root/fs/buffer.c

/* [previous][next][first][last][top][bottom][index][help] */

DEFINITIONS

This source file includes following 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. after_unlock_page
  34. free_async_buffers
  35. brw_page
  36. mark_buffer_uptodate
  37. unlock_buffer
  38. generic_readpage
  39. grow_buffers
  40. try_to_free_buffer
  41. age_buffer
  42. maybe_shrink_lav_buffers
  43. shrink_specific_buffers
  44. show_buffers
  45. try_to_reassign
  46. reassign_cluster
  47. try_to_generate_cluster
  48. generate_cluster
  49. buffer_init
  50. wakeup_bdflush
  51. sync_old_buffers
  52. sys_bdflush
  53. 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 static struct buffer_head * lru_list[NR_LIST] = {NULL, };
  53 /* next_to_age is an array of pointers into the lru lists, used to
  54    cycle through the buffers aging their contents when deciding which
  55    buffers to discard when more memory is needed */
  56 static struct buffer_head * next_to_age[NR_LIST] = {NULL, };
  57 static struct buffer_head * free_list[NR_SIZES] = {NULL, };
  58 
  59 static struct buffer_head * unused_list = NULL;
  60 struct buffer_head * reuse_list = NULL;
  61 static struct wait_queue * buffer_wait = NULL;
  62 
  63 int nr_buffers = 0;
  64 int nr_buffers_type[NR_LIST] = {0,};
  65 int nr_buffers_size[NR_SIZES] = {0,};
  66 int nr_buffers_st[NR_SIZES][NR_LIST] = {{0,},};
  67 int buffer_usage[NR_SIZES] = {0,};  /* Usage counts used to determine load average */
  68 int buffers_lav[NR_SIZES] = {0,};  /* Load average of buffer usage */
  69 int nr_free[NR_SIZES] = {0,};
  70 int buffermem = 0;
  71 int nr_buffer_heads = 0;
  72 extern int *blksize_size[];
  73 
  74 /* Here is the parameter block for the bdflush process. If you add or
  75  * remove any of the parameters, make sure to update kernel/sysctl.c.
  76  */
  77 
  78 static void wakeup_bdflush(int);
  79 
  80 #define N_PARAM 9
  81 #define LAV
  82 
  83 union bdflush_param{
  84         struct {
  85                 int nfract;  /* Percentage of buffer cache dirty to 
  86                                 activate bdflush */
  87                 int ndirty;  /* Maximum number of dirty blocks to write out per
  88                                 wake-cycle */
  89                 int nrefill; /* Number of clean buffers to try and obtain
  90                                 each time we call refill */
  91                 int nref_dirt; /* Dirty buffer threshold for activating bdflush
  92                                   when trying to refill buffers. */
  93                 int clu_nfract;  /* Percentage of buffer cache to scan to 
  94                                     search for free clusters */
  95                 int age_buffer;  /* Time for normal buffer to age before 
  96                                     we flush it */
  97                 int age_super;  /* Time for superblock to age before we 
  98                                    flush it */
  99                 int lav_const;  /* Constant used for load average (time
 100                                    constant */
 101                 int lav_ratio;  /* Used to determine how low a lav for a
 102                                    particular size can go before we start to
 103                                    trim back the buffers */
 104         } b_un;
 105         unsigned int data[N_PARAM];
 106 } bdf_prm = {{60, 500, 64, 256, 15, 30*HZ, 5*HZ, 1884, 2}};
 107 
 108 /* The lav constant is set for 1 minute, as long as the update process runs
 109    every 5 seconds.  If you change the frequency of update, the time
 110    constant will also change. */
 111 
 112 
 113 /* These are the min and max parameter values that we will allow to be assigned */
 114 int bdflush_min[N_PARAM] = {  0,  10,    5,   25,  0,   100,   100, 1, 1};
 115 int bdflush_max[N_PARAM] = {100,5000, 2000, 2000,100, 60000, 60000, 2047, 5};
 116 
 117 /*
 118  * Rewrote the wait-routines to use the "new" wait-queue functionality,
 119  * and getting rid of the cli-sti pairs. The wait-queue routines still
 120  * need cli-sti, but now it's just a couple of 386 instructions or so.
 121  *
 122  * Note that the real wait_on_buffer() is an inline function that checks
 123  * if 'b_wait' is set before calling this, so that the queues aren't set
 124  * up unnecessarily.
 125  */
 126 void __wait_on_buffer(struct buffer_head * bh)
     /* [previous][next][first][last][top][bottom][index][help] */
 127 {
 128         struct wait_queue wait = { current, NULL };
 129 
 130         bh->b_count++;
 131         add_wait_queue(&bh->b_wait, &wait);
 132 repeat:
 133         run_task_queue(&tq_disk);
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 250 {
 251         fsync_dev(0);
 252         return 0;
 253 }
 254 
 255 int file_fsync (struct inode *inode, struct file *filp)
     /* [previous][next][first][last][top][bottom][index][help] */
 256 {
 257         return fsync_dev(inode->i_dev);
 258 }
 259 
 260 asmlinkage int sys_fsync(unsigned int fd)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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)
     /* [previous][next][first][last][top][bottom][index][help] */
 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 /*
 807  * A buffer may need to be moved from one buffer list to another
 808  * (e.g. in case it is not shared any more). Handle this.
 809  */
 810 void refile_buffer(struct buffer_head * buf)
     /* [previous][next][first][last][top][bottom][index][help] */
 811 {
 812         int dispose;
 813 
 814         if(buf->b_dev == B_FREE) {
 815                 printk("Attempt to refile free buffer\n");
 816                 return;
 817         }
 818         if (buffer_dirty(buf))
 819                 dispose = BUF_DIRTY;
 820         else if ((mem_map[MAP_NR((unsigned long) buf->b_data)].count > 1) || buffer_protected(buf))
 821                 dispose = BUF_SHARED;
 822         else if (buffer_locked(buf))
 823                 dispose = BUF_LOCKED;
 824         else if (buf->b_list == BUF_SHARED)
 825                 dispose = BUF_UNSHARED;
 826         else
 827                 dispose = BUF_CLEAN;
 828         if(dispose == BUF_CLEAN) buf->b_lru_time = jiffies;
 829         if(dispose != buf->b_list)  {
 830                 if(dispose == BUF_DIRTY || dispose == BUF_UNSHARED)
 831                          buf->b_lru_time = jiffies;
 832                 if(dispose == BUF_LOCKED && 
 833                    (buf->b_flushtime - buf->b_lru_time) <= bdf_prm.b_un.age_super)
 834                          dispose = BUF_LOCKED1;
 835                 remove_from_queues(buf);
 836                 buf->b_list = dispose;
 837                 insert_into_queues(buf);
 838                 if(dispose == BUF_DIRTY && nr_buffers_type[BUF_DIRTY] > 
 839                    (nr_buffers - nr_buffers_type[BUF_SHARED]) *
 840                    bdf_prm.b_un.nfract/100)
 841                          wakeup_bdflush(0);
 842         }
 843 }
 844 
 845 /*
 846  * Release a buffer head
 847  */
 848 void __brelse(struct buffer_head * buf)
     /* [previous][next][first][last][top][bottom][index][help] */
 849 {
 850         wait_on_buffer(buf);
 851 
 852         /* If dirty, mark the time this buffer should be written back */
 853         set_writetime(buf, 0);
 854         refile_buffer(buf);
 855 
 856         if (buf->b_count) {
 857                 buf->b_count--;
 858                 return;
 859         }
 860         printk("VFS: brelse: Trying to free free buffer\n");
 861 }
 862 
 863 /*
 864  * bforget() is like brelse(), except it removes the buffer
 865  * from the hash-queues (so that it won't be re-used if it's
 866  * shared).
 867  */
 868 void __bforget(struct buffer_head * buf)
     /* [previous][next][first][last][top][bottom][index][help] */
 869 {
 870         wait_on_buffer(buf);
 871         mark_buffer_clean(buf);
 872         clear_bit(BH_Protected, &buf->b_state);
 873         buf->b_count--;
 874         remove_from_hash_queue(buf);
 875         buf->b_dev = NODEV;
 876         refile_buffer(buf);
 877 }
 878 
 879 /*
 880  * bread() reads a specified block and returns the buffer that contains
 881  * it. It returns NULL if the block was unreadable.
 882  */
 883 struct buffer_head * bread(kdev_t dev, int block, int size)
     /* [previous][next][first][last][top][bottom][index][help] */
 884 {
 885         struct buffer_head * bh;
 886 
 887         if (!(bh = getblk(dev, block, size))) {
 888                 printk("VFS: bread: READ error on device %s\n",
 889                         kdevname(dev));
 890                 return NULL;
 891         }
 892         if (buffer_uptodate(bh))
 893                 return bh;
 894         ll_rw_block(READ, 1, &bh);
 895         wait_on_buffer(bh);
 896         if (buffer_uptodate(bh))
 897                 return bh;
 898         brelse(bh);
 899         return NULL;
 900 }
 901 
 902 /*
 903  * Ok, breada can be used as bread, but additionally to mark other
 904  * blocks for reading as well. End the argument list with a negative
 905  * number.
 906  */
 907 
 908 #define NBUF 16
 909 
 910 struct buffer_head * breada(kdev_t dev, int block, int bufsize,
     /* [previous][next][first][last][top][bottom][index][help] */
 911         unsigned int pos, unsigned int filesize)
 912 {
 913         struct buffer_head * bhlist[NBUF];
 914         unsigned int blocks;
 915         struct buffer_head * bh;
 916         int index;
 917         int i, j;
 918 
 919         if (pos >= filesize)
 920                 return NULL;
 921 
 922         if (block < 0 || !(bh = getblk(dev,block,bufsize)))
 923                 return NULL;
 924 
 925         index = BUFSIZE_INDEX(bh->b_size);
 926 
 927         if (buffer_uptodate(bh))
 928                 return bh;
 929 
 930         blocks = ((filesize & (bufsize - 1)) - (pos & (bufsize - 1))) >> (9+index);
 931 
 932         if (blocks > (read_ahead[MAJOR(dev)] >> index))
 933                 blocks = read_ahead[MAJOR(dev)] >> index;
 934         if (blocks > NBUF)
 935                 blocks = NBUF;
 936         
 937         bhlist[0] = bh;
 938         j = 1;
 939         for(i=1; i<blocks; i++) {
 940                 bh = getblk(dev,block+i,bufsize);
 941                 if (buffer_uptodate(bh)) {
 942                         brelse(bh);
 943                         break;
 944                 }
 945                 bhlist[j++] = bh;
 946         }
 947 
 948         /* Request the read for these buffers, and then release them */
 949         ll_rw_block(READ, j, bhlist);
 950 
 951         for(i=1; i<j; i++)
 952                 brelse(bhlist[i]);
 953 
 954         /* Wait for this buffer, and then continue on */
 955         bh = bhlist[0];
 956         wait_on_buffer(bh);
 957         if (buffer_uptodate(bh))
 958                 return bh;
 959         brelse(bh);
 960         return NULL;
 961 }
 962 
 963 /*
 964  * See fs/inode.c for the weird use of volatile..
 965  */
 966 static void put_unused_buffer_head(struct buffer_head * bh)
     /* [previous][next][first][last][top][bottom][index][help] */
 967 {
 968         struct wait_queue * wait;
 969 
 970         wait = ((volatile struct buffer_head *) bh)->b_wait;
 971         memset(bh,0,sizeof(*bh));
 972         ((volatile struct buffer_head *) bh)->b_wait = wait;
 973         bh->b_next_free = unused_list;
 974         unused_list = bh;
 975         wake_up(&buffer_wait);
 976 }
 977 
 978 static void get_more_buffer_heads(void)
     /* [previous][next][first][last][top][bottom][index][help] */
 979 {
 980         int i;
 981         struct buffer_head * bh;
 982 
 983         for (;;) {
 984                 if (unused_list)
 985                         return;
 986 
 987                 /*
 988                  * This is critical.  We can't swap out pages to get
 989                  * more buffer heads, because the swap-out may need
 990                  * more buffer-heads itself.  Thus GFP_ATOMIC.
 991                  */
 992                 bh = (struct buffer_head *) get_free_page(GFP_ATOMIC);
 993                 if (bh)
 994                         break;
 995 
 996                 /*
 997                  * Uhhuh. We're _really_ low on memory. Now we just
 998                  * wait for old buffer heads to become free due to
 999                  * finishing IO..
1000                  */
1001                 run_task_queue(&tq_disk);
1002                 sleep_on(&buffer_wait);
1003         }
1004 
1005         for (nr_buffer_heads+=i=PAGE_SIZE/sizeof*bh ; i>0; i--) {
1006                 bh->b_next_free = unused_list;  /* only make link */
1007                 unused_list = bh++;
1008         }
1009 }
1010 
1011 /* 
1012  * We can't put completed temporary IO buffer_heads directly onto the
1013  * unused_list when they become unlocked, since the device driver
1014  * end_request routines still expect access to the buffer_head's
1015  * fields after the final unlock.  So, the device driver puts them on
1016  * the reuse_list instead once IO completes, and we recover these to
1017  * the unused_list here.
1018  *
1019  * The reuse_list receives buffers from interrupt routines, so we need
1020  * to be IRQ-safe here (but note that interrupts only _add_ to the
1021  * reuse_list, never take away. So we don't need to worry about the
1022  * reuse_list magically emptying).
1023  */
1024 static inline void recover_reusable_buffer_heads(void)
     /* [previous][next][first][last][top][bottom][index][help] */
1025 {
1026         if (reuse_list) {
1027                 struct buffer_head *bh;
1028                 unsigned long flags;
1029         
1030                 save_flags(flags);
1031                 do {
1032                         cli();
1033                         bh = reuse_list;
1034                         reuse_list = bh->b_next_free;
1035                         restore_flags(flags);
1036                         put_unused_buffer_head(bh);
1037                 } while (reuse_list);
1038         }
1039 }
1040 
1041 static struct buffer_head * get_unused_buffer_head(void)
     /* [previous][next][first][last][top][bottom][index][help] */
1042 {
1043         struct buffer_head * bh;
1044 
1045         recover_reusable_buffer_heads();
1046         get_more_buffer_heads();
1047         if (!unused_list)
1048                 return NULL;
1049         bh = unused_list;
1050         unused_list = bh->b_next_free;
1051         bh->b_next_free = NULL;
1052         bh->b_data = NULL;
1053         bh->b_size = 0;
1054         bh->b_state = 0;
1055         return bh;
1056 }
1057 
1058 /*
1059  * Create the appropriate buffers when given a page for data area and
1060  * the size of each buffer.. Use the bh->b_this_page linked list to
1061  * follow the buffers created.  Return NULL if unable to create more
1062  * buffers.
1063  */
1064 static struct buffer_head * create_buffers(unsigned long page, unsigned long size)
     /* [previous][next][first][last][top][bottom][index][help] */
1065 {
1066         struct buffer_head *bh, *head;
1067         unsigned long offset;
1068 
1069         head = NULL;
1070         offset = PAGE_SIZE;
1071         while ((offset -= size) < PAGE_SIZE) {
1072                 bh = get_unused_buffer_head();
1073                 if (!bh)
1074                         goto no_grow;
1075                 bh->b_this_page = head;
1076                 head = bh;
1077                 bh->b_data = (char *) (page+offset);
1078                 bh->b_size = size;
1079                 bh->b_dev = B_FREE;  /* Flag as unused */
1080         }
1081         return head;
1082 /*
1083  * In case anything failed, we just free everything we got.
1084  */
1085 no_grow:
1086         bh = head;
1087         while (bh) {
1088                 head = bh;
1089                 bh = bh->b_this_page;
1090                 put_unused_buffer_head(head);
1091         }
1092         return NULL;
1093 }
1094 
1095 /* Run the hooks that have to be done when a page I/O has completed. */
1096 static inline void after_unlock_page (struct page * page)
     /* [previous][next][first][last][top][bottom][index][help] */
1097 {
1098         if (clear_bit(PG_decr_after, &page->flags))
1099                 nr_async_pages--;
1100         if (clear_bit(PG_free_after, &page->flags))
1101                 free_page(page_address(page));
1102         if (clear_bit(PG_swap_unlock_after, &page->flags))
1103                 swap_after_unlock_page(page->swap_unlock_entry);
1104 }
1105 
1106 /* Free all temporary buffers belonging to a page. */
1107 static inline void free_async_buffers (struct buffer_head * bh)
     /* [previous][next][first][last][top][bottom][index][help] */
1108 {
1109         struct buffer_head * tmp;
1110         unsigned long flags;
1111 
1112         tmp = bh;
1113         save_flags(flags);
1114         cli();
1115         do {
1116                 if (!test_bit(BH_FreeOnIO, &tmp->b_state)) {
1117                         printk ("Whoops: unlock_buffer: "
1118                                 "async IO mismatch on page.\n");
1119                         restore_flags(flags);
1120                         return;
1121                 }
1122                 tmp->b_next_free = reuse_list;
1123                 reuse_list = tmp;
1124                 clear_bit(BH_FreeOnIO, &tmp->b_state);
1125                 tmp = tmp->b_this_page;
1126         } while (tmp != bh);
1127         restore_flags(flags);
1128 }
1129 
1130 /*
1131  * Start I/O on a page.
1132  * This function expects the page to be locked and may return before I/O is complete.
1133  * You then have to check page->locked, page->uptodate, and maybe wait on page->wait.
1134  */
1135 int brw_page(int rw, unsigned long address, kdev_t dev, int b[], int size, int bmap)
     /* [previous][next][first][last][top][bottom][index][help] */
1136 {
1137         struct buffer_head *bh, *prev, *next, *arr[MAX_BUF_PER_PAGE];
1138         int block, nr;
1139         struct page *page;
1140 
1141         page = mem_map + MAP_NR(address);
1142         if (!PageLocked(page))
1143                 panic("brw_page: page not locked for I/O");
1144         clear_bit(PG_uptodate, &page->flags);
1145         /*
1146          * Allocate buffer heads pointing to this page, just for I/O.
1147          * They do _not_ show up in the buffer hash table!
1148          * They are _not_ registered in page->buffers either!
1149          */
1150         bh = create_buffers(address, size);
1151         if (!bh) {
1152                 clear_bit(PG_locked, &page->flags);
1153                 wake_up(&page->wait);
1154                 return -ENOMEM;
1155         }
1156         nr = 0;
1157         next = bh;
1158         do {
1159                 struct buffer_head * tmp;
1160                 block = *(b++);
1161 
1162                 set_bit(BH_FreeOnIO, &next->b_state);
1163                 next->b_list = BUF_CLEAN;
1164                 next->b_dev = dev;
1165                 next->b_blocknr = block;
1166                 next->b_count = 1;
1167                 next->b_flushtime = 0;
1168                 set_bit(BH_Uptodate, &next->b_state);
1169 
1170                 /* When we use bmap, we define block zero to represent
1171                    a hole.  ll_rw_page, however, may legitimately
1172                    access block zero, and we need to distinguish the
1173                    two cases. 
1174                    */
1175                 if (bmap && !block) {
1176                         memset(next->b_data, 0, size);
1177                         next->b_count--;
1178                         continue;
1179                 }
1180                 tmp = get_hash_table(dev, block, size);
1181                 if (tmp) {
1182                         if (!buffer_uptodate(tmp)) {
1183                                 if (rw == READ)
1184                                         ll_rw_block(READ, 1, &tmp);
1185                                 wait_on_buffer(tmp);
1186                         }
1187                         if (rw == READ) 
1188                                 memcpy(next->b_data, tmp->b_data, size);
1189                         else {
1190                                 memcpy(tmp->b_data, next->b_data, size);
1191                                 mark_buffer_dirty(tmp, 0);
1192                         }
1193                         brelse(tmp);
1194                         next->b_count--;
1195                         continue;
1196                 }
1197                 if (rw == READ)
1198                         clear_bit(BH_Uptodate, &next->b_state);
1199                 else
1200                         set_bit(BH_Dirty, &next->b_state);
1201                 arr[nr++] = next;
1202         } while (prev = next, (next = next->b_this_page) != NULL);
1203         prev->b_this_page = bh;
1204         
1205         if (nr) {
1206                 ll_rw_block(rw, nr, arr);
1207                 /* The rest of the work is done in mark_buffer_uptodate()
1208                  * and unlock_buffer(). */
1209         } else {
1210                 clear_bit(PG_locked, &page->flags);
1211                 set_bit(PG_uptodate, &page->flags);
1212                 wake_up(&page->wait);
1213                 free_async_buffers(bh);
1214                 after_unlock_page(page);
1215         }
1216         ++current->maj_flt;
1217         return 0;
1218 }
1219 
1220 /*
1221  * This is called by end_request() when I/O has completed.
1222  */
1223 void mark_buffer_uptodate(struct buffer_head * bh, int on)
     /* [previous][next][first][last][top][bottom][index][help] */
1224 {
1225         if (on) {
1226                 struct buffer_head *tmp = bh;
1227                 int page_uptodate = 1;
1228                 set_bit(BH_Uptodate, &bh->b_state);
1229                 /* If a page has buffers and all these buffers are uptodate,
1230                  * then the page is uptodate. */
1231                 do {
1232                         if (!test_bit(BH_Uptodate, &tmp->b_state)) {
1233                                 page_uptodate = 0;
1234                                 break;
1235                         }
1236                         tmp=tmp->b_this_page;
1237                 } while (tmp && tmp != bh);
1238                 if (page_uptodate)
1239                         set_bit(PG_uptodate, &mem_map[MAP_NR(bh->b_data)].flags);
1240         } else
1241                 clear_bit(BH_Uptodate, &bh->b_state);
1242 }
1243 
1244 /*
1245  * This is called by end_request() when I/O has completed.
1246  */
1247 void unlock_buffer(struct buffer_head * bh)
     /* [previous][next][first][last][top][bottom][index][help] */
1248 {
1249         struct buffer_head *tmp;
1250         struct page *page;
1251 
1252         clear_bit(BH_Lock, &bh->b_state);
1253         wake_up(&bh->b_wait);
1254 
1255         if (!test_bit(BH_FreeOnIO, &bh->b_state))
1256                 return;
1257         /* This is a temporary buffer used for page I/O. */
1258         page = mem_map + MAP_NR(bh->b_data);
1259         if (!PageLocked(page)) {
1260                 printk ("Whoops: unlock_buffer: "
1261                         "async io complete on unlocked page\n");
1262                 return;
1263         }
1264         if (bh->b_count != 1) {
1265                 printk ("Whoops: unlock_buffer: b_count != 1 on async io.\n");
1266                 return;
1267         }
1268         /* Async buffer_heads are here only as labels for IO, and get
1269            thrown away once the IO for this page is complete.  IO is
1270            deemed complete once all buffers have been visited
1271            (b_count==0) and are now unlocked. */
1272         bh->b_count--;
1273         for (tmp = bh; tmp=tmp->b_this_page, tmp!=bh; ) {
1274                 if (test_bit(BH_Lock, &tmp->b_state) || tmp->b_count)
1275                         return;
1276         }
1277         /* OK, the async IO on this page is complete. */
1278         clear_bit(PG_locked, &page->flags);
1279         wake_up(&page->wait);
1280         free_async_buffers(bh);
1281         after_unlock_page(page);
1282         wake_up(&buffer_wait);
1283 }
1284 
1285 /*
1286  * Generic "readpage" function for block devices that have the normal
1287  * bmap functionality. This is most of the block device filesystems.
1288  * Reads the page asynchronously --- the unlock_buffer() and
1289  * mark_buffer_uptodate() functions propagate buffer state into the
1290  * page struct once IO has completed.
1291  */
1292 int generic_readpage(struct inode * inode, struct page * page)
     /* [previous][next][first][last][top][bottom][index][help] */
1293 {
1294         unsigned long block, address;
1295         int *p, nr[PAGE_SIZE/512];
1296         int i;
1297 
1298         address = page_address(page);
1299         page->count++;
1300         set_bit(PG_locked, &page->flags);
1301         set_bit(PG_free_after, &page->flags);
1302         
1303         i = PAGE_SIZE >> inode->i_sb->s_blocksize_bits;
1304         block = page->offset >> inode->i_sb->s_blocksize_bits;
1305         p = nr;
1306         do {
1307                 *p = inode->i_op->bmap(inode, block);
1308                 i--;
1309                 block++;
1310                 p++;
1311         } while (i > 0);
1312 
1313         /* IO start */
1314         brw_page(READ, address, inode->i_dev, nr, inode->i_sb->s_blocksize, 1);
1315         return 0;
1316 }
1317 
1318 /*
1319  * Try to increase the number of buffers available: the size argument
1320  * is used to determine what kind of buffers we want.
1321  */
1322 static int grow_buffers(int pri, int size)
     /* [previous][next][first][last][top][bottom][index][help] */
1323 {
1324         unsigned long page;
1325         struct buffer_head *bh, *tmp;
1326         struct buffer_head * insert_point;
1327         int isize;
1328 
1329         if ((size & 511) || (size > PAGE_SIZE)) {
1330                 printk("VFS: grow_buffers: size = %d\n",size);
1331                 return 0;
1332         }
1333 
1334         isize = BUFSIZE_INDEX(size);
1335 
1336         if (!(page = __get_free_page(pri)))
1337                 return 0;
1338         bh = create_buffers(page, size);
1339         if (!bh) {
1340                 free_page(page);
1341                 return 0;
1342         }
1343 
1344         insert_point = free_list[isize];
1345 
1346         tmp = bh;
1347         while (1) {
1348                 nr_free[isize]++;
1349                 if (insert_point) {
1350                         tmp->b_next_free = insert_point->b_next_free;
1351                         tmp->b_prev_free = insert_point;
1352                         insert_point->b_next_free->b_prev_free = tmp;
1353                         insert_point->b_next_free = tmp;
1354                 } else {
1355                         tmp->b_prev_free = tmp;
1356                         tmp->b_next_free = tmp;
1357                 }
1358                 insert_point = tmp;
1359                 ++nr_buffers;
1360                 if (tmp->b_this_page)
1361                         tmp = tmp->b_this_page;
1362                 else
1363                         break;
1364         }
1365         free_list[isize] = bh;
1366         mem_map[MAP_NR(page)].buffers = bh;
1367         tmp->b_this_page = bh;
1368         buffermem += PAGE_SIZE;
1369         return 1;
1370 }
1371 
1372 
1373 /* =========== Reduce the buffer memory ============= */
1374 
1375 /*
1376  * try_to_free_buffer() checks if all the buffers on this particular page
1377  * are unused, and free's the page if so.
1378  */
1379 int try_to_free_buffer(struct buffer_head * bh, struct buffer_head ** bhp,
     /* [previous][next][first][last][top][bottom][index][help] */
1380                        int priority)
1381 {
1382         unsigned long page;
1383         struct buffer_head * tmp, * p;
1384         int isize = BUFSIZE_INDEX(bh->b_size);
1385 
1386         *bhp = bh;
1387         page = (unsigned long) bh->b_data;
1388         page &= PAGE_MASK;
1389         tmp = bh;
1390         do {
1391                 if (!tmp)
1392                         return 0;
1393                 if (tmp->b_count || buffer_protected(tmp) ||
1394                     buffer_dirty(tmp) || buffer_locked(tmp) || tmp->b_wait)
1395                         return 0;
1396                 if (priority && buffer_touched(tmp))
1397                         return 0;
1398                 tmp = tmp->b_this_page;
1399         } while (tmp != bh);
1400         tmp = bh;
1401         do {
1402                 p = tmp;
1403                 tmp = tmp->b_this_page;
1404                 nr_buffers--;
1405                 nr_buffers_size[isize]--;
1406                 if (p == *bhp)
1407                   {
1408                     *bhp = p->b_prev_free;
1409                     if (p == *bhp) /* Was this the last in the list? */
1410                       *bhp = NULL;
1411                   }
1412                 remove_from_queues(p);
1413                 put_unused_buffer_head(p);
1414         } while (tmp != bh);
1415         buffermem -= PAGE_SIZE;
1416         mem_map[MAP_NR(page)].buffers = NULL;
1417         free_page(page);
1418         return !mem_map[MAP_NR(page)].count;
1419 }
1420 
1421 /* Age buffers on a given page, according to whether they have been
1422    visited recently or not. */
1423 static inline void age_buffer(struct buffer_head *bh)
     /* [previous][next][first][last][top][bottom][index][help] */
1424 {
1425         struct buffer_head *tmp = bh;
1426         int touched = 0;
1427 
1428         /*
1429          * When we age a page, we mark all other buffers in the page
1430          * with the "has_aged" flag.  Then, when these aliased buffers
1431          * come up for aging, we skip them until next pass.  This
1432          * ensures that a page full of multiple buffers only gets aged
1433          * once per pass through the lru lists. 
1434          */
1435         if (clear_bit(BH_Has_aged, &bh->b_state))
1436                 return;
1437         
1438         do {
1439                 touched |= clear_bit(BH_Touched, &tmp->b_state);
1440                 tmp = tmp->b_this_page;
1441                 set_bit(BH_Has_aged, &tmp->b_state);
1442         } while (tmp != bh);
1443         clear_bit(BH_Has_aged, &bh->b_state);
1444 
1445         if (touched) 
1446                 touch_page(mem_map + MAP_NR((unsigned long) bh->b_data));
1447         else
1448                 age_page(mem_map + MAP_NR((unsigned long) bh->b_data));
1449 }
1450 
1451 /*
1452  * Consult the load average for buffers and decide whether or not
1453  * we should shrink the buffers of one size or not.  If we decide yes,
1454  * do it and return 1.  Else return 0.  Do not attempt to shrink size
1455  * that is specified.
1456  *
1457  * I would prefer not to use a load average, but the way things are now it
1458  * seems unavoidable.  The way to get rid of it would be to force clustering
1459  * universally, so that when we reclaim buffers we always reclaim an entire
1460  * page.  Doing this would mean that we all need to move towards QMAGIC.
1461  */
1462 
1463 static int maybe_shrink_lav_buffers(int size)
     /* [previous][next][first][last][top][bottom][index][help] */
1464 {          
1465         int nlist;
1466         int isize;
1467         int total_lav, total_n_buffers, n_sizes;
1468         
1469         /* Do not consider the shared buffers since they would not tend
1470            to have getblk called very often, and this would throw off
1471            the lav.  They are not easily reclaimable anyway (let the swapper
1472            make the first move). */
1473   
1474         total_lav = total_n_buffers = n_sizes = 0;
1475         for(nlist = 0; nlist < NR_SIZES; nlist++)
1476          {
1477                  total_lav += buffers_lav[nlist];
1478                  if(nr_buffers_size[nlist]) n_sizes++;
1479                  total_n_buffers += nr_buffers_size[nlist];
1480                  total_n_buffers -= nr_buffers_st[nlist][BUF_SHARED]; 
1481          }
1482         
1483         /* See if we have an excessive number of buffers of a particular
1484            size - if so, victimize that bunch. */
1485   
1486         isize = (size ? BUFSIZE_INDEX(size) : -1);
1487         
1488         if (n_sizes > 1)
1489                  for(nlist = 0; nlist < NR_SIZES; nlist++)
1490                   {
1491                           if(nlist == isize) continue;
1492                           if(nr_buffers_size[nlist] &&
1493                              bdf_prm.b_un.lav_const * buffers_lav[nlist]*total_n_buffers < 
1494                              total_lav * (nr_buffers_size[nlist] - nr_buffers_st[nlist][BUF_SHARED]))
1495                                    if(shrink_specific_buffers(6, bufferindex_size[nlist])) 
1496                                             return 1;
1497                   }
1498         return 0;
1499 }
1500 
1501 /*
1502  * Try to free up some pages by shrinking the buffer-cache
1503  *
1504  * Priority tells the routine how hard to try to shrink the
1505  * buffers: 6 means "don't bother too much", while a value
1506  * of 0 means "we'd better get some free pages now".
1507  *
1508  * "limit" is meant to limit the shrink-action only to pages
1509  * that are in the 0 - limit address range, for DMA re-allocations.
1510  * We ignore that right now.
1511  */
1512 
1513 static int shrink_specific_buffers(unsigned int priority, int size)
     /* [previous][next][first][last][top][bottom][index][help] */
1514 {
1515         struct buffer_head *bh;
1516         int nlist;
1517         int i, isize, isize1;
1518 
1519 #ifdef DEBUG
1520         if(size) printk("Shrinking buffers of size %d\n", size);
1521 #endif
1522         /* First try the free lists, and see if we can get a complete page
1523            from here */
1524         isize1 = (size ? BUFSIZE_INDEX(size) : -1);
1525 
1526         for(isize = 0; isize<NR_SIZES; isize++){
1527                 if(isize1 != -1 && isize1 != isize) continue;
1528                 bh = free_list[isize];
1529                 if(!bh) continue;
1530                 for (i=0 ; !i || bh != free_list[isize]; bh = bh->b_next_free, i++) {
1531                         if (bh->b_count || buffer_protected(bh) ||
1532                             !bh->b_this_page)
1533                                  continue;
1534                         if (!age_of((unsigned long) bh->b_data) &&
1535                             try_to_free_buffer(bh, &bh, 6))
1536                                  return 1;
1537                         if(!bh) break;
1538                         /* Some interrupt must have used it after we
1539                            freed the page.  No big deal - keep looking */
1540                 }
1541         }
1542         
1543         /* Not enough in the free lists, now try the lru list */
1544         
1545         for(nlist = 0; nlist < NR_LIST; nlist++) {
1546         repeat1:
1547                 if(priority > 2 && nlist == BUF_SHARED) continue;
1548                 i = nr_buffers_type[nlist];
1549                 i = ((BUFFEROUT_WEIGHT * i) >> 10) >> priority;
1550                 for ( ; i > 0; i-- ) {
1551                         bh = next_to_age[nlist];
1552                         if (!bh)
1553                                 break;
1554                         next_to_age[nlist] = bh->b_next_free;
1555 
1556                         /* First, age the buffer. */
1557                         age_buffer(bh);
1558                         /* We may have stalled while waiting for I/O
1559                            to complete. */
1560                         if(bh->b_list != nlist) goto repeat1;
1561                         if (bh->b_count || buffer_protected(bh) ||
1562                             !bh->b_this_page)
1563                                  continue;
1564                         if(size && bh->b_size != size) continue;
1565                         if (buffer_locked(bh))
1566                                  if (priority)
1567                                           continue;
1568                                  else
1569                                           wait_on_buffer(bh);
1570                         if (buffer_dirty(bh)) {
1571                                 bh->b_count++;
1572                                 bh->b_flushtime = 0;
1573                                 ll_rw_block(WRITEA, 1, &bh);
1574                                 bh->b_count--;
1575                                 continue;
1576                         }
1577                         /* At priority 6, only consider really old
1578                            (age==0) buffers for reclaiming.  At
1579                            priority 0, consider any buffers. */
1580                         if ((age_of((unsigned long) bh->b_data) >>
1581                              (6-priority)) > 0)
1582                                 continue;                               
1583                         if (try_to_free_buffer(bh, &bh, 0))
1584                                  return 1;
1585                         if(!bh) break;
1586                 }
1587         }
1588         return 0;
1589 }
1590 
1591 
1592 /* ================== Debugging =================== */
1593 
1594 void show_buffers(void)
     /* [previous][next][first][last][top][bottom][index][help] */
1595 {
1596         struct buffer_head * bh;
1597         int found = 0, locked = 0, dirty = 0, used = 0, lastused = 0;
1598         int protected = 0;
1599         int shared;
1600         int nlist, isize;
1601 
1602         printk("Buffer memory:   %6dkB\n",buffermem>>10);
1603         printk("Buffer heads:    %6d\n",nr_buffer_heads);
1604         printk("Buffer blocks:   %6d\n",nr_buffers);
1605 
1606         for(nlist = 0; nlist < NR_LIST; nlist++) {
1607           shared = found = locked = dirty = used = lastused = protected = 0;
1608           bh = lru_list[nlist];
1609           if(!bh) continue;
1610           do {
1611                 found++;
1612                 if (buffer_locked(bh))
1613                         locked++;
1614                 if (buffer_protected(bh))
1615                         protected++;
1616                 if (buffer_dirty(bh))
1617                         dirty++;
1618                 if (mem_map[MAP_NR(((unsigned long) bh->b_data))].count != 1)
1619                         shared++;
1620                 if (bh->b_count)
1621                         used++, lastused = found;
1622                 bh = bh->b_next_free;
1623           } while (bh != lru_list[nlist]);
1624           printk("Buffer[%d] mem: %d buffers, %d used (last=%d), "
1625                  "%d locked, %d protected, %d dirty %d shrd\n",
1626                  nlist, found, used, lastused,
1627                  locked, protected, dirty, shared);
1628         };
1629         printk("Size    [LAV]     Free  Clean  Unshar     Lck    Lck1   Dirty  Shared \n");
1630         for(isize = 0; isize<NR_SIZES; isize++){
1631                 printk("%5d [%5d]: %7d ", bufferindex_size[isize],
1632                        buffers_lav[isize], nr_free[isize]);
1633                 for(nlist = 0; nlist < NR_LIST; nlist++)
1634                          printk("%7d ", nr_buffers_st[isize][nlist]);
1635                 printk("\n");
1636         }
1637 }
1638 
1639 
1640 /* ====================== Cluster patches for ext2 ==================== */
1641 
1642 /*
1643  * try_to_reassign() checks if all the buffers on this particular page
1644  * are unused, and reassign to a new cluster them if this is true.
1645  */
1646 static inline int try_to_reassign(struct buffer_head * bh, struct buffer_head ** bhp,
     /* [previous][next][first][last][top][bottom][index][help] */
1647                            kdev_t dev, unsigned int starting_block)
1648 {
1649         unsigned long page;
1650         struct buffer_head * tmp, * p;
1651 
1652         *bhp = bh;
1653         page = (unsigned long) bh->b_data;
1654         page &= PAGE_MASK;
1655         if(mem_map[MAP_NR(page)].count != 1) return 0;
1656         tmp = bh;
1657         do {
1658                 if (!tmp)
1659                          return 0;
1660                 
1661                 if (tmp->b_count || buffer_protected(tmp) ||
1662                     buffer_dirty(tmp) || buffer_locked(tmp))
1663                          return 0;
1664                 tmp = tmp->b_this_page;
1665         } while (tmp != bh);
1666         tmp = bh;
1667         
1668         while((unsigned long) tmp->b_data & (PAGE_SIZE - 1)) 
1669                  tmp = tmp->b_this_page;
1670         
1671         /* This is the buffer at the head of the page */
1672         bh = tmp;
1673         do {
1674                 p = tmp;
1675                 tmp = tmp->b_this_page;
1676                 remove_from_queues(p);
1677                 p->b_dev = dev;
1678                 mark_buffer_uptodate(p, 0);
1679                 clear_bit(BH_Req, &p->b_state);
1680                 p->b_blocknr = starting_block++;
1681                 insert_into_queues(p);
1682         } while (tmp != bh);
1683         return 1;
1684 }
1685 
1686 /*
1687  * Try to find a free cluster by locating a page where
1688  * all of the buffers are unused.  We would like this function
1689  * to be atomic, so we do not call anything that might cause
1690  * the process to sleep.  The priority is somewhat similar to
1691  * the priority used in shrink_buffers.
1692  * 
1693  * My thinking is that the kernel should end up using whole
1694  * pages for the buffer cache as much of the time as possible.
1695  * This way the other buffers on a particular page are likely
1696  * to be very near each other on the free list, and we will not
1697  * be expiring data prematurely.  For now we only cannibalize buffers
1698  * of the same size to keep the code simpler.
1699  */
1700 static int reassign_cluster(kdev_t dev, 
     /* [previous][next][first][last][top][bottom][index][help] */
1701                      unsigned int starting_block, int size)
1702 {
1703         struct buffer_head *bh;
1704         int isize = BUFSIZE_INDEX(size);
1705         int i;
1706 
1707         /* We want to give ourselves a really good shot at generating
1708            a cluster, and since we only take buffers from the free
1709            list, we "overfill" it a little. */
1710 
1711         while(nr_free[isize] < 32) refill_freelist(size);
1712 
1713         bh = free_list[isize];
1714         if(bh)
1715                  for (i=0 ; !i || bh != free_list[isize] ; bh = bh->b_next_free, i++) {
1716                          if (!bh->b_this_page)  continue;
1717                          if (try_to_reassign(bh, &bh, dev, starting_block))
1718                                  return 4;
1719                  }
1720         return 0;
1721 }
1722 
1723 /* This function tries to generate a new cluster of buffers
1724  * from a new page in memory.  We should only do this if we have
1725  * not expanded the buffer cache to the maximum size that we allow.
1726  */
1727 static unsigned long try_to_generate_cluster(kdev_t dev, int block, int size)
     /* [previous][next][first][last][top][bottom][index][help] */
1728 {
1729         struct buffer_head * bh, * tmp, * arr[MAX_BUF_PER_PAGE];
1730         int isize = BUFSIZE_INDEX(size);
1731         unsigned long offset;
1732         unsigned long page;
1733         int nblock;
1734 
1735         page = get_free_page(GFP_NOBUFFER);
1736         if(!page) return 0;
1737 
1738         bh = create_buffers(page, size);
1739         if (!bh) {
1740                 free_page(page);
1741                 return 0;
1742         };
1743         nblock = block;
1744         for (offset = 0 ; offset < PAGE_SIZE ; offset += size) {
1745                 if (find_buffer(dev, nblock++, size))
1746                          goto not_aligned;
1747         }
1748         tmp = bh;
1749         nblock = 0;
1750         while (1) {
1751                 arr[nblock++] = bh;
1752                 bh->b_count = 1;
1753                 bh->b_flushtime = 0;
1754                 bh->b_state = 0;
1755                 bh->b_dev = dev;
1756                 bh->b_list = BUF_CLEAN;
1757                 bh->b_blocknr = block++;
1758                 nr_buffers++;
1759                 nr_buffers_size[isize]++;
1760                 insert_into_queues(bh);
1761                 if (bh->b_this_page)
1762                         bh = bh->b_this_page;
1763                 else
1764                         break;
1765         }
1766         buffermem += PAGE_SIZE;
1767         mem_map[MAP_NR(page)].buffers = bh;
1768         bh->b_this_page = tmp;
1769         while (nblock-- > 0)
1770                 brelse(arr[nblock]);
1771         return 4; /* ?? */
1772 not_aligned:
1773         while ((tmp = bh) != NULL) {
1774                 bh = bh->b_this_page;
1775                 put_unused_buffer_head(tmp);
1776         }
1777         free_page(page);
1778         return 0;
1779 }
1780 
1781 unsigned long generate_cluster(kdev_t dev, int b[], int size)
     /* [previous][next][first][last][top][bottom][index][help] */
1782 {
1783         int i, offset;
1784         
1785         for (i = 0, offset = 0 ; offset < PAGE_SIZE ; i++, offset += size) {
1786                 if(i && b[i]-1 != b[i-1]) return 0;  /* No need to cluster */
1787                 if(find_buffer(dev, b[i], size)) return 0;
1788         };
1789 
1790         /* OK, we have a candidate for a new cluster */
1791         
1792         /* See if one size of buffer is over-represented in the buffer cache,
1793            if so reduce the numbers of buffers */
1794         if(maybe_shrink_lav_buffers(size))
1795          {
1796                  int retval;
1797                  retval = try_to_generate_cluster(dev, b[0], size);
1798                  if(retval) return retval;
1799          };
1800         
1801         if (nr_free_pages > min_free_pages*2) 
1802                  return try_to_generate_cluster(dev, b[0], size);
1803         else
1804                  return reassign_cluster(dev, b[0], size);
1805 }
1806 
1807 
1808 /* ===================== Init ======================= */
1809 
1810 /*
1811  * This initializes the initial buffer free list.  nr_buffers_type is set
1812  * to one less the actual number of buffers, as a sop to backwards
1813  * compatibility --- the old code did this (I think unintentionally,
1814  * but I'm not sure), and programs in the ps package expect it.
1815  *                                      - TYT 8/30/92
1816  */
1817 void buffer_init(void)
     /* [previous][next][first][last][top][bottom][index][help] */
1818 {
1819         int i;
1820         int isize = BUFSIZE_INDEX(BLOCK_SIZE);
1821         long memsize = MAP_NR(high_memory) << PAGE_SHIFT;
1822 
1823         if (memsize >= 4*1024*1024) {
1824                 if(memsize >= 16*1024*1024)
1825                          nr_hash = 16381;
1826                 else
1827                          nr_hash = 4093;
1828         } else {
1829                 nr_hash = 997;
1830         };
1831         
1832         hash_table = (struct buffer_head **) vmalloc(nr_hash * 
1833                                                      sizeof(struct buffer_head *));
1834 
1835 
1836         for (i = 0 ; i < nr_hash ; i++)
1837                 hash_table[i] = NULL;
1838         lru_list[BUF_CLEAN] = 0;
1839         grow_buffers(GFP_KERNEL, BLOCK_SIZE);
1840         if (!free_list[isize])
1841                 panic("VFS: Unable to initialize buffer free list!");
1842         return;
1843 }
1844 
1845 
1846 /* ====================== bdflush support =================== */
1847 
1848 /* This is a simple kernel daemon, whose job it is to provide a dynamic
1849  * response to dirty buffers.  Once this process is activated, we write back
1850  * a limited number of buffers to the disks and then go back to sleep again.
1851  */
1852 struct wait_queue * bdflush_wait = NULL;
1853 struct wait_queue * bdflush_done = NULL;
1854 
1855 static void wakeup_bdflush(int wait)
     /* [previous][next][first][last][top][bottom][index][help] */
1856 {
1857         wake_up(&bdflush_wait);
1858         if (wait) {
1859                 run_task_queue(&tq_disk);
1860                 sleep_on(&bdflush_done);
1861         }
1862 }
1863 
1864 
1865 /* 
1866  * Here we attempt to write back old buffers.  We also try and flush inodes 
1867  * and supers as well, since this function is essentially "update", and 
1868  * otherwise there would be no way of ensuring that these quantities ever 
1869  * get written back.  Ideally, we would have a timestamp on the inodes
1870  * and superblocks so that we could write back only the old ones as well
1871  */
1872 
1873 asmlinkage int sync_old_buffers(void)
     /* [previous][next][first][last][top][bottom][index][help] */
1874 {
1875         int i, isize;
1876         int ndirty, nwritten;
1877         int nlist;
1878         int ncount;
1879         struct buffer_head * bh, *next;
1880 
1881         sync_supers(0);
1882         sync_inodes(0);
1883 
1884         ncount = 0;
1885 #ifdef DEBUG
1886         for(nlist = 0; nlist < NR_LIST; nlist++)
1887 #else
1888         for(nlist = BUF_DIRTY; nlist <= BUF_DIRTY; nlist++)
1889 #endif
1890         {
1891                 ndirty = 0;
1892                 nwritten = 0;
1893         repeat:
1894                 bh = lru_list[nlist];
1895                 if(bh) 
1896                          for (i = nr_buffers_type[nlist]; i-- > 0; bh = next) {
1897                                  /* We may have stalled while waiting for I/O to complete. */
1898                                  if(bh->b_list != nlist) goto repeat;
1899                                  next = bh->b_next_free;
1900                                  if(!lru_list[nlist]) {
1901                                          printk("Dirty list empty %d\n", i);
1902                                          break;
1903                                  }
1904                                  
1905                                  /* Clean buffer on dirty list?  Refile it */
1906                                  if (nlist == BUF_DIRTY && !buffer_dirty(bh) && !buffer_locked(bh))
1907                                   {
1908                                           refile_buffer(bh);
1909                                           continue;
1910                                   }
1911                                  
1912                                  if (buffer_locked(bh) || !buffer_dirty(bh))
1913                                           continue;
1914                                  ndirty++;
1915                                  if(bh->b_flushtime > jiffies) continue;
1916                                  nwritten++;
1917                                  bh->b_count++;
1918                                  bh->b_flushtime = 0;
1919 #ifdef DEBUG
1920                                  if(nlist != BUF_DIRTY) ncount++;
1921 #endif
1922                                  ll_rw_block(WRITE, 1, &bh);
1923                                  bh->b_count--;
1924                          }
1925         }
1926 #ifdef DEBUG
1927         if (ncount) printk("sync_old_buffers: %d dirty buffers not on dirty list\n", ncount);
1928         printk("Wrote %d/%d buffers\n", nwritten, ndirty);
1929 #endif
1930         
1931         /* We assume that we only come through here on a regular
1932            schedule, like every 5 seconds.  Now update load averages.  
1933            Shift usage counts to prevent overflow. */
1934         for(isize = 0; isize<NR_SIZES; isize++){
1935                 CALC_LOAD(buffers_lav[isize], bdf_prm.b_un.lav_const, buffer_usage[isize]);
1936                 buffer_usage[isize] = 0;
1937         };
1938         return 0;
1939 }
1940 
1941 
1942 /* This is the interface to bdflush.  As we get more sophisticated, we can
1943  * pass tuning parameters to this "process", to adjust how it behaves. 
1944  * We would want to verify each parameter, however, to make sure that it 
1945  * is reasonable. */
1946 
1947 asmlinkage int sys_bdflush(int func, long data)
     /* [previous][next][first][last][top][bottom][index][help] */
1948 {
1949         int i, error;
1950 
1951         if (!suser())
1952                 return -EPERM;
1953 
1954         if (func == 1)
1955                  return sync_old_buffers();
1956 
1957         /* Basically func 1 means read param 1, 2 means write param 1, etc */
1958         if (func >= 2) {
1959                 i = (func-2) >> 1;
1960                 if (i < 0 || i >= N_PARAM)
1961                         return -EINVAL;
1962                 if((func & 1) == 0) {
1963                         error = verify_area(VERIFY_WRITE, (void *) data, sizeof(int));
1964                         if (error)
1965                                 return error;
1966                         put_user(bdf_prm.data[i], (int*)data);
1967                         return 0;
1968                 };
1969                 if (data < bdflush_min[i] || data > bdflush_max[i])
1970                         return -EINVAL;
1971                 bdf_prm.data[i] = data;
1972                 return 0;
1973         }
1974 
1975         /* Having func 0 used to launch the actual bdflush and then never
1976         return (unless explicitly killed). We return zero here to 
1977         remain semi-compatible with present update(8) programs. */
1978 
1979         return 0;
1980 }
1981 
1982 /* This is the actual bdflush daemon itself. It used to be started from
1983  * the syscall above, but now we launch it ourselves internally with
1984  * kernel_thread(...)  directly after the first thread in init/main.c */
1985 
1986 int bdflush(void * unused) 
     /* [previous][next][first][last][top][bottom][index][help] */
1987 {
1988         int i;
1989         int ndirty;
1990         int nlist;
1991         int ncount;
1992         struct buffer_head * bh, *next;
1993 
1994         /*
1995          *      We have a bare-bones task_struct, and really should fill
1996          *      in a few more things so "top" and /proc/2/{exe,root,cwd}
1997          *      display semi-sane things. Not real crucial though...  
1998          */
1999 
2000         current->session = 1;
2001         current->pgrp = 1;
2002         sprintf(current->comm, "kflushd");
2003 
2004         /*
2005          *      As a kernel thread we want to tamper with system buffers
2006          *      and other internals and thus be subject to the SMP locking
2007          *      rules. (On a uniprocessor box this does nothing).
2008          */
2009          
2010 #ifdef __SMP__
2011         lock_kernel();
2012         syscall_count++;
2013 #endif
2014                  
2015         for (;;) {
2016 #ifdef DEBUG
2017                 printk("bdflush() activated...");
2018 #endif
2019                 
2020                 ncount = 0;
2021 #ifdef DEBUG
2022                 for(nlist = 0; nlist < NR_LIST; nlist++)
2023 #else
2024                 for(nlist = BUF_DIRTY; nlist <= BUF_DIRTY; nlist++)
2025 #endif
2026                  {
2027                          ndirty = 0;
2028                  repeat:
2029                          bh = lru_list[nlist];
2030                          if(bh) 
2031                                   for (i = nr_buffers_type[nlist]; i-- > 0 && ndirty < bdf_prm.b_un.ndirty; 
2032                                        bh = next) {
2033                                           /* We may have stalled while waiting for I/O to complete. */
2034                                           if(bh->b_list != nlist) goto repeat;
2035                                           next = bh->b_next_free;
2036                                           if(!lru_list[nlist]) {
2037                                                   printk("Dirty list empty %d\n", i);
2038                                                   break;
2039                                           }
2040                                           
2041                                           /* Clean buffer on dirty list?  Refile it */
2042                                           if (nlist == BUF_DIRTY && !buffer_dirty(bh) && !buffer_locked(bh))
2043                                            {
2044                                                    refile_buffer(bh);
2045                                                    continue;
2046                                            }
2047                                           
2048                                           if (buffer_locked(bh) || !buffer_dirty(bh))
2049                                                    continue;
2050                                           /* Should we write back buffers that are shared or not??
2051                                              currently dirty buffers are not shared, so it does not matter */
2052                                           bh->b_count++;
2053                                           ndirty++;
2054                                           bh->b_flushtime = 0;
2055                                           ll_rw_block(WRITE, 1, &bh);
2056 #ifdef DEBUG
2057                                           if(nlist != BUF_DIRTY) ncount++;
2058 #endif
2059                                           bh->b_count--;
2060                                   }
2061                  }
2062 #ifdef DEBUG
2063                 if (ncount) printk("sys_bdflush: %d dirty buffers not on dirty list\n", ncount);
2064                 printk("sleeping again.\n");
2065 #endif
2066                 run_task_queue(&tq_disk);
2067                 wake_up(&bdflush_done);
2068                 
2069                 /* If there are still a lot of dirty buffers around, skip the sleep
2070                    and flush some more */
2071                 
2072                 if(nr_buffers_type[BUF_DIRTY] <= (nr_buffers - nr_buffers_type[BUF_SHARED]) * 
2073                    bdf_prm.b_un.nfract/100) {
2074                         current->signal = 0;
2075                         interruptible_sleep_on(&bdflush_wait);
2076                 }
2077         }
2078 }
2079 
2080 
2081 /*
2082  * Overrides for Emacs so that we follow Linus's tabbing style.
2083  * Emacs will notice this stuff at the end of the file and automatically
2084  * adjust the settings for this buffer only.  This must remain at the end
2085  * of the file.
2086  * ---------------------------------------------------------------------------
2087  * Local variables:
2088  * c-indent-level: 8
2089  * c-brace-imaginary-offset: 0
2090  * c-brace-offset: -8
2091  * c-argdecl-indent: 8
2092  * c-label-offset: -8
2093  * c-continued-statement-offset: 8
2094  * c-continued-brace-offset: 0
2095  * End:
2096  */

/* [previous][next][first][last][top][bottom][index][help] */