST] = {{0,},};
  68 int buffer_usage[NR_SIZES] = {0,};  /* Usage counts used to determine load average */
  69 int buffers_lav[NR_SIZES] = {0,};  /* Load average of buffer usage */
  70 int nr_free[NR_SIZES] = {0,};
  71 int buffermem = 0;
  72 int nr_buffer_heads = 0;
  73 extern int *blksize_size[];
  74 
  75 /* Here is the parameter block for the bdflush process. If you add or
  76  * remove any of the parameters, make sure to update kernel/sysctl.c.
  77  */
  78 
  79 static void wakeup_bdflush(int);
  80 
  81 #define N_PARAM 9
  82 #define LAV
  83 
  84 union bdflush_param{
  85         struct {
  86                 int nfract;  /* Percentage of buffer cache dirty to 
  87                                 activate bdflush */
  88                 int ndirty;  /* Maximum number of dirty blocks to write out per
  89                                 wake-cycle */
  90                 int nrefill; /* Number of clean buffers to try and obtain
  91                                 each time we call refill */
  92                 int nref_dirt; /* Dirty buffer threshold for activating bdflush
  93                                   when trying to refill buffers. */
  94                 int clu_nfract;  /* Percentage of buffer cache to scan to 
  95                                     search for free clusters */
  96                 int age_buffer;  /* Time for normal buffer to age before 
  97                                     we flush it */
  98                 int age_super;  /* Time for superblock to age before we 
  99                                    flush it */
 100                 int lav_const;  /* Constant used for load average (time
 101                                    constant */
 102                 int lav_ratio;  /* Used to determine how low a lav for a
 103                                    particular size can go before we start to
 104                                    trim back the buffers */
 105         } b_un;
 106         unsigned int data[N_PARAM];
 107 } bdf_prm = {{25, 500, 64, 256, 15, 30*HZ, 5*HZ, 1884, 2}};
 108 
 109 /* The lav constant is set for 1 minute, as long as the update process runs
 110    every 5 seconds.  If you change the frequency of update, the time
 111    constant will also change. */
 112 
 113 
 114 /* These are the min and max parameter values that we will allow to be assigned */
 115 int bdflush_min[N_PARAM] = {  0,  10,    5,   60,  0,   100,   100, 1, 1};
 116 int bdflush_max[N_PARAM] = {100,5000, 2000, 2000,100, 60000, 60000, 2047, 5};
 117 
 118 /*
 119  * Rewrote the wait-routines to use the "new" wait-queue functionality,
 120  * and getting rid of the cli-sti pairs. The wait-queue routines still
 121  * need cli-sti, but now it's just a couple of 386 instructions or so.
 122  *
 123  * Note that the real wait_on_buffer() is an inline function that checks
 124  * if 'b_wait' is set before calling this, so that the queues aren't set
 125  * up unnecessarily.
 126  */
 127 void __wait_on_buffer(struct buffer_head * bh)
     /*  */
 128 {
 129         struct wait_queue wait = { current, NULL };
 130 
 131         bh->b_count++;
 132         add_wait_queue(&bh->b_wait, &wait);
 133 repeat:
 134         current->state = TASK_UNINTERRUPTIBLE;
 135         if (buffer_locked(bh)) {
 136                 schedule();
 137                 goto repeat;
 138         }
 139         remove_wait_queue(&bh->b_wait, &wait);
 140         bh->b_count--;
 141         current->state = TASK_RUNNING;
 142 }
 143 
 144 /* Call sync_buffers with wait!=0 to ensure that the call does not
 145    return until all buffer writes have completed.  Sync() may return
 146    before the writes have finished; fsync() may not. */
 147 
 148 
 149 /* Godamity-damn.  Some buffers (bitmaps for filesystems)
 150    spontaneously dirty themselves without ever brelse being called.
 151    We will ultimately want to put these in a separate list, but for
 152    now we search all of the lists for dirty buffers */
 153 
 154 static int sync_buffers(kdev_t dev, int wait)
     /*  */
 155 {
 156         int i, retry, pass = 0, err = 0;
 157         int nlist, ncount;
 158         struct buffer_head * bh, *next;
 159 
 160         /* One pass for no-wait, three for wait:
 161            0) write out all dirty, unlocked buffers;
 162            1) write out all dirty buffers, waiting if locked;
 163            2) wait for completion by waiting for all buffers to unlock. */
 164  repeat:
 165         retry = 0;
 166  repeat2:
 167         ncount = 0;
 168         /* We search all lists as a failsafe mechanism, not because we expect
 169            there to be dirty buffers on any of the other lists. */
 170         for(nlist = 0; nlist < NR_LIST; nlist++)
 171          {
 172          repeat1:
 173                  bh = lru_list[nlist];
 174                  if(!bh) continue;
 175                  for (i = nr_buffers_type[nlist]*2 ; i-- > 0 ; bh = next) {
 176                          if(bh->b_list != nlist) goto repeat1;
 177                          next = bh->b_next_free;
 178                          if(!lru_list[nlist]) break;
 179                          if (dev && bh->b_dev != dev)
 180                                   continue;
 181                          if (buffer_locked(bh))
 182                           {
 183                                   /* Buffer is locked; skip it unless wait is
 184                                      requested AND pass > 0. */
 185                                   if (!wait || !pass) {
 186                                           retry = 1;
 187                                           continue;
 188                                   }
 189                                   wait_on_buffer (bh);
 190                                   goto repeat2;
 191                           }
 192                          /* If an unlocked buffer is not uptodate, there has
 193                              been an IO error. Skip it. */
 194                          if (wait && buffer_req(bh) && !buffer_locked(bh) &&
 195                              !buffer_dirty(bh) && !buffer_uptodate(bh)) {
 196                                   err = 1;
 197                                   continue;
 198                           }
 199                          /* Don't write clean buffers.  Don't write ANY buffers
 200                             on the third pass. */
 201                          if (!buffer_dirty(bh) || pass>=2)
 202                                   continue;
 203                          /* don't bother about locked buffers */
 204                          if (buffer_locked(bh))
 205                                  continue;
 206                          bh->b_count++;
 207                          bh->b_flushtime = 0;
 208                          ll_rw_block(WRITE, 1, &bh);
 209 
 210                          if(nlist != BUF_DIRTY) { 
 211                                  printk("[%d %s %ld] ", nlist,
 212                                         kdevname(bh->b_dev), bh->b_blocknr);
 213                                  ncount++;
 214                          };
 215                          bh->b_count--;
 216                          retry = 1;
 217                  }
 218          }
 219         if (ncount)
 220           printk("sys_sync: %d dirty buffers not on dirty list\n", ncount);
 221         
 222         /* If we are waiting for the sync to succeed, and if any dirty
 223            blocks were written, then repeat; on the second pass, only
 224            wait for buffers being written (do not pass to write any
 225            more buffers on the second pass). */
 226         if (wait && retry && ++pass<=2)
 227                  goto repeat;
 228         return err;
 229 }
 230 
 231 void sync_dev(kdev_t dev)
     /*  */
 232 {
 233         sync_buffers(dev, 0);
 234         sync_supers(dev);
 235         sync_inodes(dev);
 236         sync_buffers(dev, 0);
 237         sync_dquots(dev, -1);
 238 }
 239 
 240 int fsync_dev(kdev_t dev)
     /*  */
 241 {
 242         sync_buffers(dev, 0);
 243         sync_supers(dev);
 244         sync_inodes(dev);
 245         sync_dquots(dev, -1);
 246         return sync_buffers(dev, 1);
 247 }
 248 
 249 asmlinkage int sys_sync(void)
     /*  */
 250 {
 251         fsync_dev(0);
 252         return 0;
 253 }
 254 
 255 int file_fsync (struct inode *inode, struct file *filp)
     /*  */
 256 {
 257         return fsync_dev(inode->i_dev);
 258 }
 259 
 260 asmlinkage int sys_fsync(unsigned int fd)
     /*  */
 261 {
 262         struct file * file;
 263         struct inode * inode;
 264 
 265         if (fd>=NR_OPEN || !(file=current->files->fd[fd]) || !(inode=file->f_inode))
 266                 return -EBADF;
 267         if (!file->f_op || !file->f_op->fsync)
 268                 return -EINVAL;
 269         if (file->f_op->fsync(inode,file))
 270                 return -EIO;
 271         return 0;
 272 }
 273 
 274 asmlinkage int sys_fdatasync(unsigned int fd)
     /*  */
 275 {
 276         struct file * file;
 277         struct inode * inode;
 278 
 279         if (fd>=NR_OPEN || !(file=current->files->fd[fd]) || !(inode=file->f_inode))
 280                 return -EBADF;
 281         if (!file->f_op || !file->f_op->fsync)
 282                 return -EINVAL;
 283         /* this needs further work, at the moment it is identical to fsync() */
 284         if (file->f_op->fsync(inode,file))
 285                 return -EIO;
 286         return 0;
 287 }
 288 
 289 void invalidate_buffers(kdev_t dev)
     /*  */
 290 {
 291         int i;
 292         int nlist;
 293         struct buffer_head * bh;
 294 
 295         for(nlist = 0; nlist < NR_LIST; nlist++) {
 296                 bh = lru_list[nlist];
 297                 for (i = nr_buffers_type[nlist]*2 ; --i > 0 ; bh = bh->b_next_free) {
 298                         if (bh->b_dev != dev)
 299                                 continue;
 300                         wait_on_buffer(bh);
 301                         if (bh->b_dev != dev)
 302                                 continue;
 303                         if (bh->b_count)
 304                                 continue;
 305                         bh->b_flushtime = 0;
 306                         clear_bit(BH_Protected, &bh->b_state);
 307                         clear_bit(BH_Uptodate, &bh->b_state);
 308                         clear_bit(BH_Dirty, &bh->b_state);
 309                         clear_bit(BH_Req, &bh->b_state);
 310                 }
 311         }
 312 }
 313 
 314 #define _hashfn(dev,block) (((unsigned)(HASHDEV(dev)^block))%nr_hash)
 315 #define hash(dev,block) hash_table[_hashfn(dev,block)]
 316 
 317 static inline void remove_from_hash_queue(struct buffer_head * bh)
     /*  */
 318 {
 319         if (bh->b_next)
 320                 bh->b_next->b_prev = bh->b_prev;
 321         if (bh->b_prev)
 322                 bh->b_prev->b_next = bh->b_next;
 323         if (hash(bh->b_dev,bh->b_blocknr) == bh)
 324                 hash(bh->b_dev,bh->b_blocknr) = bh->b_next;
 325         bh->b_next = bh->b_prev = NULL;
 326 }
 327 
 328 static inline void remove_from_lru_list(struct buffer_head * bh)
     /*  */
 329 {
 330         if (!(bh->b_prev_free) || !(bh->b_next_free))
 331                 panic("VFS: LRU block list corrupted");
 332         if (bh->b_dev == B_FREE)
 333                 panic("LRU list corrupted");
 334         bh->b_prev_free->b_next_free = bh->b_next_free;
 335         bh->b_next_free->b_prev_free = bh->b_prev_free;
 336 
 337         if (lru_list[bh->b_list] == bh)
 338                  lru_list[bh->b_list] = bh->b_next_free;
 339         if (lru_list[bh->b_list] == bh)
 340                  lru_list[bh->b_list] = NULL;
 341         if (next_to_age[bh->b_list] == bh)
 342                 next_to_age[bh->b_list] = bh->b_next_free;
 343         if (next_to_age[bh->b_list] == bh)
 344                 next_to_age[bh->b_list] = NULL;
 345 
 346         bh->b_next_free = bh->b_prev_free = NULL;
 347 }
 348 
 349 static inline void remove_from_free_list(struct buffer_head * bh)
     /*  */
 350 {
 351         int isize = BUFSIZE_INDEX(bh->b_size);
 352         if (!(bh->b_prev_free) || !(bh->b_next_free))
 353                 panic("VFS: Free block list corrupted");
 354         if(bh->b_dev != B_FREE)
 355                 panic("Free list corrupted");
 356         if(!free_list[isize])
 357                 panic("Free list empty");
 358         nr_free[isize]--;
 359         if(bh->b_next_free == bh)
 360                  free_list[isize] = NULL;
 361         else {
 362                 bh->b_prev_free->b_next_free = bh->b_next_free;
 363                 bh->b_next_free->b_prev_free = bh->b_prev_free;
 364                 if (free_list[isize] == bh)
 365                          free_list[isize] = bh->b_next_free;
 366         }
 367         bh->b_next_free = bh->b_prev_free = NULL;
 368 }
 369 
 370 static inline void remove_from_queues(struct buffer_head * bh)
     /*  */
 371 {
 372         if(bh->b_dev == B_FREE) {
 373                 remove_from_free_list(bh); /* Free list entries should not be
 374                                               in the hash queue */
 375                 return;
 376         };
 377         nr_buffers_type[bh->b_list]--;
 378         nr_buffers_st[BUFSIZE_INDEX(bh->b_size)][bh->b_list]--;
 379         remove_from_hash_queue(bh);
 380         remove_from_lru_list(bh);
 381 }
 382 
 383 static inline void put_last_lru(struct buffer_head * bh)
     /*  */
 384 {
 385         if (!bh)
 386                 return;
 387         if (bh == lru_list[bh->b_list]) {
 388                 lru_list[bh->b_list] = bh->b_next_free;
 389                 if (next_to_age[bh->b_list] == bh)
 390                         next_to_age[bh->b_list] = bh->b_next_free;
 391                 return;
 392         }
 393         if(bh->b_dev == B_FREE)
 394                 panic("Wrong block for lru list");
 395         remove_from_lru_list(bh);
 396 /* add to back of free list */
 397 
 398         if(!lru_list[bh->b_list]) {
 399                 lru_list[bh->b_list] = bh;
 400                 lru_list[bh->b_list]->b_prev_free = bh;
 401         }
 402         if (!next_to_age[bh->b_list])
 403                 next_to_age[bh->b_list] = bh;
 404 
 405         bh->b_next_free = lru_list[bh->b_list];
 406         bh->b_prev_free = lru_list[bh->b_list]->b_prev_free;
 407         lru_list[bh->b_list]->b_prev_free->b_next_free = bh;
 408         lru_list[bh->b_list]->b_prev_free = bh;
 409 }
 410 
 411 static inline void put_last_free(struct buffer_head * bh)
     /*  */
 412 {
 413         int isize;
 414         if (!bh)
 415                 return;
 416 
 417         isize = BUFSIZE_INDEX(bh->b_size);      
 418         bh->b_dev = B_FREE;  /* So it is obvious we are on the free list */
 419         /* add to back of free list */
 420         if(!free_list[isize]) {
 421                 free_list[isize] = bh;
 422                 bh->b_prev_free = bh;
 423         };
 424 
 425         nr_free[isize]++;
 426         bh->b_next_free = free_list[isize];
 427         bh->b_prev_free = free_list[isize]->b_prev_free;
 428         free_list[isize]->b_prev_free->b_next_free = bh;
 429         free_list[isize]->b_prev_free = bh;
 430 }
 431 
 432 static inline void insert_into_queues(struct buffer_head * bh)
     /*  */
 433 {
 434         /* put at end of free list */
 435         if(bh->b_dev == B_FREE) {
 436                 put_last_free(bh);
 437                 return;
 438         }
 439         if(!lru_list[bh->b_list]) {
 440                 lru_list[bh->b_list] = bh;
 441                 bh->b_prev_free = bh;
 442         }
 443         if (!next_to_age[bh->b_list])
 444                 next_to_age[bh->b_list] = bh;
 445         if (bh->b_next_free) panic("VFS: buffer LRU pointers corrupted");
 446         bh->b_next_free = lru_list[bh->b_list];
 447         bh->b_prev_free = lru_list[bh->b_list]->b_prev_free;
 448         lru_list[bh->b_list]->b_prev_free->b_next_free = bh;
 449         lru_list[bh->b_list]->b_prev_free = bh;
 450         nr_buffers_type[bh->b_list]++;
 451         nr_buffers_st[BUFSIZE_INDEX(bh->b_size)][bh->b_list]++;
 452 /* put the buffer in new hash-queue if it has a device */
 453         bh->b_prev = NULL;
 454         bh->b_next = NULL;
 455         if (!(bh->b_dev))
 456                 return;
 457         bh->b_next = hash(bh->b_dev,bh->b_blocknr);
 458         hash(bh->b_dev,bh->b_blocknr) = bh;
 459         if (bh->b_next)
 460                 bh->b_next->b_prev = bh;
 461 }
 462 
 463 static inline struct buffer_head * find_buffer(kdev_t dev, int block, int size)
     /*  */
 464 {               
 465         struct buffer_head * tmp;
 466 
 467         for (tmp = hash(dev,block) ; tmp != NULL ; tmp = tmp->b_next)
 468                 if (tmp->b_blocknr == block && tmp->b_dev == dev)
 469                         if (tmp->b_size == size)
 470                                 return tmp;
 471                         else {
 472                                 printk("VFS: Wrong blocksize on device %s\n",
 473                                         kdevname(dev));
 474                                 return NULL;
 475                         }
 476         return NULL;
 477 }
 478 
 479 /*
 480  * Why like this, I hear you say... The reason is race-conditions.
 481  * As we don't lock buffers (unless we are reading them, that is),
 482  * something might happen to it while we sleep (ie a read-error
 483  * will force it bad). This shouldn't really happen currently, but
 484  * the code is ready.
 485  */
 486 struct buffer_head * get_hash_table(kdev_t dev, int block, int size)
     /*  */
 487 {
 488         struct buffer_head * bh;
 489 
 490         for (;;) {
 491                 if (!(bh=find_buffer(dev,block,size)))
 492                         return NULL;
 493                 bh->b_count++;
 494                 wait_on_buffer(bh);
 495                 if (bh->b_dev == dev && bh->b_blocknr == block
 496                                              && bh->b_size == size)
 497                         return bh;
 498                 bh->b_count--;
 499         }
 500 }
 501 
 502 void set_blocksize(kdev_t dev, int size)
     /*  */
 503 {
 504         int i, nlist;
 505         struct buffer_head * bh, *bhnext;
 506 
 507         if (!blksize_size[MAJOR(dev)])
 508                 return;
 509 
 510         if (size > PAGE_SIZE)
 511                 size = 0;
 512 
 513         switch (size) {
 514                 default: panic("Invalid blocksize passed to set_blocksize");
 515                 case 512: case 1024: case 2048: case 4096: case 8192: ;
 516         }
 517 
 518         if (blksize_size[MAJOR(dev)][MINOR(dev)] == 0 && size == BLOCK_SIZE) {
 519                 blksize_size[MAJOR(dev)][MINOR(dev)] = size;
 520                 return;
 521         }
 522         if (blksize_size[MAJOR(dev)][MINOR(dev)] == size)
 523                 return;
 524         sync_buffers(dev, 2);
 525         blksize_size[MAJOR(dev)][MINOR(dev)] = size;
 526 
 527   /* We need to be quite careful how we do this - we are moving entries
 528      around on the free list, and we can get in a loop if we are not careful.*/
 529 
 530         for(nlist = 0; nlist < NR_LIST; nlist++) {
 531                 bh = lru_list[nlist];
 532                 for (i = nr_buffers_type[nlist]*2 ; --i > 0 ; bh = bhnext) {
 533                         if(!bh) break;
 534                         bhnext = bh->b_next_free; 
 535                         if (bh->b_dev != dev)
 536                                  continue;
 537                         if (bh->b_size == size)
 538                                  continue;
 539                         
 540                         wait_on_buffer(bh);
 541                         if (bh->b_dev == dev && bh->b_size != size) {
 542                                 clear_bit(BH_Dirty, &bh->b_state);
 543                                 clear_bit(BH_Uptodate, &bh->b_state);
 544                                 clear_bit(BH_Req, &bh->b_state);
 545                                 bh->b_flushtime = 0;
 546                         }
 547                         remove_from_hash_queue(bh);
 548                 }
 549         }
 550 }
 551 
 552 #define BADNESS(bh) (buffer_dirty(bh) || buffer_locked(bh))
 553 
 554 void refill_freelist(int size)
     /*  */
 555 {
 556         struct buffer_head * bh, * tmp;
 557         struct buffer_head * candidate[NR_LIST];
 558         unsigned int best_time, winner;
 559         int isize = BUFSIZE_INDEX(size);
 560         int buffers[NR_LIST];
 561         int i;
 562         int needed;
 563 
 564         /* First see if we even need this.  Sometimes it is advantageous
 565          to request some blocks in a filesystem that we know that we will
 566          be needing ahead of time. */
 567 
 568         if (nr_free[isize] > 100)
 569                 return;
 570 
 571         /* If there are too many dirty buffers, we wake up the update process
 572            now so as to ensure that there are still clean buffers available
 573            for user processes to use (and dirty) */
 574         
 575         /* We are going to try and locate this much memory */
 576         needed =bdf_prm.b_un.nrefill * size;  
 577 
 578         while (nr_free_pages > min_free_pages*2 && needed > 0 &&
 579                grow_buffers(GFP_BUFFER, size)) {
 580                 needed -= PAGE_SIZE;
 581         }
 582 
 583         if(needed <= 0) return;
 584 
 585         /* See if there are too many buffers of a different size.
 586            If so, victimize them */
 587 
 588         while(maybe_shrink_lav_buffers(size))
 589          {
 590                  if(!grow_buffers(GFP_BUFFER, size)) break;
 591                  needed -= PAGE_SIZE;
 592                  if(needed <= 0) return;
 593          };
 594 
 595         /* OK, we cannot grow the buffer cache, now try and get some
 596            from the lru list */
 597 
 598         /* First set the candidate pointers to usable buffers.  This
 599            should be quick nearly all of the time. */
 600 
 601 repeat0:
 602         for(i=0; i<NR_LIST; i++){
 603                 if(i == BUF_DIRTY || i == BUF_SHARED || 
 604                    nr_buffers_type[i] == 0) {
 605                         candidate[i] = NULL;
 606                         buffers[i] = 0;
 607                         continue;
 608                 }
 609                 buffers[i] = nr_buffers_type[i];
 610                 for (bh = lru_list[i]; buffers[i] > 0; bh = tmp, buffers[i]--)
 611                  {
 612                          if(buffers[i] < 0) panic("Here is the problem");
 613                          tmp = bh->b_next_free;
 614                          if (!bh) break;
 615                          
 616                          if (mem_map[MAP_NR((unsigned long) bh->b_data)].count != 1 ||
 617                              buffer_dirty(bh)) {
 618                                  refile_buffer(bh);
 619                                  continue;
 620                          }
 621                          
 622                          if (bh->b_count || buffer_protected(bh) || bh->b_size != size)
 623                                   continue;
 624                          
 625                          /* Buffers are written in the order they are placed 
 626                             on the locked list. If we encounter a locked
 627                             buffer here, this means that the rest of them
 628                             are also locked */
 629                          if (buffer_locked(bh) && (i == BUF_LOCKED || i == BUF_LOCKED1)) {
 630                                  buffers[i] = 0;
 631                                  break;
 632                          }
 633                          
 634                          if (BADNESS(bh)) continue;
 635                          break;
 636                  };
 637                 if(!buffers[i]) candidate[i] = NULL; /* Nothing on this list */
 638                 else candidate[i] = bh;
 639                 if(candidate[i] && candidate[i]->b_count) panic("Here is the problem");
 640         }
 641         
 642  repeat:
 643         if(needed <= 0) return;
 644         
 645         /* Now see which candidate wins the election */
 646         
 647         winner = best_time = UINT_MAX;  
 648         for(i=0; i<NR_LIST; i++){
 649                 if(!candidate[i]) continue;
 650                 if(candidate[i]->b_lru_time < best_time){
 651                         best_time = candidate[i]->b_lru_time;
 652                         winner = i;
 653                 }
 654         }
 655         
 656         /* If we have a winner, use it, and then get a new candidate from that list */
 657         if(winner != UINT_MAX) {
 658                 i = winner;
 659                 bh = candidate[i];
 660                 candidate[i] = bh->b_next_free;
 661                 if(candidate[i] == bh) candidate[i] = NULL;  /* Got last one */
 662                 if (bh->b_count || bh->b_size != size)
 663                          panic("Busy buffer in candidate list\n");
 664                 if (mem_map[MAP_NR((unsigned long) bh->b_data)].count != 1)
 665                          panic("Shared buffer in candidate list\n");
 666                 if (buffer_protected(bh))
 667                         panic("Protected buffer in candidate list\n");
 668                 if (BADNESS(bh)) panic("Buffer in candidate list with BADNESS != 0\n");
 669                 
 670                 if(bh->b_dev == B_FREE)
 671                         panic("Wrong list");
 672                 remove_from_queues(bh);
 673                 bh->b_dev = B_FREE;
 674                 put_last_free(bh);
 675                 needed -= bh->b_size;
 676                 buffers[i]--;
 677                 if(buffers[i] < 0) panic("Here is the problem");
 678                 
 679                 if(buffers[i] == 0) candidate[i] = NULL;
 680                 
 681                 /* Now all we need to do is advance the candidate pointer
 682                    from the winner list to the next usable buffer */
 683                 if(candidate[i] && buffers[i] > 0){
 684                         if(buffers[i] <= 0) panic("Here is another problem");
 685                         for (bh = candidate[i]; buffers[i] > 0; bh = tmp, buffers[i]--) {
 686                                 if(buffers[i] < 0) panic("Here is the problem");
 687                                 tmp = bh->b_next_free;
 688                                 if (!bh) break;
 689                                 
 690                                 if (mem_map[MAP_NR((unsigned long) bh->b_data)].count != 1 ||
 691                                     buffer_dirty(bh)) {
 692                                         refile_buffer(bh);
 693                                         continue;
 694                                 };
 695                                 
 696                                 if (bh->b_count || buffer_protected(bh) || bh->b_size != size)
 697                                          continue;
 698                                 
 699                                 /* Buffers are written in the order they are
 700                                    placed on the locked list.  If we encounter
 701                                    a locked buffer here, this means that the
 702                                    rest of them are also locked */
 703                                 if (buffer_locked(bh) && (i == BUF_LOCKED || i == BUF_LOCKED1)) {
 704                                         buffers[i] = 0;
 705                                         break;
 706                                 }
 707               
 708                                 if (BADNESS(bh)) continue;
 709                                 break;
 710                         };
 711                         if(!buffers[i]) candidate[i] = NULL; /* Nothing here */
 712                         else candidate[i] = bh;
 713                         if(candidate[i] && candidate[i]->b_count) 
 714                                  panic("Here is the problem");
 715                 }
 716                 
 717                 goto repeat;
 718         }
 719         
 720         if(needed <= 0) return;
 721         
 722         /* Too bad, that was not enough. Try a little harder to grow some. */
 723         
 724         if (nr_free_pages > min_free_pages + 5) {
 725                 if (grow_buffers(GFP_BUFFER, size)) {
 726                         needed -= PAGE_SIZE;
 727                         goto repeat0;
 728                 };
 729         }
 730         
 731         /* and repeat until we find something good */
 732         if (!grow_buffers(GFP_ATOMIC, size))
 733                 wakeup_bdflush(1);
 734         needed -= PAGE_SIZE;
 735         goto repeat0;
 736 }
 737 
 738 /*
 739  * Ok, this is getblk, and it isn't very clear, again to hinder
 740  * race-conditions. Most of the code is seldom used, (ie repeating),
 741  * so it should be much more efficient than it looks.
 742  *
 743  * The algorithm is changed: hopefully better, and an elusive bug removed.
 744  *
 745  * 14.02.92: changed it to sync dirty buffers a bit: better performance
 746  * when the filesystem starts to get full of dirty blocks (I hope).
 747  */
 748 struct buffer_head * getblk(kdev_t dev, int block, int size)
     /*  */
 749 {
 750         struct buffer_head * bh;
 751         int isize = BUFSIZE_INDEX(size);
 752 
 753         /* Update this for the buffer size lav. */
 754         buffer_usage[isize]++;
 755 
 756         /* If there are too many dirty buffers, we wake up the update process
 757            now so as to ensure that there are still clean buffers available
 758            for user processes to use (and dirty) */
 759 repeat:
 760         bh = get_hash_table(dev, block, size);
 761         if (bh) {
 762                 if (!buffer_dirty(bh)) {
 763                         if (buffer_uptodate(bh))
 764                                  put_last_lru(bh);
 765                         bh->b_flushtime = 0;
 766                 }
 767                 set_bit(BH_Touched, &bh->b_state);
 768                 return bh;
 769         }
 770 
 771         while(!free_list[isize]) refill_freelist(size);
 772         
 773         if (find_buffer(dev,block,size))
 774                  goto repeat;
 775 
 776         bh = free_list[isize];
 777         remove_from_free_list(bh);
 778 
 779 /* OK, FINALLY we know that this buffer is the only one of its kind, */
 780 /* and that it's unused (b_count=0), unlocked (buffer_locked=0), and clean */
 781         bh->b_count=1;
 782         bh->b_flushtime=0;
 783         bh->b_state=(1<<BH_Touched);
 784         bh->b_dev=dev;
 785         bh->b_blocknr=block;
 786         insert_into_queues(bh);
 787         return bh;
 788 }
 789 
 790 void set_writetime(struct buffer_head * buf, int flag)
     /*  */
 791 {
 792         int newtime;
 793 
 794         if (buffer_dirty(buf)) {
 795                 /* Move buffer to dirty list if jiffies is clear */
 796                 newtime = jiffies + (flag ? bdf_prm.b_un.age_super : 
 797                                      bdf_prm.b_un.age_buffer);
 798                 if(!buf->b_flushtime || buf->b_flushtime > newtime)
 799                          buf->b_flushtime = newtime;
 800         } else {
 801                 buf->b_flushtime = 0;
 802         }
 803 }
 804 
 805 
 806 void refile_buffer(struct buffer_head * buf)
     /*  */
 807 {
 808         int dispose;
 809 
 810         if(buf->b_dev == B_FREE) {
 811                 printk("Attempt to refile free buffer\n");
 812                 return;
 813         }
 814         if (buffer_dirty(buf))
 815                 dispose = BUF_DIRTY;
 816         else if ((mem_map[MAP_NR((unsigned long) buf->b_data)].count > 1) || buffer_protected(buf))
 817                 dispose = BUF_SHARED;
 818         else if (buffer_locked(buf))
 819                 dispose = BUF_LOCKED;
 820         else if (buf->b_list == BUF_SHARED)
 821                 dispose = BUF_UNSHARED;
 822         else
 823                 dispose = BUF_CLEAN;
 824         if(dispose == BUF_CLEAN) buf->b_lru_time = jiffies;
 825         if(dispose != buf->b_list)  {
 826                 if(dispose == BUF_DIRTY || dispose == BUF_UNSHARED)
 827                          buf->b_lru_time = jiffies;
 828                 if(dispose == BUF_LOCKED && 
 829                    (buf->b_flushtime - buf->b_lru_time) <= bdf_prm.b_un.age_super)
 830                          dispose = BUF_LOCKED1;
 831                 remove_from_queues(buf);
 832                 buf->b_list = dispose;
 833                 insert_into_queues(buf);
 834                 if(dispose == BUF_DIRTY && nr_buffers_type[BUF_DIRTY] > 
 835                    (nr_buffers - nr_buffers_type[BUF_SHARED]) *
 836                    bdf_prm.b_un.nfract/100)
 837                          wakeup_bdflush(0);
 838         }
 839 }
 840 
 841 /*
 842  * Release a buffer head
 843  */
 844 void __brelse(struct buffer_head * buf)
     /*  */
 845 {
 846         wait_on_buffer(buf);
 847 
 848         /* If dirty, mark the time this buffer should be written back */
 849         set_writetime(buf, 0);
 850         refile_buffer(buf);
 851 
 852         if (buf->b_count) {
 853                 buf->b_count--;
 854                 return;
 855         }
 856         printk("VFS: brelse: Trying to free free buffer\n");
 857 }
 858 
 859 /*
 860  * bforget() is like brelse(), except it removes the buffer
 861  * from the hash-queues (so that it won't be re-used if it's
 862  * shared).
 863  */
 864 void __bforget(struct buffer_head * buf)
     /*  */
 865 {
 866         wait_on_buffer(buf);
 867         mark_buffer_clean(buf);
 868         clear_bit(BH_Protected, &buf->b_state);
 869         buf->b_count--;
 870         remove_from_hash_queue(buf);
 871         buf->b_dev = NODEV;
 872         refile_buffer(buf);
 873 }
 874 
 875 /*
 876  * bread() reads a specified block and returns the buffer that contains
 877  * it. It returns NULL if the block was unreadable.
 878  */
 879 struct buffer_head * bread(kdev_t dev, int block, int size)
     /*  */
 880 {
 881         struct buffer_head * bh;
 882 
 883         if (!(bh = getblk(dev, block, size))) {
 884                 printk("VFS: bread: READ error on device %s\n",
 885                         kdevname(dev));
 886                 return NULL;
 887         }
 888         if (buffer_uptodate(bh))
 889                 return bh;
 890         ll_rw_block(READ, 1, &bh);
 891         wait_on_buffer(bh);
 892         if (buffer_uptodate(bh))
 893                 return bh;
 894         brelse(bh);
 895         return NULL;
 896 }
 897 
 898 /*
 899  * Ok, breada can be used as bread, but additionally to mark other
 900  * blocks for reading as well. End the argument list with a negative
 901  * number.
 902  */
 903 
 904 #define NBUF 16
 905 
 906 struct buffer_head * breada(kdev_t dev, int block, int bufsize,
     /*  */
 907         unsigned int pos, unsigned int filesize)
 908 {
 909         struct buffer_head * bhlist[NBUF];
 910         unsigned int blocks;
 911         struct buffer_head * bh;
 912         int index;
 913         int i, j;
 914 
 915         if (pos >= filesize)
 916                 return NULL;
 917 
 918         if (block < 0 || !(bh = getblk(dev,block,bufsize)))
 919                 return NULL;
 920 
 921         index = BUFSIZE_INDEX(bh->b_size);
 922 
 923         if (buffer_uptodate(bh))
 924                 return bh;
 925 
 926         blocks = ((filesize & (bufsize - 1)) - (pos & (bufsize - 1))) >> (9+index);
 927 
 928         if (blocks > (read_ahead[MAJOR(dev)] >> index))
 929                 blocks = read_ahead[MAJOR(dev)] >> index;
 930         if (blocks > NBUF)
 931                 blocks = NBUF;
 932         
 933         bhlist[0] = bh;
 934         j = 1;
 935         for(i=1; i<blocks; i++) {
 936                 bh = getblk(dev,block+i,bufsize);
 937                 if (buffer_uptodate(bh)) {
 938                         brelse(bh);
 939                         break;
 940                 }
 941                 bhlist[j++] = bh;
 942         }
 943 
 944         /* Request the read for these buffers, and then release them */
 945         ll_rw_block(READ, j, bhlist);
 946 
 947         for(i=1; i<j; i++)
 948                 brelse(bhlist[i]);
 949 
 950         /* Wait for this buffer, and then continue on */
 951         bh = bhlist[0];
 952         wait_on_buffer(bh);
 953         if (buffer_uptodate(bh))
 954                 return bh;
 955         brelse(bh);
 956         return NULL;
 957 }
 958 
 959 /*
 960  * See fs/inode.c for the weird use of volatile..
 961  */
 962 static void put_unused_buffer_head(struct buffer_head * bh)
     /*  */
 963 {
 964         struct wait_queue * wait;
 965 
 966         wait = ((volatile struct buffer_head *) bh)->b_wait;
 967         memset(bh,0,sizeof(*bh));
 968         ((volatile struct buffer_head *) bh)->b_wait = wait;
 969         bh->b_next_free = unused_list;
 970         unused_list = bh;
 971         wake_up(&buffer_wait);
 972 }
 973 
 974 static void get_more_buffer_heads(void)
     /*  */
 975 {
 976         int i;
 977         struct buffer_head * bh;
 978 
 979         for (;;) {
 980                 if (unused_list)
 981                         return;
 982 
 983                 /*
 984                  * This is critical.  We can't swap out pages to get
 985                  * more buffer heads, because the swap-out may need
 986                  * more buffer-heads itself.  Thus GFP_ATOMIC.
 987                  */
 988                 bh = (struct buffer_head *) get_free_page(GFP_ATOMIC);
 989                 if (bh)
 990                         break;
 991 
 992                 /*
 993                  * Uhhuh. We're _really_ low on memory. Now we just
 994                  * wait for old buffer heads to become free due to
 995                  * finishing IO..
 996                  */
 997                 sleep_on(&buffer_wait);
 998         }
 999 
1000         for (nr_buffer_heads+=i=PAGE_SIZE/sizeof*bh ; i>0; i--) {
1001                 bh->b_next_free = unused_list;  /* only make link */
1002                 unused_list = bh++;
1003         }
1004 }
1005 
1006 /* 
1007  * We can't put completed temporary IO buffer_heads directly onto the
1008  * unused_list when they become unlocked, since the device driver
1009  * end_request routines still expect access to the buffer_head's
1010  * fields after the final unlock.  So, the device driver puts them on
1011  * the reuse_list instead once IO completes, and we recover these to
1012  * the unused_list here.
1013  *
1014  * The reuse_list receives buffers from interrupt routines, so we need
1015  * to be IRQ-safe here.
1016  */
1017 static inline void recover_reusable_buffer_heads(void)
     /*  */
1018 {
1019         struct buffer_head *bh;
1020         unsigned long flags;
1021         
1022         save_flags(flags);
1023         while (reuse_list) {
1024                 cli();
1025                 bh = reuse_list;
1026                 reuse_list = bh->b_next_free;
1027                 restore_flags(flags);
1028                 put_unused_buffer_head(bh);
1029         }
1030 }
1031 
1032 static struct buffer_head * get_unused_buffer_head(void)
     /*  */
1033 {
1034         struct buffer_head * bh;
1035 
1036         recover_reusable_buffer_heads();
1037         get_more_buffer_heads();
1038         if (!unused_list)
1039                 return NULL;
1040         bh = unused_list;
1041         unused_list = bh->b_next_free;
1042         bh->b_next_free = NULL;
1043         bh->b_data = NULL;
1044         bh->b_size = 0;
1045         bh->b_state = 0;
1046         return bh;
1047 }
1048 
1049 /*
1050  * Create the appropriate buffers when given a page for data area and
1051  * the size of each buffer.. Use the bh->b_this_page linked list to
1052  * follow the buffers created.  Return NULL if unable to create more
1053  * buffers.
1054  */
1055 static struct buffer_head * create_buffers(unsigned long page, unsigned long size)
     /*  */
1056 {
1057         struct buffer_head *bh, *head;
1058         unsigned long offset;
1059 
1060         head = NULL;
1061         offset = PAGE_SIZE;
1062         while ((offset -= size) < PAGE_SIZE) {
1063                 bh = get_unused_buffer_head();
1064                 if (!bh)
1065                         goto no_grow;
1066                 bh->b_this_page = head;
1067                 head = bh;
1068                 bh->b_data = (char *) (page+offset);
1069                 bh->b_size = size;
1070                 bh->b_dev = B_FREE;  /* Flag as unused */
1071         }
1072         return head;
1073 /*
1074  * In case anything failed, we just free everything we got.
1075  */
1076 no_grow:
1077         bh = head;
1078         while (bh) {
1079                 head = bh;
1080                 bh = bh->b_this_page;
1081                 put_unused_buffer_head(head);
1082         }
1083         return NULL;
1084 }
1085 
1086 int brw_page(int rw, unsigned long address, kdev_t dev, int b[], int size, int bmap)
     /*  */
1087 {
1088         struct buffer_head *bh, *prev, *next, *arr[MAX_BUF_PER_PAGE];
1089         int block, nr;
1090         struct page *page;
1091 
1092         page = mem_map + MAP_NR(address);
1093         page->uptodate = 0;
1094         bh = create_buffers(address, size);
1095         if (!bh)
1096                 return -ENOMEM;
1097         nr = 0;
1098         next = bh;
1099         do {
1100                 struct buffer_head * tmp;
1101                 block = *(b++);
1102 
1103                 set_bit(BH_FreeOnIO, &next->b_state);
1104                 next->b_list = BUF_CLEAN;
1105                 next->b_dev = dev;
1106                 next->b_blocknr = block;
1107                 next->b_count = 1;
1108                 next->b_flushtime = 0;
1109                 set_bit(BH_Uptodate, &next->b_state);
1110 
1111                 /* When we use bmap, we define block zero to represent
1112                    a hole.  ll_rw_page, however, may legitimately
1113                    access block zero, and we need to distinguish the
1114                    two cases. 
1115                    */
1116                 if (bmap && !block) {
1117                         memset(next->b_data, 0, size);
1118                         next->b_count--;
1119                         continue;
1120                 }
1121                 tmp = get_hash_table(dev, block, size);
1122                 if (tmp) {
1123                         if (!buffer_uptodate(tmp)) {
1124                                 if (rw == READ)
1125                                         ll_rw_block(READ, 1, &tmp);
1126                                 wait_on_buffer(tmp);
1127                         }
1128                         if (rw == READ) 
1129                                 memcpy(next->b_data, tmp->b_data, size);
1130                         else {
1131                                 memcpy(tmp->b_data, next->b_data, size);
1132                                 mark_buffer_dirty(tmp, 0);
1133                         }
1134                         brelse(tmp);
1135                         next->b_count--;
1136                         continue;
1137                 }
1138                 if (rw == READ)
1139                         clear_bit(BH_Uptodate, &next->b_state);
1140                 else
1141                         set_bit(BH_Dirty, &next->b_state);
1142                 arr[nr++] = next;
1143         } while (prev = next, (next = next->b_this_page) != NULL);
1144         prev->b_this_page = bh;
1145         
1146         if (nr)
1147                 ll_rw_block(rw, nr, arr);
1148         else {
1149                 page->locked = 0;
1150                 page->uptodate = 1;
1151                 wake_up(&page->wait);
1152                 next = bh;
1153                 do {
1154                         next->b_next_free = reuse_list;
1155                         reuse_list = next;
1156                         next = next->b_this_page;
1157                 } while (next != bh);
1158         }
1159         ++current->maj_flt;
1160         return 0;
1161 }
1162 
1163 void mark_buffer_uptodate(struct buffer_head * bh, int on)
     /*  */
1164 {
1165         if (on) {
1166                 struct buffer_head *tmp = bh;
1167                 int page_uptodate = 1;
1168                 set_bit(BH_Uptodate, &bh->b_state);
1169                 do {
1170                         if (!test_bit(BH_Uptodate, &tmp->b_state)) {
1171                                 page_uptodate = 0;
1172                                 break;
1173                         }
1174                         tmp=tmp->b_this_page;
1175                 } while (tmp && tmp != bh);
1176                 if (page_uptodate)
1177                         mem_map[MAP_NR(bh->b_data)].uptodate = 1;
1178         } else
1179                 clear_bit(BH_Uptodate, &bh->b_state);
1180 }
1181 
1182 void unlock_buffer(struct buffer_head * bh)
     /*  */
1183 {
1184         struct buffer_head *tmp;
1185         unsigned long flags;
1186         struct page *page;
1187 
1188         clear_bit(BH_Lock, &bh->b_state);
1189         wake_up(&bh->b_wait);
1190 
1191         if (!test_bit(BH_FreeOnIO, &bh->b_state))
1192                 return;
1193         page = mem_map + MAP_NR(bh->b_data);
1194         if (!page->locked) {
1195                 printk ("Whoops: unlock_buffer: "
1196                         "async io complete on unlocked page\n");
1197                 return;
1198         }
1199         if (bh->b_count != 1) {
1200                 printk ("Whoops: unlock_buffer: b_count != 1 on async io.\n");
1201                 return;
1202         }
1203         /* Async buffer_heads are here only as labels for IO, and get
1204            thrown away once the IO for this page is complete.  IO is
1205            deemed complete once all buffers have been visited
1206            (b_count==0) and are now unlocked. */
1207         bh->b_count--;
1208         for (tmp = bh; tmp=tmp->b_this_page, tmp!=bh; ) {
1209                 if (test_bit(BH_Lock, &tmp->b_state) || tmp->b_count)
1210                         return;
1211         }
1212 
1213         /* OK, go ahead and complete the async IO on this page. */
1214         save_flags(flags);
1215         page->locked = 0;
1216         wake_up(&page->wait);
1217         cli();
1218         tmp = bh;
1219         do {
1220                 if (!test_bit(BH_FreeOnIO, &tmp->b_state)) {
1221                         printk ("Whoops: unlock_buffer: "
1222                                 "async IO mismatch on page.\n");
1223                         restore_flags(flags);
1224                         return;
1225                 }
1226                 tmp->b_next_free = reuse_list;
1227                 reuse_list = tmp;
1228                 clear_bit(BH_FreeOnIO, &tmp->b_state);
1229                 tmp = tmp->b_this_page;
1230         } while (tmp != bh);
1231         restore_flags(flags);
1232         if (page->free_after) {
1233                 extern int nr_async_pages;
1234                 nr_async_pages--;
1235                 page->free_after = 0;
1236                 free_page(page_address(page));
1237         }
1238         wake_up(&buffer_wait);
1239 }
1240 
1241 /*
1242  * Generic "readpage" function for block devices that have the normal
1243  * bmap functionality. This is most of the block device filesystems.
1244  * Reads the page asynchronously --- the unlock_buffer() and
1245  * mark_buffer_uptodate() functions propogate buffer state into the
1246  * page struct once IO has completed.
1247  */
1248 int generic_readpage(struct inode * inode, struct page * page)
     /*  */
1249 {
1250         unsigned long block, address;
1251         int *p, nr[PAGE_SIZE/512];
1252         int i;
1253 
1254         address = page_address(page);
1255         page->count++;
1256         page->locked = 1;
1257         
1258         i = PAGE_SIZE >> inode->i_sb->s_blocksize_bits;
1259         block = page->offset >> inode->i_sb->s_blocksize_bits;
1260         p = nr;
1261         do {
1262                 *p = inode->i_op->bmap(inode, block);
1263                 i--;
1264                 block++;
1265                 p++;
1266         } while (i > 0);
1267 
1268         /* IO start */
1269         brw_page(READ, address, inode->i_dev, nr, inode->i_sb->s_blocksize, 1);
1270         free_page(address);
1271         return 0;
1272 }
1273 
1274 /*
1275  * Try to increase the number of buffers available: the size argument
1276  * is used to determine what kind of buffers we want.
1277  */
1278 static int grow_buffers(int pri, int size)
     /*  */
1279 {
1280         unsigned long page;
1281         struct buffer_head *bh, *tmp;
1282         struct buffer_head * insert_point;
1283         int isize;
1284 
1285         if ((size & 511) || (size > PAGE_SIZE)) {
1286                 printk("VFS: grow_buffers: size = %d\n",size);
1287                 return 0;
1288         }
1289 
1290         isize = BUFSIZE_INDEX(size);
1291 
1292         if (!(page = __get_free_page(pri)))
1293                 return 0;
1294         bh = create_buffers(page, size);
1295         if (!bh) {
1296                 free_page(page);
1297                 return 0;
1298         }
1299 
1300         insert_point = free_list[isize];
1301 
1302         tmp = bh;
1303         while (1) {
1304                 nr_free[isize]++;
1305                 if (insert_point) {
1306                         tmp->b_next_free = insert_point->b_next_free;
1307                         tmp->b_prev_free = insert_point;
1308                         insert_point->b_next_free->b_prev_free = tmp;
1309                         insert_point->b_next_free = tmp;
1310                 } else {
1311                         tmp->b_prev_free = tmp;
1312                         tmp->b_next_free = tmp;
1313                 }
1314                 insert_point = tmp;
1315                 ++nr_buffers;
1316                 if (tmp->b_this_page)
1317                         tmp = tmp->b_this_page;
1318                 else
1319                         break;
1320         }
1321         free_list[isize] = bh;
1322         buffer_pages[MAP_NR(page)] = bh;
1323         tmp->b_this_page = bh;
1324         buffermem += PAGE_SIZE;
1325         return 1;
1326 }
1327 
1328 
1329 /* =========== Reduce the buffer memory ============= */
1330 
1331 /*
1332  * try_to_free_buffer() checks if all the buffers on this particular page
1333  * are unused, and free's the page if so.
1334  */
1335 int try_to_free_buffer(struct buffer_head * bh, struct buffer_head ** bhp,
     /*  */
1336                        int priority)
1337 {
1338         unsigned long page;
1339         struct buffer_head * tmp, * p;
1340         int isize = BUFSIZE_INDEX(bh->b_size);
1341 
1342         *bhp = bh;
1343         page = (unsigned long) bh->b_data;
1344         page &= PAGE_MASK;
1345         tmp = bh;
1346         do {
1347                 if (!tmp)
1348                         return 0;
1349                 if (tmp->b_count || buffer_protected(tmp) ||
1350                     buffer_dirty(tmp) || buffer_locked(tmp) || tmp->b_wait)
1351                         return 0;
1352                 if (priority && buffer_touched(tmp))
1353                         return 0;
1354                 tmp = tmp->b_this_page;
1355         } while (tmp != bh);
1356         tmp = bh;
1357         do {
1358                 p = tmp;
1359                 tmp = tmp->b_this_page;
1360                 nr_buffers--;
1361                 nr_buffers_size[isize]--;
1362                 if (p == *bhp)
1363                   {
1364                     *bhp = p->b_prev_free;
1365                     if (p == *bhp) /* Was this the last in the list? */
1366                       *bhp = NULL;
1367                   }
1368                 remove_from_queues(p);
1369                 put_unused_buffer_head(p);
1370         } while (tmp != bh);
1371         buffermem -= PAGE_SIZE;
1372         buffer_pages[MAP_NR(page)] = NULL;
1373         free_page(page);
1374         return !mem_map[MAP_NR(page)].count;
1375 }
1376 
1377 /* Age buffers on a given page, according to whether they have been
1378    visited recently or not. */
1379 static inline void age_buffer(struct buffer_head *bh)
     /*  */
1380 {
1381         struct buffer_head *tmp = bh;
1382         int touched = 0;
1383 
1384         /*
1385          * When we age a page, we mark all other buffers in the page
1386          * with the "has_aged" flag.  Then, when these aliased buffers
1387          * come up for aging, we skip them until next pass.  This
1388          * ensures that a page full of multiple buffers only gets aged
1389          * once per pass through the lru lists. 
1390          */
1391         if (clear_bit(BH_Has_aged, &bh->b_state))
1392                 return;
1393         
1394         do {
1395                 touched |= clear_bit(BH_Touched, &tmp->b_state);
1396                 tmp = tmp->b_this_page;
1397                 set_bit(BH_Has_aged, &tmp->b_state);
1398         } while (tmp != bh);
1399         clear_bit(BH_Has_aged, &bh->b_state);
1400 
1401         if (touched) 
1402                 touch_page(mem_map + MAP_NR((unsigned long) bh->b_data));
1403         else
1404                 age_page(mem_map + MAP_NR((unsigned long) bh->b_data));
1405 }
1406 
1407 /*
1408  * Consult the load average for buffers and decide whether or not
1409  * we should shrink the buffers of one size or not.  If we decide yes,
1410  * do it and return 1.  Else return 0.  Do not attempt to shrink size
1411  * that is specified.
1412  *
1413  * I would prefer not to use a load average, but the way things are now it
1414  * seems unavoidable.  The way to get rid of it would be to force clustering
1415  * universally, so that when we reclaim buffers we always reclaim an entire
1416  * page.  Doing this would mean that we all need to move towards QMAGIC.
1417  */
1418 
1419 static int maybe_shrink_lav_buffers(int size)
     /*  */
1420 {          
1421         int nlist;
1422         int isize;
1423         int total_lav, total_n_buffers, n_sizes;
1424         
1425         /* Do not consider the shared buffers since they would not tend
1426            to have getblk called very often, and this would throw off
1427            the lav.  They are not easily reclaimable anyway (let the swapper
1428            make the first move). */
1429   
1430         total_lav = total_n_buffers = n_sizes = 0;
1431         for(nlist = 0; nlist < NR_SIZES; nlist++)
1432          {
1433                  total_lav += buffers_lav[nlist];
1434                  if(nr_buffers_size[nlist]) n_sizes++;
1435                  total_n_buffers += nr_buffers_size[nlist];
1436                  total_n_buffers -= nr_buffers_st[nlist][BUF_SHARED]; 
1437          }
1438         
1439         /* See if we have an excessive number of buffers of a particular
1440            size - if so, victimize that bunch. */
1441   
1442         isize = (size ? BUFSIZE_INDEX(size) : -1);
1443         
1444         if (n_sizes > 1)
1445                  for(nlist = 0; nlist < NR_SIZES; nlist++)
1446                   {
1447                           if(nlist == isize) continue;
1448                           if(nr_buffers_size[nlist] &&
1449                              bdf_prm.b_un.lav_const * buffers_lav[nlist]*total_n_buffers < 
1450                              total_lav * (nr_buffers_size[nlist] - nr_buffers_st[nlist][BUF_SHARED]))
1451                                    if(shrink_specific_buffers(6, bufferindex_size[nlist])) 
1452                                             return 1;
1453                   }
1454         return 0;
1455 }
1456 
1457 /*
1458  * Try to free up some pages by shrinking the buffer-cache
1459  *
1460  * Priority tells the routine how hard to try to shrink the
1461  * buffers: 6 means "don't bother too much", while a value
1462  * of 0 means "we'd better get some free pages now".
1463  *
1464  * "limit" is meant to limit the shrink-action only to pages
1465  * that are in the 0 - limit address range, for DMA re-allocations.
1466  * We ignore that right now.
1467  */
1468 
1469 static int shrink_specific_buffers(unsigned int priority, int size)
     /*  */
1470 {
1471         struct buffer_head *bh;
1472         int nlist;
1473         int i, isize, isize1;
1474 
1475 #ifdef DEBUG
1476         if(size) printk("Shrinking buffers of size %d\n", size);
1477 #endif
1478         /* First try the free lists, and see if we can get a complete page
1479            from here */
1480         isize1 = (size ? BUFSIZE_INDEX(size) : -1);
1481 
1482         for(isize = 0; isize<NR_SIZES; isize++){
1483                 if(isize1 != -1 && isize1 != isize) continue;
1484                 bh = free_list[isize];
1485                 if(!bh) continue;
1486                 for (i=0 ; !i || bh != free_list[isize]; bh = bh->b_next_free, i++) {
1487                         if (bh->b_count || buffer_protected(bh) ||
1488                             !bh->b_this_page)
1489                                  continue;
1490                         if (!age_of((unsigned long) bh->b_data) &&
1491                             try_to_free_buffer(bh, &bh, 6))
1492                                  return 1;
1493                         if(!bh) break;
1494                         /* Some interrupt must have used it after we
1495                            freed the page.  No big deal - keep looking */
1496                 }
1497         }
1498         
1499         /* Not enough in the free lists, now try the lru list */
1500         
1501         for(nlist = 0; nlist < NR_LIST; nlist++) {
1502         repeat1:
1503                 if(priority > 2 && nlist == BUF_SHARED) continue;
1504                 i = nr_buffers_type[nlist];
1505                 i = ((BUFFEROUT_WEIGHT * i) >> 10) >> priority;
1506                 for ( ; i > 0; i-- ) {
1507                         bh = next_to_age[nlist];
1508                         if (!bh)
1509                                 break;
1510                         next_to_age[nlist] = bh->b_next_free;
1511 
1512                         /* First, age the buffer. */
1513                         age_buffer(bh);
1514                         /* We may have stalled while waiting for I/O
1515                            to complete. */
1516                         if(bh->b_list != nlist) goto repeat1;
1517                         if (bh->b_count || buffer_protected(bh) ||
1518                             !bh->b_this_page)
1519                                  continue;
1520                         if(size && bh->b_size != size) continue;
1521                         if (buffer_locked(bh))
1522                                  if (priority)
1523                                           continue;
1524                                  else
1525                                           wait_on_buffer(bh);
1526                         if (buffer_dirty(bh)) {
1527                                 bh->b_count++;
1528                                 bh->b_flushtime = 0;
1529                                 ll_rw_block(WRITEA, 1, &bh);
1530                                 bh->b_count--;
1531                                 continue;
1532                         }
1533                         /* At priority 6, only consider really old
1534                            (age==0) buffers for reclaiming.  At
1535                            priority 0, consider any buffers. */
1536                         if ((age_of((unsigned long) bh->b_data) >>
1537                              (6-priority)) > 0)
1538                                 continue;                               
1539                         if (try_to_free_buffer(bh, &bh, 0))
1540                                  return 1;
1541                         if(!bh) break;
1542                 }
1543         }
1544         return 0;
1545 }
1546 
1547 
1548 /* ================== Debugging =================== */
1549 
1550 void show_buffers(void)
     /*  */
1551 {
1552         struct buffer_head * bh;
1553         int found = 0, locked = 0, dirty = 0, used = 0, lastused = 0;
1554         int protected = 0;
1555         int shared;
1556         int nlist, isize;
1557 
1558         printk("Buffer memory:   %6dkB\n",buffermem>>10);
1559         printk("Buffer heads:    %6d\n",nr_buffer_heads);
1560         printk("Buffer blocks:   %6d\n",nr_buffers);
1561 
1562         for(nlist = 0; nlist < NR_LIST; nlist++) {
1563           shared = found = locked = dirty = used = lastused = protected = 0;
1564           bh = lru_list[nlist];
1565           if(!bh) continue;
1566           do {
1567                 found++;
1568                 if (buffer_locked(bh))
1569                         locked++;
1570                 if (buffer_protected(bh))
1571                         protected++;
1572                 if (buffer_dirty(bh))
1573                         dirty++;
1574                 if (mem_map[MAP_NR(((unsigned long) bh->b_data))].count != 1)
1575                         shared++;
1576                 if (bh->b_count)
1577                         used++, lastused = found;
1578                 bh = bh->b_next_free;
1579           } while (bh != lru_list[nlist]);
1580           printk("Buffer[%d] mem: %d buffers, %d used (last=%d), "
1581                  "%d locked, %d protected, %d dirty %d shrd\n",
1582                  nlist, found, used, lastused,
1583                  locked, protected, dirty, shared);
1584         };
1585         printk("Size    [LAV]     Free  Clean  Unshar     Lck    Lck1   Dirty  Shared \n");
1586         for(isize = 0; isize<NR_SIZES; isize++){
1587                 printk("%5d [%5d]: %7d ", bufferindex_size[isize],
1588                        buffers_lav[isize], nr_free[isize]);
1589                 for(nlist = 0; nlist < NR_LIST; nlist++)
1590                          printk("%7d ", nr_buffers_st[isize][nlist]);
1591                 printk("\n");
1592         }
1593 }
1594 
1595 
1596 /* ====================== Cluster patches for ext2 ==================== */
1597 
1598 /*
1599  * try_to_reassign() checks if all the buffers on this particular page
1600  * are unused, and reassign to a new cluster them if this is true.
1601  */
1602 static inline int try_to_reassign(struct buffer_head * bh, struct buffer_head ** bhp,
     /*  */
1603                            kdev_t dev, unsigned int starting_block)
1604 {
1605         unsigned long page;
1606         struct buffer_head * tmp, * p;
1607 
1608         *bhp = bh;
1609         page = (unsigned long) bh->b_data;
1610         page &= PAGE_MASK;
1611         if(mem_map[MAP_NR(page)].count != 1) return 0;
1612         tmp = bh;
1613         do {
1614                 if (!tmp)
1615                          return 0;
1616                 
1617                 if (tmp->b_count || buffer_protected(tmp) ||
1618                     buffer_dirty(tmp) || buffer_locked(tmp))
1619                          return 0;
1620                 tmp = tmp->b_this_page;
1621         } while (tmp != bh);
1622         tmp = bh;
1623         
1624         while((unsigned long) tmp->b_data & (PAGE_SIZE - 1)) 
1625                  tmp = tmp->b_this_page;
1626         
1627         /* This is the buffer at the head of the page */
1628         bh = tmp;
1629         do {
1630                 p = tmp;
1631                 tmp = tmp->b_this_page;
1632                 remove_from_queues(p);
1633                 p->b_dev = dev;
1634                 mark_buffer_uptodate(p, 0);
1635                 clear_bit(BH_Req, &p->b_state);
1636                 p->b_blocknr = starting_block++;
1637                 insert_into_queues(p);
1638         } while (tmp != bh);
1639         return 1;
1640 }
1641 
1642 /*
1643  * Try to find a free cluster by locating a page where
1644  * all of the buffers are unused.  We would like this function
1645  * to be atomic, so we do not call anything that might cause
1646  * the process to sleep.  The priority is somewhat similar to
1647  * the priority used in shrink_buffers.
1648  * 
1649  * My thinking is that the kernel should end up using whole
1650  * pages for the buffer cache as much of the time as possible.
1651  * This way the other buffers on a particular page are likely
1652  * to be very near each other on the free list, and we will not
1653  * be expiring data prematurely.  For now we only cannibalize buffers
1654  * of the same size to keep the code simpler.
1655  */
1656 static int reassign_cluster(kdev_t dev, 
     /*  */
1657                      unsigned int starting_block, int size)
1658 {
1659         struct buffer_head *bh;
1660         int isize = BUFSIZE_INDEX(size);
1661         int i;
1662 
1663         /* We want to give ourselves a really good shot at generating
1664            a cluster, and since we only take buffers from the free
1665            list, we "overfill" it a little. */
1666 
1667         while(nr_free[isize] < 32) refill_freelist(size);
1668 
1669         bh = free_list[isize];
1670         if(bh)
1671                  for (i=0 ; !i || bh != free_list[isize] ; bh = bh->b_next_free, i++) {
1672                          if (!bh->b_this_page)  continue;
1673                          if (try_to_reassign(bh, &bh, dev, starting_block))
1674                                  return 4;
1675                  }
1676         return 0;
1677 }
1678 
1679 /* This function tries to generate a new cluster of buffers
1680  * from a new page in memory.  We should only do this if we have
1681  * not expanded the buffer cache to the maximum size that we allow.
1682  */
1683 static unsigned long try_to_generate_cluster(kdev_t dev, int block, int size)
     /*  */
1684 {
1685         struct buffer_head * bh, * tmp, * arr[MAX_BUF_PER_PAGE];
1686         int isize = BUFSIZE_INDEX(size);
1687         unsigned long offset;
1688         unsigned long page;
1689         int nblock;
1690 
1691         page = get_free_page(GFP_NOBUFFER);
1692         if(!page) return 0;
1693 
1694         bh = create_buffers(page, size);
1695         if (!bh) {
1696                 free_page(page);
1697                 return 0;
1698         };
1699         nblock = block;
1700         for (offset = 0 ; offset < PAGE_SIZE ; offset += size) {
1701                 if (find_buffer(dev, nblock++, size))
1702                          goto not_aligned;
1703         }
1704         tmp = bh;
1705         nblock = 0;
1706         while (1) {
1707                 arr[nblock++] = bh;
1708                 bh->b_count = 1;
1709                 bh->b_flushtime = 0;
1710                 bh->b_state = 0;
1711                 bh->b_dev = dev;
1712                 bh->b_list = BUF_CLEAN;
1713                 bh->b_blocknr = block++;
1714                 nr_buffers++;
1715                 nr_buffers_size[isize]++;
1716                 insert_into_queues(bh);
1717                 if (bh->b_this_page)
1718                         bh = bh->b_this_page;
1719                 else
1720                         break;
1721         }
1722         buffermem += PAGE_SIZE;
1723         buffer_pages[MAP_NR(page)] = bh;
1724         bh->b_this_page = tmp;
1725         while (nblock-- > 0)
1726                 brelse(arr[nblock]);
1727         return 4; /* ?? */
1728 not_aligned:
1729         while ((tmp = bh) != NULL) {
1730                 bh = bh->b_this_page;
1731                 put_unused_buffer_head(tmp);
1732         }
1733         free_page(page);
1734         return 0;
1735 }
1736 
1737 unsigned long generate_cluster(kdev_t dev, int b[], int size)
     /*  */
1738 {
1739         int i, offset;
1740         
1741         for (i = 0, offset = 0 ; offset < PAGE_SIZE ; i++, offset += size) {
1742                 if(i && b[i]-1 != b[i-1]) return 0;  /* No need to cluster */
1743                 if(find_buffer(dev, b[i], size)) return 0;
1744         };
1745 
1746         /* OK, we have a candidate for a new cluster */
1747         
1748         /* See if one size of buffer is over-represented in the buffer cache,
1749            if so reduce the numbers of buffers */
1750         if(maybe_shrink_lav_buffers(size))
1751          {
1752                  int retval;
1753                  retval = try_to_generate_cluster(dev, b[0], size);
1754                  if(retval) return retval;
1755          };
1756         
1757         if (nr_free_pages > min_free_pages*2) 
1758                  return try_to_generate_cluster(dev, b[0], size);
1759         else
1760                  return reassign_cluster(dev, b[0], size);
1761 }
1762 
1763 
1764 /* ===================== Init ======================= */
1765 
1766 /*
1767  * This initializes the initial buffer free list.  nr_buffers_type is set
1768  * to one less the actual number of buffers, as a sop to backwards
1769  * compatibility --- the old code did this (I think unintentionally,
1770  * but I'm not sure), and programs in the ps package expect it.
1771  *                                      - TYT 8/30/92
1772  */
1773 void buffer_init(void)
     /*  */
1774 {
1775         int i;
1776         int isize = BUFSIZE_INDEX(BLOCK_SIZE);
1777         long memsize = MAP_NR(high_memory) << PAGE_SHIFT;
1778 
1779         if (memsize >= 4*1024*1024) {
1780                 if(memsize >= 16*1024*1024)
1781                          nr_hash = 16381;
1782                 else
1783                          nr_hash = 4093;
1784         } else {
1785                 nr_hash = 997;
1786         };
1787         
1788         hash_table = (struct buffer_head **) vmalloc(nr_hash * 
1789                                                      sizeof(struct buffer_head *));
1790 
1791 
1792         buffer_pages = (struct buffer_head **) vmalloc(MAP_NR(high_memory) * 
1793                                                      sizeof(struct buffer_head *));
1794         for (i = 0 ; i < MAP_NR(high_memory) ; i++)
1795                 buffer_pages[i] = NULL;
1796 
1797         for (i = 0 ; i < nr_hash ; i++)
1798                 hash_table[i] = NULL;
1799         lru_list[BUF_CLEAN] = 0;
1800         grow_buffers(GFP_KERNEL, BLOCK_SIZE);
1801         if (!free_list[isize])
1802                 panic("VFS: Unable to initialize buffer free list!");
1803         return;
1804 }
1805 
1806 
1807 /* ====================== bdflush support =================== */
1808 
1809 /* This is a simple kernel daemon, whose job it is to provide a dynamic
1810  * response to dirty buffers.  Once this process is activated, we write back
1811  * a limited number of buffers to the disks and then go back to sleep again.
1812  */
1813 struct wait_queue * bdflush_wait = NULL;
1814 struct wait_queue * bdflush_done = NULL;
1815 
1816 static void wakeup_bdflush(int wait)
     /*  */
1817 {
1818         wake_up(&bdflush_wait);
1819         if(wait) sleep_on(&bdflush_done);
1820 }
1821 
1822 
1823 /* 
1824  * Here we attempt to write back old buffers.  We also try and flush inodes 
1825  * and supers as well, since this function is essentially "update", and 
1826  * otherwise there would be no way of ensuring that these quantities ever 
1827  * get written back.  Ideally, we would have a timestamp on the inodes
1828  * and superblocks so that we could write back only the old ones as well
1829  */
1830 
1831 asmlinkage int sync_old_buffers(void)
     /*  */
1832 {
1833         int i, isize;
1834         int ndirty, nwritten;
1835         int nlist;
1836         int ncount;
1837         struct buffer_head * bh, *next;
1838 
1839         sync_supers(0);
1840         sync_inodes(0);
1841 
1842         ncount = 0;
1843 #ifdef DEBUG
1844         for(nlist = 0; nlist < NR_LIST; nlist++)
1845 #else
1846         for(nlist = BUF_DIRTY; nlist <= BUF_DIRTY; nlist++)
1847 #endif
1848         {
1849                 ndirty = 0;
1850                 nwritten = 0;
1851         repeat:
1852                 bh = lru_list[nlist];
1853                 if(bh) 
1854                          for (i = nr_buffers_type[nlist]; i-- > 0; bh = next) {
1855                                  /* We may have stalled while waiting for I/O to complete. */
1856                                  if(bh->b_list != nlist) goto repeat;
1857                                  next = bh->b_next_free;
1858                                  if(!lru_list[nlist]) {
1859                                          printk("Dirty list empty %d\n", i);
1860                                          break;
1861                                  }
1862                                  
1863                                  /* Clean buffer on dirty list?  Refile it */
1864                                  if (nlist == BUF_DIRTY && !buffer_dirty(bh) && !buffer_locked(bh))
1865                                   {
1866                                           refile_buffer(bh);
1867                                           continue;
1868                                   }
1869                                  
1870                                  if (buffer_locked(bh) || !buffer_dirty(bh))
1871                                           continue;
1872                                  ndirty++;
1873                                  if(bh->b_flushtime > jiffies) continue;
1874                                  nwritten++;
1875                                  bh->b_count++;
1876                                  bh->b_flushtime = 0;
1877 #ifdef DEBUG
1878                                  if(nlist != BUF_DIRTY) ncount++;
1879 #endif
1880                                  ll_rw_block(WRITE, 1, &bh);
1881                                  bh->b_count--;
1882                          }
1883         }
1884 #ifdef DEBUG
1885         if (ncount) printk("sync_old_buffers: %d dirty buffers not on dirty list\n", ncount);
1886         printk("Wrote %d/%d buffers\n", nwritten, ndirty);
1887 #endif
1888         
1889         /* We assume that we only come through here on a regular
1890            schedule, like every 5 seconds.  Now update load averages.  
1891            Shift usage counts to prevent overflow. */
1892         for(isize = 0; isize<NR_SIZES; isize++){
1893                 CALC_LOAD(buffers_lav[isize], bdf_prm.b_un.lav_const, buffer_usage[isize]);
1894                 buffer_usage[isize] = 0;
1895         };
1896         return 0;
1897 }
1898 
1899 
1900 /* This is the interface to bdflush.  As we get more sophisticated, we can
1901  * pass tuning parameters to this "process", to adjust how it behaves. 
1902  * We would want to verify each parameter, however, to make sure that it 
1903  * is reasonable. */
1904 
1905 asmlinkage int sys_bdflush(int func, long data)
     /*  */
1906 {
1907         int i, error;
1908 
1909         if (!suser())
1910                 return -EPERM;
1911 
1912         if (func == 1)
1913                  return sync_old_buffers();
1914 
1915         /* Basically func 1 means read param 1, 2 means write param 1, etc */
1916         if (func >= 2) {
1917                 i = (func-2) >> 1;
1918                 if (i < 0 || i >= N_PARAM)
1919                         return -EINVAL;
1920                 if((func & 1) == 0) {
1921                         error = verify_area(VERIFY_WRITE, (void *) data, sizeof(int));
1922                         if (error)
1923                                 return error;
1924                         put_user(bdf_prm.data[i], (int*)data);
1925                         return 0;
1926                 };
1927                 if (data < bdflush_min[i] || data > bdflush_max[i])
1928                         return -EINVAL;
1929                 bdf_prm.data[i] = data;
1930                 return 0;
1931         };
1932 
1933         /* Having func 0 used to launch the actual bdflush and then never
1934         return (unless explicitly killed). We return zero here to 
1935         remain semi-compatible with present update(8) programs. */
1936 
1937         return 0;
1938 }
1939 
1940 /* This is the actual bdflush daemon itself. It used to be started from
1941  * the syscall above, but now we launch it ourselves internally with
1942  * kernel_thread(...)  directly after the first thread in init/main.c */
1943 
1944 int bdflush(void * unused) 
     /*  */
1945 {
1946         int i;
1947         int ndirty;
1948         int nlist;
1949         int ncount;
1950         struct buffer_head * bh, *next;
1951 
1952         /*
1953          *      We have a bare-bones task_struct, and really should fill
1954          *      in a few more things so "top" and /proc/2/{exe,root,cwd}
1955          *      display semi-sane things. Not real crucial though...  
1956          */
1957 
1958         current->session = 1;
1959         current->pgrp = 1;
1960         sprintf(current->comm, "kflushd");
1961 
1962         /*
1963          *      As a kernel thread we want to tamper with system buffers
1964          *      and other internals and thus be subject to the SMP locking
1965          *      rules. (On a uniprocessor box this does nothing).
1966          */
1967          
1968 #ifdef __SMP__
1969         lock_kernel();
1970         syscall_count++;
1971 #endif
1972                  
1973         for (;;) {
1974 #ifdef DEBUG
1975                 printk("bdflush() activated...");
1976 #endif
1977                 
1978                 ncount = 0;
1979 #ifdef DEBUG
1980                 for(nlist = 0; nlist < NR_LIST; nlist++)
1981 #else
1982                 for(nlist = BUF_DIRTY; nlist <= BUF_DIRTY; nlist++)
1983 #endif
1984                  {
1985                          ndirty = 0;
1986                  repeat:
1987                          bh = lru_list[nlist];
1988                          if(bh) 
1989                                   for (i = nr_buffers_type[nlist]; i-- > 0 && ndirty < bdf_prm.b_un.ndirty; 
1990                                        bh = next) {
1991                                           /* We may have stalled while waiting for I/O to complete. */
1992                                           if(bh->b_list != nlist) goto repeat;
1993                                           next = bh->b_next_free;
1994                                           if(!lru_list[nlist]) {
1995                                                   printk("Dirty list empty %d\n", i);
1996                                                   break;
1997                                           }
1998                                           
1999                                           /* Clean buffer on dirty list?  Refile it */
2000                                           if (nlist == BUF_DIRTY && !buffer_dirty(bh) && !buffer_locked(bh))
2001                                            {
2002                                                    refile_buffer(bh);
2003                                                    continue;
2004                                            }
2005                                           
2006                                           if (buffer_locked(bh) || !buffer_dirty(bh))
2007                                                    continue;
2008                                           /* Should we write back buffers that are shared or not??
2009                                              currently dirty buffers are not shared, so it does not matter */
2010                                           bh->b_count++;
2011                                           ndirty++;
2012                                           bh->b_flushtime = 0;
2013                                           ll_rw_block(WRITE, 1, &bh);
2014 #ifdef DEBUG
2015                                           if(nlist != BUF_DIRTY) ncount++;
2016 #endif
2017                                           bh->b_count--;
2018                                   }
2019                  }
2020 #ifdef DEBUG
2021                 if (ncount) printk("sys_bdflush: %d dirty buffers not on dirty list\n", ncount);
2022                 printk("sleeping again.\n");
2023 #endif
2024                 wake_up(&bdflush_done);
2025                 
2026                 /* If there are still a lot of dirty buffers around, skip the sleep
2027                    and flush some more */
2028                 
2029                 if(nr_buffers_type[BUF_DIRTY] <= (nr_buffers - nr_buffers_type[BUF_SHARED]) * 
2030                    bdf_prm.b_un.nfract/100) {
2031                         current->signal = 0;
2032                         interruptible_sleep_on(&bdflush_wait);
2033                 }
2034         }
2035 }
2036 
2037 
2038 /*
2039  * Overrides for Emacs so that we follow Linus's tabbing style.
2040  * Emacs will notice this stuff at the end of the file and automatically
2041  * adjust the settings for this buffer only.  This must remain at the end
2042  * of the file.
2043  * ---------------------------------------------------------------------------
2044  * Local variables:
2045  * c-indent-level: 8
2046  * c-brace-imaginary-offset: 0
2047  * c-brace-offset: -8
2048  * c-argdecl-indent: 8
2049  * c-label-offset: -8
2050  * c-continued-statement-offset: 8
2051  * c-continued-brace-offset: 0
2052  * End:
2053  */

/* */