Add missing commit for the VM load heuristic and page allocation rate
[dragonfly.git] / sys / vfs / gnu / ext2fs / ext2_linux_balloc.c
1 /*
2  *  modified for Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  *
7  * $FreeBSD: src/sys/gnu/ext2fs/ext2_linux_balloc.c,v 1.11.2.3 2001/08/14 18:03:19 gallatin Exp $
8  * $DragonFly: src/sys/vfs/gnu/ext2fs/ext2_linux_balloc.c,v 1.6 2006/01/13 21:09:27 swildner Exp $
9  */
10 /*
11  *  linux/fs/ext2/balloc.c
12  *
13  * Copyright (C) 1992, 1993, 1994, 1995
14  * Remy Card (card@masi.ibp.fr)
15  * Laboratoire MASI - Institut Blaise Pascal
16  * Universite Pierre et Marie Curie (Paris VI)
17  *
18  *  Enhanced block allocation by Stephen Tweedie (sct@dcs.ed.ac.uk), 1993
19  */
20
21 /*
22  * The free blocks are managed by bitmaps.  A file system contains several
23  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
24  * block for inodes, N blocks for the inode table and data blocks.
25  *
26  * The file system contains group descriptors which are located after the
27  * super block.  Each descriptor contains the number of the bitmap block and
28  * the free blocks count in the block.  The descriptors are loaded in memory
29  * when a file system is mounted (see ext2_read_super).
30  */
31
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/buf.h>
35 #include <sys/proc.h>
36 #include <sys/mount.h>
37 #include <sys/vnode.h>
38 #include <sys/buf2.h>
39 #include <sys/thread2.h>
40
41 #include <vfs/ufs/quota.h>
42 #include <vfs/ufs/ufsmount.h>
43 #include "ext2_extern.h"
44 #include "ext2_fs.h"
45 #include "ext2_fs_sb.h"
46 #include "fs.h"
47
48 #ifdef __i386__
49 #include "i386-bitops.h"
50 #elif defined (__alpha__)
51 #include "alpha-bitops.h"
52 #else
53 #error Provide an bitops.h file, please !
54 #endif
55
56 #define in_range(b, first, len)         ((b) >= (first) && (b) <= (first) + (len) - 1)
57
58 /* got rid of get_group_desc since it can already be found in 
59  * ext2_linux_ialloc.c
60  */
61
62 static void
63 read_block_bitmap(struct mount *mp, unsigned int block_group,
64                   unsigned long bitmap_nr)
65 {
66         struct ext2_sb_info *sb = VFSTOUFS(mp)->um_e2fs;
67         struct ext2_group_desc * gdp;
68         struct buffer_head * bh;
69         int    error;
70         
71         gdp = get_group_desc (mp, block_group, NULL);
72         if ((error = bread (VFSTOUFS(mp)->um_devvp, 
73                 fsbtodb(sb, gdp->bg_block_bitmap),sb->s_blocksize, &bh)) != 0)
74                 panic ( "read_block_bitmap: "
75                             "Cannot read block bitmap - "
76                             "block_group = %d, block_bitmap = %lu",
77                             block_group, (unsigned long) gdp->bg_block_bitmap);
78         sb->s_block_bitmap_number[bitmap_nr] = block_group;
79         sb->s_block_bitmap[bitmap_nr] = bh;
80         LCK_BUF(bh)
81 }
82
83 /*
84  * load_block_bitmap loads the block bitmap for a blocks group
85  *
86  * It maintains a cache for the last bitmaps loaded.  This cache is managed
87  * with a LRU algorithm.
88  *
89  * Notes:
90  * 1/ There is one cache per mounted file system.
91  * 2/ If the file system contains less than EXT2_MAX_GROUP_LOADED groups,
92  *    this function reads the bitmap without maintaining a LRU cache.
93  */
94 static int
95 load__block_bitmap(struct mount *mp, unsigned int block_group)
96 {
97         int i, j;
98         struct ext2_sb_info *sb = VFSTOUFS(mp)->um_e2fs;
99         unsigned long block_bitmap_number;
100         struct buffer_head * block_bitmap;
101
102         if (block_group >= sb->s_groups_count)
103                 panic ( "load_block_bitmap: "
104                             "block_group >= groups_count - "
105                             "block_group = %d, groups_count = %lu",
106                             block_group, sb->s_groups_count);
107
108         if (sb->s_groups_count <= EXT2_MAX_GROUP_LOADED) {
109                 if (sb->s_block_bitmap[block_group]) {
110                         if (sb->s_block_bitmap_number[block_group] !=
111                             block_group)
112                                 panic ( "load_block_bitmap: "
113                                             "block_group != block_bitmap_number");
114                         else
115                                 return block_group;
116                 } else {
117                         read_block_bitmap (mp, block_group, block_group);
118                         return block_group;
119                 }
120         }
121
122         for (i = 0; i < sb->s_loaded_block_bitmaps &&
123                     sb->s_block_bitmap_number[i] != block_group; i++)
124                 ;
125         if (i < sb->s_loaded_block_bitmaps &&
126             sb->s_block_bitmap_number[i] == block_group) {
127                 block_bitmap_number = sb->s_block_bitmap_number[i];
128                 block_bitmap = sb->s_block_bitmap[i];
129                 for (j = i; j > 0; j--) {
130                         sb->s_block_bitmap_number[j] =
131                                 sb->s_block_bitmap_number[j - 1];
132                         sb->s_block_bitmap[j] =
133                                 sb->s_block_bitmap[j - 1];
134                 }
135                 sb->s_block_bitmap_number[0] = block_bitmap_number;
136                 sb->s_block_bitmap[0] = block_bitmap;
137         } else {
138                 if (sb->s_loaded_block_bitmaps < EXT2_MAX_GROUP_LOADED)
139                         sb->s_loaded_block_bitmaps++;
140                 else
141                         ULCK_BUF(sb->s_block_bitmap[EXT2_MAX_GROUP_LOADED - 1])
142
143                 for (j = sb->s_loaded_block_bitmaps - 1; j > 0;  j--) {
144                         sb->s_block_bitmap_number[j] =
145                                 sb->s_block_bitmap_number[j - 1];
146                         sb->s_block_bitmap[j] =
147                                 sb->s_block_bitmap[j - 1];
148                 }
149                 read_block_bitmap (mp, block_group, 0);
150         }
151         return 0;
152 }
153
154 static __inline int
155 load_block_bitmap(struct mount * mp, unsigned int block_group)
156 {
157         struct ext2_sb_info *sb = VFSTOUFS(mp)->um_e2fs;
158         if (sb->s_loaded_block_bitmaps > 0 &&
159             sb->s_block_bitmap_number[0] == block_group)
160                 return 0;
161         
162         if (sb->s_groups_count <= EXT2_MAX_GROUP_LOADED && 
163             sb->s_block_bitmap_number[block_group] == block_group &&
164             sb->s_block_bitmap[block_group]) 
165                 return block_group;
166
167         return load__block_bitmap (mp, block_group);
168 }
169
170 void
171 ext2_free_blocks(struct mount * mp, unsigned long block,
172                  unsigned long count)
173 {
174         struct ext2_sb_info *sb = VFSTOUFS(mp)->um_e2fs;
175         struct buffer_head * bh;
176         struct buffer_head * bh2;
177         unsigned long block_group;
178         unsigned long bit;
179         unsigned long i;
180         int bitmap_nr;
181         struct ext2_group_desc * gdp;
182         struct ext2_super_block * es = sb->s_es;
183
184         if (!sb) {
185                 printf ("ext2_free_blocks: nonexistent device");
186                 return;
187         }
188         lock_super (VFSTOUFS(mp)->um_devvp);
189         if (block < es->s_first_data_block || 
190             (block + count) > es->s_blocks_count) {
191                 printf ( "ext2_free_blocks: "
192                             "Freeing blocks not in datazone - "
193                             "block = %lu, count = %lu", block, count);
194                 unlock_super (VFSTOUFS(mp)->um_devvp);
195                 return;
196         }
197
198         ext2_debug ("freeing blocks %lu to %lu\n", block, block+count-1);
199
200         block_group = (block - es->s_first_data_block) /
201                       EXT2_BLOCKS_PER_GROUP(sb);
202         bit = (block - es->s_first_data_block) % EXT2_BLOCKS_PER_GROUP(sb);
203         if (bit + count > EXT2_BLOCKS_PER_GROUP(sb))
204                 panic ( "ext2_free_blocks: "
205                             "Freeing blocks across group boundary - "
206                             "Block = %lu, count = %lu",
207                             block, count);
208         bitmap_nr = load_block_bitmap (mp, block_group);
209         bh = sb->s_block_bitmap[bitmap_nr];
210         gdp = get_group_desc (mp, block_group, &bh2);
211
212         if (/* test_opt (sb, CHECK_STRICT) &&   assume always strict ! */
213             (in_range (gdp->bg_block_bitmap, block, count) ||
214              in_range (gdp->bg_inode_bitmap, block, count) ||
215              in_range (block, gdp->bg_inode_table,
216                        sb->s_itb_per_group) ||
217              in_range (block + count - 1, gdp->bg_inode_table,
218                        sb->s_itb_per_group)))
219                 panic ( "ext2_free_blocks: "
220                             "Freeing blocks in system zones - "
221                             "Block = %lu, count = %lu",
222                             block, count);
223
224         for (i = 0; i < count; i++) {
225                 if (!clear_bit (bit + i, bh->b_data))
226                         printf ("ext2_free_blocks: "
227                                       "bit already cleared for block %lu", 
228                                       block);
229                 else {
230                         gdp->bg_free_blocks_count++;
231                         es->s_free_blocks_count++;
232                 }
233         }
234
235         mark_buffer_dirty(bh2);
236         mark_buffer_dirty(bh);
237 /****
238         if (sb->s_flags & MS_SYNCHRONOUS) {
239                 ll_rw_block (WRITE, 1, &bh);
240                 wait_on_buffer (bh);
241         }
242 ****/
243         sb->s_dirt = 1;
244         unlock_super (VFSTOUFS(mp)->um_devvp);
245         return;
246 }
247
248 /*
249  * ext2_new_block uses a goal block to assist allocation.  If the goal is
250  * free, or there is a free block within 32 blocks of the goal, that block
251  * is allocated.  Otherwise a forward search is made for a free block; within 
252  * each block group the search first looks for an entire free byte in the block
253  * bitmap, and then for any free bit if that fails.
254  */
255 int
256 ext2_new_block(struct mount * mp, unsigned long goal,
257                u_int32_t * prealloc_count,
258                u_int32_t * prealloc_block)
259 {
260         struct ext2_sb_info *sb = VFSTOUFS(mp)->um_e2fs;
261         struct buffer_head * bh;
262         struct buffer_head * bh2;
263         char * p, * r;
264         int i, j, k, tmp;
265         int bitmap_nr;
266         struct ext2_group_desc * gdp;
267         struct ext2_super_block * es = sb->s_es;
268
269 #ifdef EXT2FS_DEBUG
270         static int goal_hits = 0, goal_attempts = 0;
271 #endif
272         if (!sb) {
273                 printf ("ext2_new_block: nonexistent device");
274                 return 0;
275         }
276         lock_super (VFSTOUFS(mp)->um_devvp);
277
278         ext2_debug ("goal=%lu.\n", goal);
279
280 repeat:
281         /*
282          * First, test whether the goal block is free.
283          */
284         if (goal < es->s_first_data_block || goal >= es->s_blocks_count)
285                 goal = es->s_first_data_block;
286         i = (goal - es->s_first_data_block) / EXT2_BLOCKS_PER_GROUP(sb);
287         gdp = get_group_desc (mp, i, &bh2);
288         if (gdp->bg_free_blocks_count > 0) {
289                 j = ((goal - es->s_first_data_block) % EXT2_BLOCKS_PER_GROUP(sb));
290 #ifdef EXT2FS_DEBUG
291                 if (j)
292                         goal_attempts++;
293 #endif
294                 bitmap_nr = load_block_bitmap (mp, i);
295                 bh = sb->s_block_bitmap[bitmap_nr];
296
297                 ext2_debug ("goal is at %d:%d.\n", i, j); 
298
299                 if (!test_bit(j, bh->b_data)) {
300 #ifdef EXT2FS_DEBUG
301                         goal_hits++;
302                         ext2_debug ("goal bit allocated.\n");
303 #endif
304                         goto got_block;
305                 }
306                 if (j) {
307                         /*
308                          * The goal was occupied; search forward for a free 
309                          * block within the next XX blocks.
310                          *
311                          * end_goal is more or less random, but it has to be
312                          * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
313                          * next 64-bit boundary is simple..
314                          */
315                         int end_goal = (j + 63) & ~63;
316                         j = find_next_zero_bit(bh->b_data, end_goal, j);
317                         if (j < end_goal)
318                                 goto got_block;
319                 }
320         
321                 ext2_debug ("Bit not found near goal\n");
322
323                 /*
324                  * There has been no free block found in the near vicinity
325                  * of the goal: do a search forward through the block groups,
326                  * searching in each group first for an entire free byte in
327                  * the bitmap and then for any free bit.
328                  * 
329                  * Search first in the remainder of the current group; then,
330                  * cyclicly search through the rest of the groups.
331                  */
332                 p = ((char *) bh->b_data) + (j >> 3);
333                 r = memscan(p, 0, (EXT2_BLOCKS_PER_GROUP(sb) - j + 7) >> 3);
334                 k = (r - ((char *) bh->b_data)) << 3;
335                 if (k < EXT2_BLOCKS_PER_GROUP(sb)) {
336                         j = k;
337                         goto search_back;
338                 }
339                 k = find_next_zero_bit ((unsigned long *) bh->b_data, 
340                                         EXT2_BLOCKS_PER_GROUP(sb),
341                                         j);
342                 if (k < EXT2_BLOCKS_PER_GROUP(sb)) {
343                         j = k;
344                         goto got_block;
345                 }
346         }
347
348         ext2_debug ("Bit not found in block group %d.\n", i); 
349
350         /*
351          * Now search the rest of the groups.  We assume that 
352          * i and gdp correctly point to the last group visited.
353          */
354         for (k = 0; k < sb->s_groups_count; k++) {
355                 i++;
356                 if (i >= sb->s_groups_count)
357                         i = 0;
358                 gdp = get_group_desc (mp, i, &bh2);
359                 if (gdp->bg_free_blocks_count > 0)
360                         break;
361         }
362         if (k >= sb->s_groups_count) {
363                 unlock_super (VFSTOUFS(mp)->um_devvp);
364                 return 0;
365         }
366         bitmap_nr = load_block_bitmap (mp, i);
367         bh = sb->s_block_bitmap[bitmap_nr];
368         r = memscan(bh->b_data, 0, EXT2_BLOCKS_PER_GROUP(sb) >> 3);
369         j = (r - bh->b_data) << 3;
370
371         if (j < EXT2_BLOCKS_PER_GROUP(sb))
372                 goto search_back;
373         else
374                 j = find_first_zero_bit ((unsigned long *) bh->b_data,
375                                          EXT2_BLOCKS_PER_GROUP(sb));
376         if (j >= EXT2_BLOCKS_PER_GROUP(sb)) {
377                 printf ( "ext2_new_block: "
378                          "Free blocks count corrupted for block group %d", i);
379                 unlock_super (VFSTOUFS(mp)->um_devvp);
380                 return 0;
381         }
382
383 search_back:
384         /* 
385          * We have succeeded in finding a free byte in the block
386          * bitmap.  Now search backwards up to 7 bits to find the
387          * start of this group of free blocks.
388          */
389         for (k = 0; k < 7 && j > 0 && !test_bit (j - 1, bh->b_data); k++, j--);
390         
391 got_block:
392
393         ext2_debug ("using block group %d(%d)\n", i, gdp->bg_free_blocks_count);
394
395         tmp = j + i * EXT2_BLOCKS_PER_GROUP(sb) + es->s_first_data_block;
396
397         if (/* test_opt (sb, CHECK_STRICT) && we are always strict. */
398             (tmp == gdp->bg_block_bitmap ||
399              tmp == gdp->bg_inode_bitmap ||
400              in_range (tmp, gdp->bg_inode_table, sb->s_itb_per_group)))
401                 panic ( "ext2_new_block: "
402                             "Allocating block in system zone - "
403                             "%dth block = %u in group %u", j, tmp, i);
404
405         if (set_bit (j, bh->b_data)) {
406                 printf ( "ext2_new_block: "
407                          "bit already set for block %d", j);
408                 goto repeat;
409         }
410
411         ext2_debug ("found bit %d\n", j);
412
413         /*
414          * Do block preallocation now if required.
415          */
416 #ifdef EXT2_PREALLOCATE
417         if (prealloc_block) {
418                 *prealloc_count = 0;
419                 *prealloc_block = tmp + 1;
420                 for (k = 1;
421                      k < 8 && (j + k) < EXT2_BLOCKS_PER_GROUP(sb); k++) {
422                         if (set_bit (j + k, bh->b_data))
423                                 break;
424                         (*prealloc_count)++;
425                 }       
426                 gdp->bg_free_blocks_count -= *prealloc_count;
427                 es->s_free_blocks_count -= *prealloc_count;
428                 ext2_debug ("Preallocated a further %lu bits.\n",
429                             *prealloc_count); 
430         }
431 #endif
432
433         j = tmp;
434
435         mark_buffer_dirty(bh);
436 /****
437         if (sb->s_flags & MS_SYNCHRONOUS) {
438                 ll_rw_block (WRITE, 1, &bh);
439                 wait_on_buffer (bh);
440         }
441 ****/
442         if (j >= es->s_blocks_count) {
443                 printf ( "ext2_new_block: "
444                             "block >= blocks count - "
445                             "block_group = %d, block=%d", i, j);
446                 unlock_super (VFSTOUFS(mp)->um_devvp);
447                 return 0;
448         }
449
450         ext2_debug ("allocating block %d. "
451                     "Goal hits %d of %d.\n", j, goal_hits, goal_attempts);
452
453         gdp->bg_free_blocks_count--;
454         mark_buffer_dirty(bh2);
455         es->s_free_blocks_count--;
456         sb->s_dirt = 1;
457         unlock_super (VFSTOUFS(mp)->um_devvp);
458         return j;
459 }
460
461 #ifdef unused
462 static unsigned long
463 ext2_count_free_blocks(struct mount * mp)
464 {
465         struct ext2_sb_info *sb = VFSTOUFS(mp)->um_e2fs;
466 #ifdef EXT2FS_DEBUG
467         struct ext2_super_block * es;
468         unsigned long desc_count, bitmap_count, x;
469         int bitmap_nr;
470         struct ext2_group_desc * gdp;
471         int i;
472         
473         lock_super (VFSTOUFS(mp)->um_devvp);
474         es = sb->s_es;
475         desc_count = 0;
476         bitmap_count = 0;
477         gdp = NULL;
478         for (i = 0; i < sb->s_groups_count; i++) {
479                 gdp = get_group_desc (mp, i, NULL);
480                 desc_count += gdp->bg_free_blocks_count;
481                 bitmap_nr = load_block_bitmap (mp, i);
482                 x = ext2_count_free (sb->s_block_bitmap[bitmap_nr],
483                                      sb->s_blocksize);
484                 ext2_debug ("group %d: stored = %d, counted = %lu\n",
485                         i, gdp->bg_free_blocks_count, x);
486                 bitmap_count += x;
487         }
488         ext2_debug( "stored = %lu, computed = %lu, %lu\n",
489                es->s_free_blocks_count, desc_count, bitmap_count);
490         unlock_super (VFSTOUFS(mp)->um_devvp);
491         return bitmap_count;
492 #else
493         return sb->s_es->s_free_blocks_count;
494 #endif
495 }
496 #endif /* unused */
497
498 static __inline int
499 block_in_use (unsigned long block, struct ext2_sb_info *sb,
500               unsigned char * map)
501 {
502         return test_bit ((block - sb->s_es->s_first_data_block) %
503                          EXT2_BLOCKS_PER_GROUP(sb), map);
504 }
505
506 static int
507 test_root(int a, int b)
508 {
509         if (a == 0)
510                 return 1;
511         while (1) {
512                 if (a == 1)
513                         return 1;
514                 if (a % b)
515                         return 0;
516                 a = a / b;
517         }
518 }
519
520 int
521 ext2_group_sparse(int group)
522 {
523         return (test_root(group, 3) || test_root(group, 5) ||
524                 test_root(group, 7));
525 }
526
527 #ifdef unused
528 static void
529 ext2_check_blocks_bitmap(struct mount * mp)
530 {
531         struct ext2_sb_info *sb = VFSTOUFS(mp)->um_e2fs;
532         struct buffer_head * bh;
533         struct ext2_super_block * es;
534         unsigned long desc_count, bitmap_count, x;
535         unsigned long desc_blocks;
536         int bitmap_nr;
537         struct ext2_group_desc * gdp;
538         int i, j;
539
540         lock_super (VFSTOUFS(mp)->um_devvp);
541         es = sb->s_es;
542         desc_count = 0;
543         bitmap_count = 0;
544         gdp = NULL;
545         desc_blocks = (sb->s_groups_count + EXT2_DESC_PER_BLOCK(sb) - 1) /
546                       EXT2_DESC_PER_BLOCK(sb);
547         for (i = 0; i < sb->s_groups_count; i++) {
548                 gdp = get_group_desc (mp, i, NULL);
549                 desc_count += gdp->bg_free_blocks_count;
550                 bitmap_nr = load_block_bitmap (mp, i);
551                 bh = sb->s_block_bitmap[bitmap_nr];
552
553                 if (!(es->s_feature_ro_compat &
554                      EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER) ||
555                     ext2_group_sparse(i)) {
556                         if (!test_bit (0, bh->b_data))
557                                 printf ("ext2_check_blocks_bitmap: "
558                                             "Superblock in group %d "
559                                             "is marked free", i);
560
561                         for (j = 0; j < desc_blocks; j++)
562                                 if (!test_bit (j + 1, bh->b_data))
563                                         printf ("ext2_check_blocks_bitmap: "
564                                             "Descriptor block #%d in group "
565                                             "%d is marked free", j, i);
566                 }
567
568                 if (!block_in_use (gdp->bg_block_bitmap, sb, bh->b_data))
569                         printf ("ext2_check_blocks_bitmap: "
570                                     "Block bitmap for group %d is marked free",
571                                     i);
572
573                 if (!block_in_use (gdp->bg_inode_bitmap, sb, bh->b_data))
574                         printf ("ext2_check_blocks_bitmap: "
575                                     "Inode bitmap for group %d is marked free",
576                                     i);
577
578                 for (j = 0; j < sb->s_itb_per_group; j++)
579                         if (!block_in_use (gdp->bg_inode_table + j, sb, bh->b_data))
580                                 printf ("ext2_check_blocks_bitmap: "
581                                             "Block #%d of the inode table in "
582                                             "group %d is marked free", j, i);
583
584                 x = ext2_count_free (bh, sb->s_blocksize);
585                 if (gdp->bg_free_blocks_count != x)
586                         printf ("ext2_check_blocks_bitmap: "
587                                     "Wrong free blocks count for group %d, "
588                                     "stored = %d, counted = %lu", i,
589                                     gdp->bg_free_blocks_count, x);
590                 bitmap_count += x;
591         }
592         if (es->s_free_blocks_count != bitmap_count)
593                 printf ("ext2_check_blocks_bitmap: "
594                             "Wrong free blocks count in super block, "
595                             "stored = %lu, counted = %lu",
596                             (unsigned long) es->s_free_blocks_count, bitmap_count);
597         unlock_super (VFSTOUFS(mp)->um_devvp);
598 }
599 #endif /* unused */
600
601 /*
602  *  this function is taken from 
603  *  linux/fs/ext2/bitmap.c
604  */
605
606 static int nibblemap[] = {4, 3, 3, 2, 3, 2, 2, 1, 3, 2, 2, 1, 2, 1, 1, 0};
607
608 unsigned long ext2_count_free (struct buffer_head * map, unsigned int numchars)
609 {
610         unsigned int i;
611         unsigned long sum = 0;
612
613         if (!map)
614                 return (0);
615         for (i = 0; i < numchars; i++)
616                 sum += nibblemap[map->b_data[i] & 0xf] +
617                         nibblemap[(map->b_data[i] >> 4) & 0xf];
618         return (sum);
619 }