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 (grow_buffers(GFP_ATOMIC, size)) {
 725                 needed -= PAGE_SIZE;
 726                 goto repeat0;
 727         }
 728         wakeup_bdflush(1);
 729         goto repeat0;
 730 }
 731 
 732 /*
 733  * Ok, this is getblk, and it isn't very clear, again to hinder
 734  * race-conditions. Most of the code is seldom used, (ie repeating),
 735  * so it should be much more efficient than it looks.
 736  *
 737  * The algorithm is changed: hopefully better, and an elusive bug removed.
 738  *
 739  * 14.02.92: changed it to sync dirty buffers a bit: better performance
 740  * when the filesystem starts to get full of dirty blocks (I hope).
 741  */
 742 struct buffer_head * getblk(kdev_t dev, int block, int size)
     /* [previous][next][first][last][top][bottom][index][help] */
 743 {
 744         struct buffer_head * bh;
 745         int isize = BUFSIZE_INDEX(size);
 746 
 747         /* Update this for the buffer size lav. */
 748         buffer_usage[isize]++;
 749 
 750         /* If there are too many dirty buffers, we wake up the update process
 751            now so as to ensure that there are still clean buffers available
 752            for user processes to use (and dirty) */
 753 repeat:
 754         bh = get_hash_table(dev, block, size);
 755         if (bh) {
 756                 if (!buffer_dirty(bh)) {
 757                         if (buffer_uptodate(bh))
 758                                  put_last_lru(bh);
 759                         bh->b_flushtime = 0;
 760                 }
 761                 set_bit(BH_Touched, &bh->b_state);
 762                 return bh;
 763         }
 764 
 765         while(!free_list[isize]) refill_freelist(size);
 766         
 767         if (find_buffer(dev,block,size))
 768                  goto repeat;
 769 
 770         bh = free_list[isize];
 771         remove_from_free_list(bh);
 772 
 773 /* OK, FINALLY we know that this buffer is the only one of its kind, */
 774 /* and that it's unused (b_count=0), unlocked (buffer_locked=0), and clean */
 775         bh->b_count=1;
 776         bh->b_flushtime=0;
 777         bh->b_state=(1<<BH_Touched);
 778         bh->b_dev=dev;
 779         bh->b_blocknr=block;
 780         insert_into_queues(bh);
 781         return bh;
 782 }
 783 
 784 void set_writetime(struct buffer_head * buf, int flag)
     /* [previous][next][first][last][top][bottom][index][help] */
 785 {
 786         int newtime;
 787 
 788         if (buffer_dirty(buf)) {
 789                 /* Move buffer to dirty list if jiffies is clear */
 790                 newtime = jiffies + (flag ? bdf_prm.b_un.age_super : 
 791                                      bdf_prm.b_un.age_buffer);
 792                 if(!buf->b_flushtime || buf->b_flushtime > newtime)
 793                          buf->b_flushtime = newtime;
 794         } else {
 795                 buf->b_flushtime = 0;
 796         }
 797 }
 798 
 799 
 800 /*
 801  * A buffer may need to be moved from one buffer list to another
 802  * (e.g. in case it is not shared any more). Handle this.
 803  */
 804 void refile_buffer(struct buffer_head * buf)
     /* [previous][next][first][last][top][bottom][index][help] */
 805 {
 806         int dispose;
 807 
 808         if(buf->b_dev == B_FREE) {
 809                 printk("Attempt to refile free buffer\n");
 810                 return;
 811         }
 812         if (buffer_dirty(buf))
 813                 dispose = BUF_DIRTY;
 814         else if ((mem_map[MAP_NR((unsigned long) buf->b_data)].count > 1) || buffer_protected(buf))
 815                 dispose = BUF_SHARED;
 816         else if (buffer_locked(buf))
 817                 dispose = BUF_LOCKED;
 818         else if (buf->b_list == BUF_SHARED)
 819                 dispose = BUF_UNSHARED;
 820         else
 821                 dispose = BUF_CLEAN;
 822         if(dispose == BUF_CLEAN) buf->b_lru_time = jiffies;
 823         if(dispose != buf->b_list)  {
 824                 if(dispose == BUF_DIRTY || dispose == BUF_UNSHARED)
 825                          buf->b_lru_time = jiffies;
 826                 if(dispose == BUF_LOCKED && 
 827                    (buf->b_flushtime - buf->b_lru_time) <= bdf_prm.b_un.age_super)
 828                          dispose = BUF_LOCKED1;
 829                 remove_from_queues(buf);
 830                 buf->b_list = dispose;
 831                 insert_into_queues(buf);
 832                 if(dispose == BUF_DIRTY && nr_buffers_type[BUF_DIRTY] > 
 833                    (nr_buffers - nr_buffers_type[BUF_SHARED]) *
 834                    bdf_prm.b_un.nfract/100)
 835                          wakeup_bdflush(0);
 836         }
 837 }
 838 
 839 /*
 840  * Release a buffer head
 841  */
 842 void __brelse(struct buffer_head * buf)
     /* [previous][next][first][last][top][bottom][index][help] */
 843 {
 844         wait_on_buffer(buf);
 845 
 846         /* If dirty, mark the time this buffer should be written back */
 847         set_writetime(buf, 0);
 848         refile_buffer(buf);
 849 
 850         if (buf->b_count) {
 851                 buf->b_count--;
 852                 return;
 853         }
 854         printk("VFS: brelse: Trying to free free buffer\n");
 855 }
 856 
 857 /*
 858  * bforget() is like brelse(), except it removes the buffer
 859  * from the hash-queues (so that it won't be re-used if it's
 860  * shared).
 861  */
 862 void __bforget(struct buffer_head * buf)
     /* [previous][next][first][last][top][bottom][index][help] */
 863 {
 864         wait_on_buffer(buf);
 865         mark_buffer_clean(buf);
 866         clear_bit(BH_Protected, &buf->b_state);
 867         buf->b_count--;
 868         remove_from_hash_queue(buf);
 869         buf->b_dev = NODEV;
 870         refile_buffer(buf);
 871 }
 872 
 873 /*
 874  * bread() reads a specified block and returns the buffer that contains
 875  * it. It returns NULL if the block was unreadable.
 876  */
 877 struct buffer_head * bread(kdev_t dev, int block, int size)
     /* [previous][next][first][last][top][bottom][index][help] */
 878 {
 879         struct buffer_head * bh;
 880 
 881         if (!(bh = getblk(dev, block, size))) {
 882                 printk("VFS: bread: READ error on device %s\n",
 883                         kdevname(dev));
 884                 return NULL;
 885         }
 886         if (buffer_uptodate(bh))
 887                 return bh;
 888         ll_rw_block(READ, 1, &bh);
 889         wait_on_buffer(bh);
 890         if (buffer_uptodate(bh))
 891                 return bh;
 892         brelse(bh);
 893         return NULL;
 894 }
 895 
 896 /*
 897  * Ok, breada can be used as bread, but additionally to mark other
 898  * blocks for reading as well. End the argument list with a negative
 899  * number.
 900  */
 901 
 902 #define NBUF 16
 903 
 904 struct buffer_head * breada(kdev_t dev, int block, int bufsize,
     /* [previous][next][first][last][top][bottom][index][help] */
 905         unsigned int pos, unsigned int filesize)
 906 {
 907         struct buffer_head * bhlist[NBUF];
 908         unsigned int blocks;
 909         struct buffer_head * bh;
 910         int index;
 911         int i, j;
 912 
 913         if (pos >= filesize)
 914                 return NULL;
 915 
 916         if (block < 0 || !(bh = getblk(dev,block,bufsize)))
 917                 return NULL;
 918 
 919         index = BUFSIZE_INDEX(bh->b_size);
 920 
 921         if (buffer_uptodate(bh))
 922                 return(bh);   
 923         else ll_rw_block(READ, 1, &bh);
 924 
 925         blocks = (filesize - pos) >> (9+index);
 926 
 927         if (blocks < (read_ahead[MAJOR(dev)] >> index))
 928                 blocks = read_ahead[MAJOR(dev)] >> index;
 929         if (blocks > NBUF) 
 930                 blocks = NBUF;
 931 
 932 /*      if (blocks) printk("breada (new) %d blocks\n",blocks); */
 933 
 934 
 935         bhlist[0] = bh;
 936         j = 1;
 937         for(i=1; i<blocks; i++) {
 938                 bh = getblk(dev,block+i,bufsize);
 939                 if (buffer_uptodate(bh)) {
 940                         brelse(bh);
 941                         break;
 942                 }
 943                 else bhlist[j++] = bh;
 944         }
 945 
 946         /* Request the read for these buffers, and then release them */
 947         if (j>1)  
 948                 ll_rw_block(READA, (j-1), bhlist+1); 
 949         for(i=1; i<j; i++)
 950                 brelse(bhlist[i]);
 951 
 952         /* Wait for this buffer, and then continue on */
 953         bh = bhlist[0];
 954         wait_on_buffer(bh);
 955         if (buffer_uptodate(bh))
 956                 return bh;
 957         brelse(bh);
 958         return NULL;
 959 }
 960 
 961 /*
 962  * See fs/inode.c for the weird use of volatile..
 963  */
 964 static void put_unused_buffer_head(struct buffer_head * bh)
     /* [previous][next][first][last][top][bottom][index][help] */
 965 {
 966         struct wait_queue * wait;
 967 
 968         wait = ((volatile struct buffer_head *) bh)->b_wait;
 969         memset(bh,0,sizeof(*bh));
 970         ((volatile struct buffer_head *) bh)->b_wait = wait;
 971         bh->b_next_free = unused_list;
 972         unused_list = bh;
 973         wake_up(&buffer_wait);
 974 }
 975 
 976 static void get_more_buffer_heads(void)
     /* [previous][next][first][last][top][bottom][index][help] */
 977 {
 978         int i;
 979         struct buffer_head * bh;
 980 
 981         for (;;) {
 982                 if (unused_list)
 983                         return;
 984 
 985                 /*
 986                  * This is critical.  We can't swap out pages to get
 987                  * more buffer heads, because the swap-out may need
 988                  * more buffer-heads itself.  Thus GFP_ATOMIC.
 989                  */
 990                 bh = (struct buffer_head *) get_free_page(GFP_ATOMIC);
 991                 if (bh)
 992                         break;
 993 
 994                 /*
 995                  * Uhhuh. We're _really_ low on memory. Now we just
 996                  * wait for old buffer heads to become free due to
 997                  * finishing IO..
 998                  */
 999                 run_task_queue(&tq_disk);
1000                 sleep_on(&buffer_wait);
1001         }
1002 
1003         for (nr_buffer_heads+=i=PAGE_SIZE/sizeof*bh ; i>0; i--) {
1004                 bh->b_next_free = unused_list;  /* only make link */
1005                 unused_list = bh++;
1006         }
1007 }
1008 
1009 /* 
1010  * We can't put completed temporary IO buffer_heads directly onto the
1011  * unused_list when they become unlocked, since the device driver
1012  * end_request routines still expect access to the buffer_head's
1013  * fields after the final unlock.  So, the device driver puts them on
1014  * the reuse_list instead once IO completes, and we recover these to
1015  * the unused_list here.
1016  *
1017  * The reuse_list receives buffers from interrupt routines, so we need
1018  * to be IRQ-safe here (but note that interrupts only _add_ to the
1019  * reuse_list, never take away. So we don't need to worry about the
1020  * reuse_list magically emptying).
1021  */
1022 static inline void recover_reusable_buffer_heads(void)
     /* [previous][next][first][last][top][bottom][index][help] */
1023 {
1024         if (reuse_list) {
1025                 struct buffer_head *bh;
1026                 unsigned long flags;
1027         
1028                 save_flags(flags);
1029                 do {
1030                         cli();
1031                         bh = reuse_list;
1032                         reuse_list = bh->b_next_free;
1033                         restore_flags(flags);
1034                         put_unused_buffer_head(bh);
1035                 } while (reuse_list);
1036         }
1037 }
1038 
1039 static struct buffer_head * get_unused_buffer_head(void)
     /* [previous][next][first][last][top][bottom][index][help] */
1040 {
1041         struct buffer_head * bh;
1042 
1043         recover_reusable_buffer_heads();
1044         get_more_buffer_heads();
1045         if (!unused_list)
1046                 return NULL;
1047         bh = unused_list;
1048         unused_list = bh->b_next_free;
1049         bh->b_next_free = NULL;
1050         bh->b_data = NULL;
1051         bh->b_size = 0;
1052         bh->b_state = 0;
1053         return bh;
1054 }
1055 
1056 /*
1057  * Create the appropriate buffers when given a page for data area and
1058  * the size of each buffer.. Use the bh->b_this_page linked list to
1059  * follow the buffers created.  Return NULL if unable to create more
1060  * buffers.
1061  */
1062 static struct buffer_head * create_buffers(unsigned long page, unsigned long size)
     /* [previous][next][first][last][top][bottom][index][help] */
1063 {
1064         struct buffer_head *bh, *head;
1065         unsigned long offset;
1066 
1067         head = NULL;
1068         offset = PAGE_SIZE;
1069         while ((offset -= size) < PAGE_SIZE) {
1070                 bh = get_unused_buffer_head();
1071                 if (!bh)
1072                         goto no_grow;
1073                 bh->b_this_page = head;
1074                 head = bh;
1075                 bh->b_data = (char *) (page+offset);
1076                 bh->b_size = size;
1077                 bh->b_dev = B_FREE;  /* Flag as unused */
1078         }
1079         return head;
1080 /*
1081  * In case anything failed, we just free everything we got.
1082  */
1083 no_grow:
1084         bh = head;
1085         while (bh) {
1086                 head = bh;
1087                 bh = bh->b_this_page;
1088                 put_unused_buffer_head(head);
1089         }
1090         return NULL;
1091 }
1092 
1093 /* Run the hooks that have to be done when a page I/O has completed. */
1094 static inline void after_unlock_page (struct page * page)
     /* [previous][next][first][last][top][bottom][index][help] */
1095 {
1096         if (clear_bit(PG_decr_after, &page->flags))
1097                 nr_async_pages--;
1098         if (clear_bit(PG_free_after, &page->flags))
1099                 free_page(page_address(page));
1100         if (clear_bit(PG_swap_unlock_after, &page->flags))
1101                 swap_after_unlock_page(page->swap_unlock_entry);
1102 }
1103 
1104 /* Free all temporary buffers belonging to a page. */
1105 static inline void free_async_buffers (struct buffer_head * bh)
     /* [previous][next][first][last][top][bottom][index][help] */
1106 {
1107         struct buffer_head * tmp;
1108         unsigned long flags;
1109 
1110         tmp = bh;
1111         save_flags(flags);
1112         cli();
1113         do {
1114                 if (!test_bit(BH_FreeOnIO, &tmp->b_state)) {
1115                         printk ("Whoops: unlock_buffer: "
1116                                 "async IO mismatch on page.\n");
1117                         restore_flags(flags);
1118                         return;
1119                 }
1120                 tmp->b_next_free = reuse_list;
1121                 reuse_list = tmp;
1122                 clear_bit(BH_FreeOnIO, &tmp->b_state);
1123                 tmp = tmp->b_this_page;
1124         } while (tmp != bh);
1125         restore_flags(flags);
1126 }
1127 
1128 /*
1129  * Start I/O on a page.
1130  * This function expects the page to be locked and may return before I/O is complete.
1131  * You then have to check page->locked, page->uptodate, and maybe wait on page->wait.
1132  */
1133 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] */
1134 {
1135         struct buffer_head *bh, *prev, *next, *arr[MAX_BUF_PER_PAGE];
1136         int block, nr;
1137         struct page *page;
1138 
1139         page = mem_map + MAP_NR(address);
1140         if (!PageLocked(page))
1141                 panic("brw_page: page not locked for I/O");
1142         clear_bit(PG_uptodate, &page->flags);
1143         /*
1144          * Allocate buffer heads pointing to this page, just for I/O.
1145          * They do _not_ show up in the buffer hash table!
1146          * They are _not_ registered in page->buffers either!
1147          */
1148         bh = create_buffers(address, size);
1149         if (!bh) {
1150                 clear_bit(PG_locked, &page->flags);
1151                 wake_up(&page->wait);
1152                 return -ENOMEM;
1153         }
1154         nr = 0;
1155         next = bh;
1156         do {
1157                 struct buffer_head * tmp;
1158                 block = *(b++);
1159 
1160                 set_bit(BH_FreeOnIO, &next->b_state);
1161                 next->b_list = BUF_CLEAN;
1162                 next->b_dev = dev;
1163                 next->b_blocknr = block;
1164                 next->b_count = 1;
1165                 next->b_flushtime = 0;
1166                 set_bit(BH_Uptodate, &next->b_state);
1167 
1168                 /* When we use bmap, we define block zero to represent
1169                    a hole.  ll_rw_page, however, may legitimately
1170                    access block zero, and we need to distinguish the
1171                    two cases. 
1172                    */
1173                 if (bmap && !block) {
1174                         memset(next->b_data, 0, size);
1175                         next->b_count--;
1176                         continue;
1177                 }
1178                 tmp = get_hash_table(dev, block, size);
1179                 if (tmp) {
1180                         if (!buffer_uptodate(tmp)) {
1181                                 if (rw == READ)
1182                                         ll_rw_block(READ, 1, &tmp);
1183                                 wait_on_buffer(tmp);
1184                         }
1185                         if (rw == READ) 
1186                                 memcpy(next->b_data, tmp->b_data, size);
1187                         else {
1188                                 memcpy(tmp->b_data, next->b_data, size);
1189                                 mark_buffer_dirty(tmp, 0);
1190                         }
1191                         brelse(tmp);
1192                         next->b_count--;
1193                         continue;
1194                 }
1195                 if (rw == READ)
1196                         clear_bit(BH_Uptodate, &next->b_state);
1197                 else
1198                         set_bit(BH_Dirty, &next->b_state);
1199                 arr[nr++] = next;
1200         } while (prev = next, (next = next->b_this_page) != NULL);
1201         prev->b_this_page = bh;
1202         
1203         if (nr) {
1204                 ll_rw_block(rw, nr, arr);
1205                 /* The rest of the work is done in mark_buffer_uptodate()
1206                  * and unlock_buffer(). */
1207         } else {
1208                 clear_bit(PG_locked, &page->flags);
1209                 set_bit(PG_uptodate, &page->flags);
1210                 wake_up(&page->wait);
1211                 free_async_buffers(bh);
1212                 after_unlock_page(page);
1213         }
1214         ++current->maj_flt;
1215         return 0;
1216 }
1217 
1218 /*
1219  * This is called by end_request() when I/O has completed.
1220  */
1221 void mark_buffer_uptodate(struct buffer_head * bh, int on)
     /* [previous][next][first][last][top][bottom][index][help] */
1222 {
1223         if (on) {
1224                 struct buffer_head *tmp = bh;
1225                 int page_uptodate = 1;
1226                 set_bit(BH_Uptodate, &bh->b_state);
1227                 /* If a page has buffers and all these buffers are uptodate,
1228                  * then the page is uptodate. */
1229                 do {
1230                         if (!test_bit(BH_Uptodate, &tmp->b_state)) {
1231                                 page_uptodate = 0;
1232                                 break;
1233                         }
1234                         tmp=tmp->b_this_page;
1235                 } while (tmp && tmp != bh);
1236                 if (page_uptodate)
1237                         set_bit(PG_uptodate, &mem_map[MAP_NR(bh->b_data)].flags);
1238         } else
1239                 clear_bit(BH_Uptodate, &bh->b_state);
1240 }
1241 
1242 /*
1243  * This is called by end_request() when I/O has completed.
1244  */
1245 void unlock_buffer(struct buffer_head * bh)
     /* [previous][next][first][last][top][bottom][index][help] */
1246 {
1247         struct buffer_head *tmp;
1248         struct page *page;
1249 
1250         clear_bit(BH_Lock, &bh->b_state);
1251         wake_up(&bh->b_wait);
1252 
1253         if (!test_bit(BH_FreeOnIO, &bh->b_state))
1254                 return;
1255         /* This is a temporary buffer used for page I/O. */
1256         page = mem_map + MAP_NR(bh->b_data);
1257         if (!PageLocked(page)) {
1258                 printk ("Whoops: unlock_buffer: "
1259                         "async io complete on unlocked page\n");
1260                 return;
1261         }
1262         if (bh->b_count != 1) {
1263                 printk ("Whoops: unlock_buffer: b_count != 1 on async io.\n");
1264                 return;
1265         }
1266         /* Async buffer_heads are here only as labels for IO, and get
1267            thrown away once the IO for this page is complete.  IO is
1268            deemed complete once all buffers have been visited
1269            (b_count==0) and are now unlocked. */
1270         bh->b_count--;
1271         for (tmp = bh; tmp=tmp->b_this_page, tmp!=bh; ) {
1272                 if (test_bit(BH_Lock, &tmp->b_state) || tmp->b_count)
1273                         return;
1274         }
1275         /* OK, the async IO on this page is complete. */
1276         clear_bit(PG_locked, &page->flags);
1277         wake_up(&page->wait);
1278         free_async_buffers(bh);
1279         after_unlock_page(page);
1280         wake_up(&buffer_wait);
1281 }
1282 
1283 /*
1284  * Generic "readpage" function for block devices that have the normal
1285  * bmap functionality. This is most of the block device filesystems.
1286  * Reads the page asynchronously --- the unlock_buffer() and
1287  * mark_buffer_uptodate() functions propagate buffer state into the
1288  * page struct once IO has completed.
1289  */
1290 int generic_readpage(struct inode * inode, struct page * page)
     /* [previous][next][first][last][top][bottom][index][help] */
1291 {
1292         unsigned long block, address;
1293         int *p, nr[PAGE_SIZE/512];
1294         int i;
1295 
1296         address = page_address(page);
1297         page->count++;
1298         set_bit(PG_locked, &page->flags);
1299         set_bit(PG_free_after, &page->flags);
1300         
1301         i = PAGE_SIZE >> inode->i_sb->s_blocksize_bits;
1302         block = page->offset >> inode->i_sb->s_blocksize_bits;
1303         p = nr;
1304         do {
1305                 *p = inode->i_op->bmap(inode, block);
1306                 i--;
1307                 block++;
1308                 p++;
1309         } while (i > 0);
1310 
1311         /* IO start */
1312         brw_page(READ, address, inode->i_dev, nr, inode->i_sb->s_blocksize, 1);
1313         return 0;
1314 }
1315 
1316 /*
1317  * Try to increase the number of buffers available: the size argument
1318  * is used to determine what kind of buffers we want.
1319  */
1320 static int grow_buffers(int pri, int size)
     /* [previous][next][first][last][top][bottom][index][help] */
1321 {
1322         unsigned long page;
1323         struct buffer_head *bh, *tmp;
1324         struct buffer_head * insert_point;
1325         int isize;
1326 
1327         if ((size & 511) || (size > PAGE_SIZE)) {
1328                 printk("VFS: grow_buffers: size = %d\n",size);
1329                 return 0;
1330         }
1331 
1332         isize = BUFSIZE_INDEX(size);
1333 
1334         if (!(page = __get_free_page(pri)))
1335                 return 0;
1336         bh = create_buffers(page, size);
1337         if (!bh) {
1338                 free_page(page);
1339                 return 0;
1340         }
1341 
1342         insert_point = free_list[isize];
1343 
1344         tmp = bh;
1345         while (1) {
1346                 nr_free[isize]++;
1347                 if (insert_point) {
1348                         tmp->b_next_free = insert_point->b_next_free;
1349                         tmp->b_prev_free = insert_point;
1350                         insert_point->b_next_free->b_prev_free = tmp;
1351                         insert_point->b_next_free = tmp;
1352                 } else {
1353                         tmp->b_prev_free = tmp;
1354                         tmp->b_next_free = tmp;
1355                 }
1356                 insert_point = tmp;
1357                 ++nr_buffers;
1358                 if (tmp->b_this_page)
1359                         tmp = tmp->b_this_page;
1360                 else
1361                         break;
1362         }
1363         free_list[isize] = bh;
1364         mem_map[MAP_NR(page)].buffers = bh;
1365         tmp->b_this_page = bh;
1366         buffermem += PAGE_SIZE;
1367         return 1;
1368 }
1369 
1370 
1371 /* =========== Reduce the buffer memory ============= */
1372 
1373 /*
1374  * try_to_free_buffer() checks if all the buffers on this particular page
1375  * are unused, and free's the page if so.
1376  */
1377 int try_to_free_buffer(struct buffer_head * bh, struct buffer_head ** bhp,
     /* [previous][next][first][last][top][bottom][index][help] */
1378                        int priority)
1379 {
1380         unsigned long page;
1381         struct buffer_head * tmp, * p;
1382         int isize = BUFSIZE_INDEX(bh->b_size);
1383 
1384         *bhp = bh;
1385         page = (unsigned long) bh->b_data;
1386         page &= PAGE_MASK;
1387         tmp = bh;
1388         do {
1389                 if (!tmp)
1390                         return 0;
1391                 if (tmp->b_count || buffer_protected(tmp) ||
1392                     buffer_dirty(tmp) || buffer_locked(tmp) || tmp->b_wait)
1393                         return 0;
1394                 if (priority && buffer_touched(tmp))
1395                         return 0;
1396                 tmp = tmp->b_this_page;
1397         } while (tmp != bh);
1398         tmp = bh;
1399         do {
1400                 p = tmp;
1401                 tmp = tmp->b_this_page;
1402                 nr_buffers--;
1403                 nr_buffers_size[isize]--;
1404                 if (p == *bhp)
1405                   {
1406                     *bhp = p->b_prev_free;
1407                     if (p == *bhp) /* Was this the last in the list? */
1408                       *bhp = NULL;
1409                   }
1410                 remove_from_queues(p);
1411                 put_unused_buffer_head(p);
1412         } while (tmp != bh);
1413         buffermem -= PAGE_SIZE;
1414         mem_map[MAP_NR(page)].buffers = NULL;
1415         free_page(page);
1416         return !mem_map[MAP_NR(page)].count;
1417 }
1418 
1419 /* Age buffers on a given page, according to whether they have been
1420    visited recently or not. */
1421 static inline void age_buffer(struct buffer_head *bh)
     /* [previous][next][first][last][top][bottom][index][help] */
1422 {
1423         struct buffer_head *tmp = bh;
1424         int touched = 0;
1425 
1426         /*
1427          * When we age a page, we mark all other buffers in the page
1428          * with the "has_aged" flag.  Then, when these aliased buffers
1429          * come up for aging, we skip them until next pass.  This
1430          * ensures that a page full of multiple buffers only gets aged
1431          * once per pass through the lru lists. 
1432          */
1433         if (clear_bit(BH_Has_aged, &bh->b_state))
1434                 return;
1435         
1436         do {
1437                 touched |= clear_bit(BH_Touched, &tmp->b_state);
1438                 tmp = tmp->b_this_page;
1439                 set_bit(BH_Has_aged, &tmp->b_state);
1440         } while (tmp != bh);
1441         clear_bit(BH_Has_aged, &bh->b_state);
1442 
1443         if (touched) 
1444                 touch_page(mem_map + MAP_NR((unsigned long) bh->b_data));
1445         else
1446                 age_page(mem_map + MAP_NR((unsigned long) bh->b_data));
1447 }
1448 
1449 /*
1450  * Consult the load average for buffers and decide whether or not
1451  * we should shrink the buffers of one size or not.  If we decide yes,
1452  * do it and return 1.  Else return 0.  Do not attempt to shrink size
1453  * that is specified.
1454  *
1455  * I would prefer not to use a load average, but the way things are now it
1456  * seems unavoidable.  The way to get rid of it would be to force clustering
1457  * universally, so that when we reclaim buffers we always reclaim an entire
1458  * page.  Doing this would mean that we all need to move towards QMAGIC.
1459  */
1460 
1461 static int maybe_shrink_lav_buffers(int size)
     /* [previous][next][first][last][top][bottom][index][help] */
1462 {          
1463         int nlist;
1464         int isize;
1465         int total_lav, total_n_buffers, n_sizes;
1466         
1467         /* Do not consider the shared buffers since they would not tend
1468            to have getblk called very often, and this would throw off
1469            the lav.  They are not easily reclaimable anyway (let the swapper
1470            make the first move). */
1471   
1472         total_lav = total_n_buffers = n_sizes = 0;
1473         for(nlist = 0; nlist < NR_SIZES; nlist++)
1474          {
1475                  total_lav += buffers_lav[nlist];
1476                  if(nr_buffers_size[nlist]) n_sizes++;
1477                  total_n_buffers += nr_buffers_size[nlist];
1478                  total_n_buffers -= nr_buffers_st[nlist][BUF_SHARED]; 
1479          }
1480         
1481         /* See if we have an excessive number of buffers of a particular
1482            size - if so, victimize that bunch. */
1483   
1484         isize = (size ? BUFSIZE_INDEX(size) : -1);
1485         
1486         if (n_sizes > 1)
1487                  for(nlist = 0; nlist < NR_SIZES; nlist++)
1488                   {
1489                           if(nlist == isize) continue;
1490                           if(nr_buffers_size[nlist] &&
1491                              bdf_prm.b_un.lav_const * buffers_lav[nlist]*total_n_buffers < 
1492                              total_lav * (nr_buffers_size[nlist] - nr_buffers_st[nlist][BUF_SHARED]))
1493                                    if(shrink_specific_buffers(6, bufferindex_size[nlist])) 
1494                                             return 1;
1495                   }
1496         return 0;
1497 }
1498 
1499 /*
1500  * Try to free up some pages by shrinking the buffer-cache
1501  *
1502  * Priority tells the routine how hard to try to shrink the
1503  * buffers: 6 means "don't bother too much", while a value
1504  * of 0 means "we'd better get some free pages now".
1505  *
1506  * "limit" is meant to limit the shrink-action only to pages
1507  * that are in the 0 - limit address range, for DMA re-allocations.
1508  * We ignore that right now.
1509  */
1510 
1511 static int shrink_specific_buffers(unsigned int priority, int size)
     /* [previous][next][first][last][top][bottom][index][help] */
1512 {
1513         struct buffer_head *bh;
1514         int nlist;
1515         int i, isize, isize1;
1516 
1517 #ifdef DEBUG
1518         if(size) printk("Shrinking buffers of size %d\n", size);
1519 #endif
1520         /* First try the free lists, and see if we can get a complete page
1521            from here */
1522         isize1 = (size ? BUFSIZE_INDEX(size) : -1);
1523 
1524         for(isize = 0; isize<NR_SIZES; isize++){
1525                 if(isize1 != -1 && isize1 != isize) continue;
1526                 bh = free_list[isize];
1527                 if(!bh) continue;
1528                 for (i=0 ; !i || bh != free_list[isize]; bh = bh->b_next_free, i++) {
1529                         if (bh->b_count || buffer_protected(bh) ||
1530                             !bh->b_this_page)
1531                                  continue;
1532                         if (!age_of((unsigned long) bh->b_data) &&
1533                             try_to_free_buffer(bh, &bh, 6))
1534                                  return 1;
1535                         if(!bh) break;
1536                         /* Some interrupt must have used it after we
1537                            freed the page.  No big deal - keep looking */
1538                 }
1539         }
1540         
1541         /* Not enough in the free lists, now try the lru list */
1542         
1543         for(nlist = 0; nlist < NR_LIST; nlist++) {
1544         repeat1:
1545                 if(priority > 2 && nlist == BUF_SHARED) continue;
1546                 i = nr_buffers_type[nlist];
1547                 i = ((BUFFEROUT_WEIGHT * i) >> 10) >> priority;
1548                 for ( ; i > 0; i-- ) {
1549                         bh = next_to_age[nlist];
1550                         if (!bh)
1551                                 break;
1552                         next_to_age[nlist] = bh->b_next_free;
1553 
1554                         /* First, age the buffer. */
1555                         age_buffer(bh);
1556                         /* We may have stalled while waiting for I/O
1557                            to complete. */
1558                         if(bh->b_list != nlist) goto repeat1;
1559                         if (bh->b_count || buffer_protected(bh) ||
1560                             !bh->b_this_page)
1561                                  continue;
1562                         if(size && bh->b_size != size) continue;
1563                         if (buffer_locked(bh))
1564                                  if (priority)
1565                                           continue;
1566                                  else
1567                                           wait_on_buffer(bh);
1568                         if (buffer_dirty(bh)) {
1569                                 bh->b_count++;
1570                                 bh->b_flushtime = 0;
1571                                 ll_rw_block(WRITEA, 1, &bh);
1572                                 bh->b_count--;
1573                                 continue;
1574                         }
1575                         /* At priority 6, only consider really old
1576                            (age==0) buffers for reclaiming.  At
1577                            priority 0, consider any buffers. */
1578                         if ((age_of((unsigned long) bh->b_data) >>
1579                              (6-priority)) > 0)
1580                                 continue;                               
1581                         if (try_to_free_buffer(bh, &bh, 0))
1582                                  return 1;
1583                         if(!bh) break;
1584                 }
1585         }
1586         return 0;
1587 }
1588 
1589 
1590 /* ================== Debugging =================== */
1591 
1592 void show_buffers(void)
     /* [previous][next][first][last][top][bottom][index][help] */
1593 {
1594         struct buffer_head * bh;
1595         int found = 0, locked = 0, dirty = 0, used = 0, lastused = 0;
1596         int protected = 0;
1597         int shared;
1598         int nlist, isize;
1599 
1600         printk("Buffer memory:   %6dkB\n",buffermem>>10);
1601         printk("Buffer heads:    %6d\n",nr_buffer_heads);
1602         printk("Buffer blocks:   %6d\n",nr_buffers);
1603 
1604         for(nlist = 0; nlist < NR_LIST; nlist++) {
1605           shared = found = locked = dirty = used = lastused = protected = 0;
1606           bh = lru_list[nlist];
1607           if(!bh) continue;
1608           do {
1609                 found++;
1610                 if (buffer_locked(bh))
1611                         locked++;
1612                 if (buffer_protected(bh))
1613                         protected++;
1614                 if (buffer_dirty(bh))
1615                         dirty++;
1616                 if (mem_map[MAP_NR(((unsigned long) bh->b_data))].count != 1)
1617                         shared++;
1618                 if (bh->b_count)
1619                         used++, lastused = found;
1620                 bh = bh->b_next_free;
1621           } while (bh != lru_list[nlist]);
1622           printk("Buffer[%d] mem: %d buffers, %d used (last=%d), "
1623                  "%d locked, %d protected, %d dirty %d shrd\n",
1624                  nlist, found, used, lastused,
1625                  locked, protected, dirty, shared);
1626         };
1627         printk("Size    [LAV]     Free  Clean  Unshar     Lck    Lck1   Dirty  Shared \n");
1628         for(isize = 0; isize<NR_SIZES; isize++){
1629                 printk("%5d [%5d]: %7d ", bufferindex_size[isize],
1630                        buffers_lav[isize], nr_free[isize]);
1631                 for(nlist = 0; nlist < NR_LIST; nlist++)
1632                          printk("%7d ", nr_buffers_st[isize][nlist]);
1633                 printk("\n");
1634         }
1635 }
1636 
1637 
1638 /* ====================== Cluster patches for ext2 ==================== */
1639 
1640 /*
1641  * try_to_reassign() checks if all the buffers on this particular page
1642  * are unused, and reassign to a new cluster them if this is true.
1643  */
1644 static inline int try_to_reassign(struct buffer_head * bh, struct buffer_head ** bhp,
     /* [previous][next][first][last][top][bottom][index][help] */
1645                            kdev_t dev, unsigned int starting_block)
1646 {
1647         unsigned long page;
1648         struct buffer_head * tmp, * p;
1649 
1650         *bhp = bh;
1651         page = (unsigned long) bh->b_data;
1652         page &= PAGE_MASK;
1653         if(mem_map[MAP_NR(page)].count != 1) return 0;
1654         tmp = bh;
1655         do {
1656                 if (!tmp)
1657                          return 0;
1658                 
1659                 if (tmp->b_count || buffer_protected(tmp) ||
1660                     buffer_dirty(tmp) || buffer_locked(tmp))
1661                          return 0;
1662                 tmp = tmp->b_this_page;
1663         } while (tmp != bh);
1664         tmp = bh;
1665         
1666         while((unsigned long) tmp->b_data & (PAGE_SIZE - 1)) 
1667                  tmp = tmp->b_this_page;
1668         
1669         /* This is the buffer at the head of the page */
1670         bh = tmp;
1671         do {
1672                 p = tmp;
1673                 tmp = tmp->b_this_page;
1674                 remove_from_queues(p);
1675                 p->b_dev = dev;
1676                 mark_buffer_uptodate(p, 0);
1677                 clear_bit(BH_Req, &p->b_state);
1678                 p->b_blocknr = starting_block++;
1679                 insert_into_queues(p);
1680         } while (tmp != bh);
1681         return 1;
1682 }
1683 
1684 /*
1685  * Try to find a free cluster by locating a page where
1686  * all of the buffers are unused.  We would like this function
1687  * to be atomic, so we do not call anything that might cause
1688  * the process to sleep.  The priority is somewhat similar to
1689  * the priority used in shrink_buffers.
1690  * 
1691  * My thinking is that the kernel should end up using whole
1692  * pages for the buffer cache as much of the time as possible.
1693  * This way the other buffers on a particular page are likely
1694  * to be very near each other on the free list, and we will not
1695  * be expiring data prematurely.  For now we only cannibalize buffers
1696  * of the same size to keep the code simpler.
1697  */
1698 static int reassign_cluster(kdev_t dev, 
     /* [previous][next][first][last][top][bottom][index][help] */
1699                      unsigned int starting_block, int size)
1700 {
1701         struct buffer_head *bh;
1702         int isize = BUFSIZE_INDEX(size);
1703         int i;
1704 
1705         /* We want to give ourselves a really good shot at generating
1706            a cluster, and since we only take buffers from the free
1707            list, we "overfill" it a little. */
1708 
1709         while(nr_free[isize] < 32) refill_freelist(size);
1710 
1711         bh = free_list[isize];
1712         if(bh)
1713                  for (i=0 ; !i || bh != free_list[isize] ; bh = bh->b_next_free, i++) {
1714                          if (!bh->b_this_page)  continue;
1715                          if (try_to_reassign(bh, &bh, dev, starting_block))
1716                                  return 4;
1717                  }
1718         return 0;
1719 }
1720 
1721 /* This function tries to generate a new cluster of buffers
1722  * from a new page in memory.  We should only do this if we have
1723  * not expanded the buffer cache to the maximum size that we allow.
1724  */
1725 static unsigned long try_to_generate_cluster(kdev_t dev, int block, int size)
     /* [previous][next][first][last][top][bottom][index][help] */
1726 {
1727         struct buffer_head * bh, * tmp, * arr[MAX_BUF_PER_PAGE];
1728         int isize = BUFSIZE_INDEX(size);
1729         unsigned long offset;
1730         unsigned long page;
1731         int nblock;
1732 
1733         page = get_free_page(GFP_NOBUFFER);
1734         if(!page) return 0;
1735 
1736         bh = create_buffers(page, size);
1737         if (!bh) {
1738                 free_page(page);
1739                 return 0;
1740         };
1741         nblock = block;
1742         for (offset = 0 ; offset < PAGE_SIZE ; offset += size) {
1743                 if (find_buffer(dev, nblock++, size))
1744                          goto not_aligned;
1745         }
1746         tmp = bh;
1747         nblock = 0;
1748         while (1) {
1749                 arr[nblock++] = bh;
1750                 bh->b_count = 1;
1751                 bh->b_flushtime = 0;
1752                 bh->b_state = 0;
1753                 bh->b_dev = dev;
1754                 bh->b_list = BUF_CLEAN;
1755                 bh->b_blocknr = block++;
1756                 nr_buffers++;
1757                 nr_buffers_size[isize]++;
1758                 insert_into_queues(bh);
1759                 if (bh->b_this_page)
1760                         bh = bh->b_this_page;
1761                 else
1762                         break;
1763         }
1764         buffermem += PAGE_SIZE;
1765         mem_map[MAP_NR(page)].buffers = bh;
1766         bh->b_this_page = tmp;
1767         while (nblock-- > 0)
1768                 brelse(arr[nblock]);
1769         return 4; /* ?? */
1770 not_aligned:
1771         while ((tmp = bh) != NULL) {
1772                 bh = bh->b_this_page;
1773                 put_unused_buffer_head(tmp);
1774         }
1775         free_page(page);
1776         return 0;
1777 }
1778 
1779 unsigned long generate_cluster(kdev_t dev, int b[], int size)
     /* [previous][next][first][last][top][bottom][index][help] */
1780 {
1781         int i, offset;
1782         
1783         for (i = 0, offset = 0 ; offset < PAGE_SIZE ; i++, offset += size) {
1784                 if(i && b[i]-1 != b[i-1]) return 0;  /* No need to cluster */
1785                 if(find_buffer(dev, b[i], size)) return 0;
1786         };
1787 
1788         /* OK, we have a candidate for a new cluster */
1789         
1790         /* See if one size of buffer is over-represented in the buffer cache,
1791            if so reduce the numbers of buffers */
1792         if(maybe_shrink_lav_buffers(size))
1793          {
1794                  int retval;
1795                  retval = try_to_generate_cluster(dev, b[0], size);
1796                  if(retval) return retval;
1797          };
1798         
1799         if (nr_free_pages > min_free_pages*2) 
1800                  return try_to_generate_cluster(dev, b[0], size);
1801         else
1802                  return reassign_cluster(dev, b[0], size);
1803 }
1804 
1805 
1806 /* ===================== Init ======================= */
1807 
1808 /*
1809  * This initializes the initial buffer free list.  nr_buffers_type is set
1810  * to one less the actual number of buffers, as a sop to backwards
1811  * compatibility --- the old code did this (I think unintentionally,
1812  * but I'm not sure), and programs in the ps package expect it.
1813  *                                      - TYT 8/30/92
1814  */
1815 void buffer_init(void)
     /* [previous][next][first][last][top][bottom][index][help] */
1816 {
1817         int i;
1818         int isize = BUFSIZE_INDEX(BLOCK_SIZE);
1819         long memsize = MAP_NR(high_memory) << PAGE_SHIFT;
1820 
1821         if (memsize >= 4*1024*1024) {
1822                 if(memsize >= 16*1024*1024)
1823                          nr_hash = 16381;
1824                 else
1825                          nr_hash = 4093;
1826         } else {
1827                 nr_hash = 997;
1828         };
1829         
1830         hash_table = (struct buffer_head **) vmalloc(nr_hash * 
1831                                                      sizeof(struct buffer_head *));
1832 
1833 
1834         for (i = 0 ; i < nr_hash ; i++)
1835                 hash_table[i] = NULL;
1836         lru_list[BUF_CLEAN] = 0;
1837         grow_buffers(GFP_KERNEL, BLOCK_SIZE);
1838         if (!free_list[isize])
1839                 panic("VFS: Unable to initialize buffer free list!");
1840         return;
1841 }
1842 
1843 
1844 /* ====================== bdflush support =================== */
1845 
1846 /* This is a simple kernel daemon, whose job it is to provide a dynamic
1847  * response to dirty buffers.  Once this process is activated, we write back
1848  * a limited number of buffers to the disks and then go back to sleep again.
1849  */
1850 struct wait_queue * bdflush_wait = NULL;
1851 struct wait_queue * bdflush_done = NULL;
1852 
1853 static void wakeup_bdflush(int wait)
     /* [previous][next][first][last][top][bottom][index][help] */
1854 {
1855         wake_up(&bdflush_wait);
1856         if (wait) {
1857                 run_task_queue(&tq_disk);
1858                 sleep_on(&bdflush_done);
1859         }
1860 }
1861 
1862 
1863 /* 
1864  * Here we attempt to write back old buffers.  We also try and flush inodes 
1865  * and supers as well, since this function is essentially "update", and 
1866  * otherwise there would be no way of ensuring that these quantities ever 
1867  * get written back.  Ideally, we would have a timestamp on the inodes
1868  * and superblocks so that we could write back only the old ones as well
1869  */
1870 
1871 asmlinkage int sync_old_buffers(void)
     /* [previous][next][first][last][top][bottom][index][help] */
1872 {
1873         int i, isize;
1874         int ndirty, nwritten;
1875         int nlist;
1876         int ncount;
1877         struct buffer_head * bh, *next;
1878 
1879         sync_supers(0);
1880         sync_inodes(0);
1881 
1882         ncount = 0;
1883 #ifdef DEBUG
1884         for(nlist = 0; nlist < NR_LIST; nlist++)
1885 #else
1886         for(nlist = BUF_DIRTY; nlist <= BUF_DIRTY; nlist++)
1887 #endif
1888         {
1889                 ndirty = 0;
1890                 nwritten = 0;
1891         repeat:
1892                 bh = lru_list[nlist];
1893                 if(bh) 
1894                          for (i = nr_buffers_type[nlist]; i-- > 0; bh = next) {
1895                                  /* We may have stalled while waiting for I/O to complete. */
1896                                  if(bh->b_list != nlist) goto repeat;
1897                                  next = bh->b_next_free;
1898                                  if(!lru_list[nlist]) {
1899                                          printk("Dirty list empty %d\n", i);
1900                                          break;
1901                                  }
1902                                  
1903                                  /* Clean buffer on dirty list?  Refile it */
1904                                  if (nlist == BUF_DIRTY && !buffer_dirty(bh) && !buffer_locked(bh))
1905                                   {
1906                                           refile_buffer(bh);
1907                                           continue;
1908                                   }
1909                                  
1910                                  if (buffer_locked(bh) || !buffer_dirty(bh))
1911                                           continue;
1912                                  ndirty++;
1913                                  if(bh->b_flushtime > jiffies) continue;
1914                                  nwritten++;
1915                                  bh->b_count++;
1916                                  bh->b_flushtime = 0;
1917 #ifdef DEBUG
1918                                  if(nlist != BUF_DIRTY) ncount++;
1919 #endif
1920                                  ll_rw_block(WRITE, 1, &bh);
1921                                  bh->b_count--;
1922                          }
1923         }
1924 #ifdef DEBUG
1925         if (ncount) printk("sync_old_buffers: %d dirty buffers not on dirty list\n", ncount);
1926         printk("Wrote %d/%d buffers\n", nwritten, ndirty);
1927 #endif
1928         
1929         /* We assume that we only come through here on a regular
1930            schedule, like every 5 seconds.  Now update load averages.  
1931            Shift usage counts to prevent overflow. */
1932         for(isize = 0; isize<NR_SIZES; isize++){
1933                 CALC_LOAD(buffers_lav[isize], bdf_prm.b_un.lav_const, buffer_usage[isize]);
1934                 buffer_usage[isize] = 0;
1935         }
1936         return 0;
1937 }
1938 
1939 
1940 /* This is the interface to bdflush.  As we get more sophisticated, we can
1941  * pass tuning parameters to this "process", to adjust how it behaves. 
1942  * We would want to verify each parameter, however, to make sure that it 
1943  * is reasonable. */
1944 
1945 asmlinkage int sys_bdflush(int func, long data)
     /* [previous][next][first][last][top][bottom][index][help] */
1946 {
1947         int i, error;
1948 
1949         if (!suser())
1950                 return -EPERM;
1951 
1952         if (func == 1)
1953                  return sync_old_buffers();
1954 
1955         /* Basically func 1 means read param 1, 2 means write param 1, etc */
1956         if (func >= 2) {
1957                 i = (func-2) >> 1;
1958                 if (i < 0 || i >= N_PARAM)
1959                         return -EINVAL;
1960                 if((func & 1) == 0) {
1961                         error = verify_area(VERIFY_WRITE, (void *) data, sizeof(int));
1962                         if (error)
1963                                 return error;
1964                         put_user(bdf_prm.data[i], (int*)data);
1965                         return 0;
1966                 };
1967                 if (data < bdflush_min[i] || data > bdflush_max[i])
1968                         return -EINVAL;
1969                 bdf_prm.data[i] = data;
1970                 return 0;
1971         };
1972 
1973         /* Having func 0 used to launch the actual bdflush and then never
1974         return (unless explicitly killed). We return zero here to 
1975         remain semi-compatible with present update(8) programs. */
1976 
1977         return 0;
1978 }
1979 
1980 /* This is the actual bdflush daemon itself. It used to be started from
1981  * the syscall above, but now we launch it ourselves internally with
1982  * kernel_thread(...)  directly after the first thread in init/main.c */
1983 
1984 int bdflush(void * unused) 
     /* [previous][next][first][last][top][bottom][index][help] */
1985 {
1986         int i;
1987         int ndirty;
1988         int nlist;
1989         int ncount;
1990         struct buffer_head * bh, *next;
1991 
1992         /*
1993          *      We have a bare-bones task_struct, and really should fill
1994          *      in a few more things so "top" and /proc/2/{exe,root,cwd}
1995          *      display semi-sane things. Not real crucial though...  
1996          */
1997 
1998         current->session = 1;
1999         current->pgrp = 1;
2000         sprintf(current->comm, "kflushd");
2001 
2002         /*
2003          *      As a kernel thread we want to tamper with system buffers
2004          *      and other internals and thus be subject to the SMP locking
2005          *      rules. (On a uniprocessor box this does nothing).
2006          */
2007          
2008 #ifdef __SMP__
2009         lock_kernel();
2010         syscall_count++;
2011 #endif
2012                  
2013         for (;;) {
2014 #ifdef DEBUG
2015                 printk("bdflush() activated...");
2016 #endif
2017                 
2018                 ncount = 0;
2019 #ifdef DEBUG
2020                 for(nlist = 0; nlist < NR_LIST; nlist++)
2021 #else
2022                 for(nlist = BUF_DIRTY; nlist <= BUF_DIRTY; nlist++)
2023 #endif
2024                  {
2025                          ndirty = 0;
2026                  repeat:
2027                          bh = lru_list[nlist];
2028                          if(bh) 
2029                                   for (i = nr_buffers_type[nlist]; i-- > 0 && ndirty < bdf_prm.b_un.ndirty; 
2030                                        bh = next) {
2031                                           /* We may have stalled while waiting for I/O to complete. */
2032                                           if(bh->b_list != nlist) goto repeat;
2033                                           next = bh->b_next_free;
2034                                           if(!lru_list[nlist]) {
2035                                                   printk("Dirty list empty %d\n", i);
2036                                                   break;
2037                                           }
2038                                           
2039                                           /* Clean buffer on dirty list?  Refile it */
2040                                           if (nlist == BUF_DIRTY && !buffer_dirty(bh) && !buffer_locked(bh))
2041                                            {
2042                                                    refile_buffer(bh);
2043                                                    continue;
2044                                            }
2045                                           
2046                                           if (buffer_locked(bh) || !buffer_dirty(bh))
2047                                                    continue;
2048                                           /* Should we write back buffers that are shared or not??
2049                                              currently dirty buffers are not shared, so it does not matter */
2050                                           bh->b_count++;
2051                                           ndirty++;
2052                                           bh->b_flushtime = 0;
2053                                           ll_rw_block(WRITE, 1, &bh);
2054 #ifdef DEBUG
2055                                           if(nlist != BUF_DIRTY) ncount++;
2056 #endif
2057                                           bh->b_count--;
2058                                   }
2059                  }
2060 #ifdef DEBUG
2061                 if (ncount) printk("sys_bdflush: %d dirty buffers not on dirty list\n", ncount);
2062                 printk("sleeping again.\n");
2063 #endif
2064                 run_task_queue(&tq_disk);
2065                 wake_up(&bdflush_done);
2066                 
2067                 /* If there are still a lot of dirty buffers around, skip the sleep
2068                    and flush some more */
2069                 
2070                 if(nr_buffers_type[BUF_DIRTY] <= (nr_buffers - nr_buffers_type[BUF_SHARED]) * 
2071                    bdf_prm.b_un.nfract/100) {
2072                         current->signal = 0;
2073                         interruptible_sleep_on(&bdflush_wait);
2074                 }
2075         }
2076 }
2077 
2078 
2079 /*
2080  * Overrides for Emacs so that we follow Linus's tabbing style.
2081  * Emacs will notice this stuff at the end of the file and automatically
2082  * adjust the settings for this buffer only.  This must remain at the end
2083  * of the file.
2084  * ---------------------------------------------------------------------------
2085  * Local variables:
2086  * c-indent-level: 8
2087  * c-brace-imaginary-offset: 0
2088  * c-brace-offset: -8
2089  * c-argdecl-indent: 8
2090  * c-label-offset: -8
2091  * c-continued-statement-offset: 8
2092  * c-continued-brace-offset: 0
2093  * End:
2094  */

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