Whamcloud - gitweb
b=12746
[fs/lustre-release.git] / ldiskfs / kernel_patches / patches / ext3-mballoc3-core.patch
1 Index: linux-2.6.9-full/include/linux/ext3_fs.h
2 ===================================================================
3 --- linux-2.6.9-full.orig/include/linux/ext3_fs.h       2007-06-08 23:44:08.000000000 +0400
4 +++ linux-2.6.9-full/include/linux/ext3_fs.h    2007-06-29 11:31:13.000000000 +0400
5 @@ -57,6 +57,30 @@ struct statfs;
6  #define ext3_debug(f, a...)    do {} while (0)
7  #endif
8  
9 +#define EXT3_MULTIBLOCK_ALLOCATOR      1
10 +
11 +#define EXT3_MB_HINT_MERGE             1       /* prefer goal again. length */
12 +#define EXT3_MB_HINT_RESERVED          2       /* blocks already reserved */
13 +#define EXT3_MB_HINT_METADATA          4       /* metadata is being allocated */
14 +#define EXT3_MB_HINT_FIRST             8       /* first blocks in the file */
15 +#define EXT3_MB_HINT_BEST              16      /* search for the best chunk */
16 +#define EXT3_MB_HINT_DATA              32      /* data is being allocated */
17 +#define EXT3_MB_HINT_NOPREALLOC                64      /* don't preallocate (for tails) */
18 +#define EXT3_MB_HINT_GROUP_ALLOC       128     /* allocate for locality group */
19 +#define EXT3_MB_HINT_GOAL_ONLY         256     /* allocate goal blocks or none */
20 +
21 +struct ext3_allocation_request {
22 +       struct inode *inode;    /* target inode for block we're allocating */
23 +       unsigned long logical;  /* logical block in target inode */
24 +       unsigned long goal;     /* phys. target (a hint) */
25 +       unsigned long lleft;    /* the closest logical allocated block to the left */
26 +       unsigned long pleft;    /* phys. block for ^^^ */
27 +       unsigned long lright;   /* the closest logical allocated block to the right */
28 +       unsigned long pright;   /* phys. block for ^^^ */
29 +       unsigned long len;      /* how many blocks we want to allocate */
30 +       unsigned long flags;    /* flags. see above EXT3_MB_HINT_* */
31 +};
32 +
33  /*
34   * Special inodes numbers
35   */
36 @@ -387,6 +411,14 @@ struct ext3_inode {
37  #define ext3_find_first_zero_bit       ext2_find_first_zero_bit
38  #define ext3_find_next_zero_bit                ext2_find_next_zero_bit
39  
40 +#ifndef ext2_find_next_le_bit
41 +#ifdef __LITTLE_ENDIAN
42 +#define ext2_find_next_le_bit(addr, size, off) find_next_bit((addr), (size), (off))
43 +#else
44 +#error "mballoc needs a patch for big-endian systems - CFS bug 10634"
45 +#endif /* __LITTLE_ENDIAN */
46 +#endif /* !ext2_find_next_le_bit */
47 +
48  /*
49   * Maximal mount counts between two filesystem checks
50   */
51 @@ -763,6 +795,20 @@ extern unsigned long ext3_count_dirs (st
52  extern void ext3_check_inodes_bitmap (struct super_block *);
53  extern unsigned long ext3_count_free (struct buffer_head *, unsigned);
54  
55 +/* mballoc.c */
56 +extern long ext3_mb_stats;
57 +extern long ext3_mb_max_to_scan;
58 +extern int ext3_mb_init(struct super_block *, int);
59 +extern int ext3_mb_release(struct super_block *);
60 +extern unsigned long ext3_mb_new_blocks(handle_t *, struct ext3_allocation_request *, int *);
61 +extern int ext3_mb_reserve_blocks(struct super_block *, int);
62 +extern void ext3_mb_release_blocks(struct super_block *, int);
63 +extern void ext3_mb_release_blocks(struct super_block *, int);
64 +extern void ext3_mb_discard_inode_preallocations(struct inode *);
65 +extern int __init init_ext3_proc(void);
66 +extern void exit_ext3_proc(void);
67 +extern void ext3_mb_free_blocks(handle_t *, struct inode *, unsigned long, unsigned long, int, int *);
68 +
69  
70  /* inode.c */
71  extern int ext3_block_truncate_page(handle_t *, struct page *,
72 Index: linux-2.6.9-full/include/linux/ext3_fs_sb.h
73 ===================================================================
74 --- linux-2.6.9-full.orig/include/linux/ext3_fs_sb.h    2007-06-08 23:44:07.000000000 +0400
75 +++ linux-2.6.9-full/include/linux/ext3_fs_sb.h 2007-06-29 11:31:13.000000000 +0400
76 @@ -81,6 +81,61 @@ struct ext3_sb_info {
77         char *s_qf_names[MAXQUOTAS];            /* Names of quota files with journalled quota */
78         int s_jquota_fmt;                       /* Format of quota to use */
79  #endif
80 +
81 +       /* for buddy allocator */
82 +       struct ext3_group_info ***s_group_info;
83 +       struct inode *s_buddy_cache;
84 +       long s_blocks_reserved;
85 +       spinlock_t s_reserve_lock;
86 +       struct list_head s_active_transaction;
87 +       struct list_head s_closed_transaction;
88 +       struct list_head s_committed_transaction;
89 +       spinlock_t s_md_lock;
90 +       tid_t s_last_transaction;
91 +       unsigned short *s_mb_offsets, *s_mb_maxs;
92 +
93 +       /* tunables */
94 +       unsigned long s_mb_factor;
95 +       unsigned long s_stripe;
96 +       unsigned long s_mb_stream_request;
97 +       unsigned long s_mb_max_to_scan;
98 +       unsigned long s_mb_min_to_scan;
99 +       unsigned long s_mb_max_groups_to_scan;
100 +       unsigned long s_mb_stats;
101 +       unsigned long s_mb_order2_reqs;
102 +
103 +       /* history to debug policy */
104 +       struct ext3_mb_history *s_mb_history;
105 +       int s_mb_history_cur;
106 +       int s_mb_history_max;
107 +       int s_mb_history_num;
108 +       struct proc_dir_entry *s_mb_proc;
109 +       spinlock_t s_mb_history_lock;
110 +       int s_mb_history_filter;
111 +
112 +       /* stats for buddy allocator */
113 +       spinlock_t s_mb_pa_lock;
114 +       atomic_t s_bal_reqs;    /* number of reqs with len > 1 */
115 +       atomic_t s_bal_success; /* we found long enough chunks */
116 +       atomic_t s_bal_allocated;       /* in blocks */
117 +       atomic_t s_bal_ex_scanned;      /* total extents scanned */
118 +       atomic_t s_bal_goals;   /* goal hits */
119 +       atomic_t s_bal_breaks;  /* too long searches */
120 +       atomic_t s_bal_2orders; /* 2^order hits */
121 +       spinlock_t s_bal_lock;
122 +       unsigned long s_mb_buddies_generated;
123 +       unsigned long long s_mb_generation_time;
124 +       atomic_t s_mb_lost_chunks;
125 +       atomic_t s_mb_preallocated;
126 +       atomic_t s_mb_discarded;
127 +
128 +       /* locality groups */
129 +       struct ext3_locality_group *s_locality_groups;
130 +
131  };
132  
133 +#define EXT3_GROUP_INFO(sb, group)                                        \
134 +       EXT3_SB(sb)->s_group_info[(group) >> EXT3_DESC_PER_BLOCK_BITS(sb)] \
135 +                                [(group) & (EXT3_DESC_PER_BLOCK(sb) - 1)]
136 +
137  #endif /* _LINUX_EXT3_FS_SB */
138 Index: linux-2.6.9-full/fs/ext3/super.c
139 ===================================================================
140 --- linux-2.6.9-full.orig/fs/ext3/super.c       2007-06-08 23:44:08.000000000 +0400
141 +++ linux-2.6.9-full/fs/ext3/super.c    2007-06-29 11:31:13.000000000 +0400
142 @@ -394,6 +394,7 @@ void ext3_put_super (struct super_block 
143         struct ext3_super_block *es = sbi->s_es;
144         int i;
145  
146 +       ext3_mb_release(sb);
147         ext3_ext_release(sb);
148         ext3_xattr_put_super(sb);
149         journal_destroy(sbi->s_journal);
150 @@ -463,6 +464,8 @@ static struct inode *ext3_alloc_inode(st
151         ei->vfs_inode.i_version = 1;
152         
153         memset(&ei->i_cached_extent, 0, sizeof(ei->i_cached_extent));
154 +       INIT_LIST_HEAD(&ei->i_prealloc_list);
155 +       spin_lock_init(&ei->i_prealloc_lock);
156         return &ei->vfs_inode;
157  }
158  
159 @@ -2576,7 +2579,13 @@ static struct file_system_type ext3_fs_t
160  
161  static int __init init_ext3_fs(void)
162  {
163 -       int err = init_ext3_xattr();
164 +       int err;
165 +
166 +       err = init_ext3_proc();
167 +       if (err)
168 +               return err;
169 +
170 +       err = init_ext3_xattr();
171         if (err)
172                 return err;
173         err = init_inodecache();
174 @@ -2598,6 +2607,7 @@ static void __exit exit_ext3_fs(void)
175         unregister_filesystem(&ext3_fs_type);
176         destroy_inodecache();
177         exit_ext3_xattr();
178 +       exit_ext3_proc();
179  }
180  
181  int ext3_prep_san_write(struct inode *inode, long *blocks,
182 Index: linux-2.6.9-full/fs/ext3/mballoc.c
183 ===================================================================
184 --- linux-2.6.9-full.orig/fs/ext3/mballoc.c     2007-06-17 11:25:39.317298699 +0400
185 +++ linux-2.6.9-full/fs/ext3/mballoc.c  2007-06-29 12:06:50.000000000 +0400
186 @@ -0,0 +1,4371 @@
187 +/*
188 + * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
189 + * Written by Alex Tomas <alex@clusterfs.com>
190 + *
191 + * This program is free software; you can redistribute it and/or modify
192 + * it under the terms of the GNU General Public License version 2 as
193 + * published by the Free Software Foundation.
194 + *
195 + * This program is distributed in the hope that it will be useful,
196 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
197 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
198 + * GNU General Public License for more details.
199 + *
200 + * You should have received a copy of the GNU General Public Licens
201 + * along with this program; if not, write to the Free Software
202 + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
203 + */
204 +
205 +
206 +/*
207 + * mballoc.c contains the multiblocks allocation routines
208 + */
209 +
210 +#include <linux/time.h>
211 +#include <linux/fs.h>
212 +#include <linux/namei.h>
213 +#include <linux/ext3_jbd.h>
214 +#include <linux/jbd.h>
215 +#include <linux/ext3_fs.h>
216 +#include <linux/quotaops.h>
217 +#include <linux/buffer_head.h>
218 +#include <linux/module.h>
219 +#include <linux/swap.h>
220 +#include <linux/proc_fs.h>
221 +#include <linux/pagemap.h>
222 +#include <linux/seq_file.h>
223 +#include <linux/version.h>
224 +
225 +/*
226 + * MUSTDO:
227 + *   - test ext3_ext_search_left() and ext3_ext_search_right()
228 + *   - search for metadata in few groups
229 + *
230 + * TODO v4:
231 + *   - normalization should take into account whether file is still open
232 + *   - discard preallocations if no free space left (policy?)
233 + *   - don't normalize tails
234 + *   - quota
235 + *   - reservation for superuser
236 + *
237 + * TODO v3:
238 + *   - bitmap read-ahead (proposed by Oleg Drokin aka green)
239 + *   - track min/max extents in each group for better group selection
240 + *   - mb_mark_used() may allocate chunk right after splitting buddy
241 + *   - tree of groups sorted by number of free blocks
242 + *   - error handling
243 + */
244 +
245 +/*
246 + * mballoc operates on the following data:
247 + *  - on-disk bitmap
248 + *  - in-core buddy (actually includes buddy and bitmap)
249 + *  - preallocation descriptors (PAs)
250 + *
251 + * there are two types of preallocations:
252 + *  - inode
253 + *    assiged to specific inode and can be used for this inode only.
254 + *    it describes part of inode's space preallocated to specific
255 + *    physical blocks. any block from that preallocated can be used
256 + *    independent. the descriptor just tracks number of blocks left
257 + *    unused. so, before taking some block from descriptor, one must
258 + *    make sure corresponded logical block isn't allocated yet. this
259 + *    also means that freeing any block within descriptor's range
260 + *    must discard all preallocated blocks.
261 + *  - locality group
262 + *    assigned to specific locality group which does not translate to
263 + *    permanent set of inodes: inode can join and leave group. space
264 + *    from this type of preallocation can be used for any inode. thus
265 + *    it's consumed from the beginning to the end.
266 + *
267 + * relation between them can be expressed as:
268 + *    in-core buddy = on-disk bitmap + preallocation descriptors
269 + *
270 + * this mean blocks mballoc considers used are:
271 + *  - allocated blocks (persistent)
272 + *  - preallocated blocks (non-persistent)
273 + *
274 + * consistency in mballoc world means that at any time a block is either
275 + * free or used in ALL structures. notice: "any time" should not be read
276 + * literally -- time is discrete and delimited by locks.
277 + *
278 + *  to keep it simple, we don't use block numbers, instead we count number of
279 + *  blocks: how many blocks marked used/free in on-disk bitmap, buddy and PA.
280 + *
281 + * all operations can be expressed as:
282 + *  - init buddy:                      buddy = on-disk + PAs
283 + *  - new PA:                          buddy += N; PA = N
284 + *  - use inode PA:                    on-disk += N; PA -= N
285 + *  - discard inode PA                 buddy -= on-disk - PA; PA = 0
286 + *  - use locality group PA            on-disk += N; PA -= N
287 + *  - discard locality group PA                buddy -= PA; PA = 0
288 + *  note: 'buddy -= on-disk - PA' is used to show that on-disk bitmap
289 + *        is used in real operation because we can't know actual used
290 + *        bits from PA, only from on-disk bitmap
291 + *
292 + * if we follow this strict logic, then all operations above should be atomic.
293 + * given some of them can block, we'd have to use something like semaphores
294 + * killing performance on high-end SMP hardware. let's try to relax it using
295 + * the following knowledge:
296 + *  1) if buddy is referenced, it's already initialized
297 + *  2) while block is used in buddy and the buddy is referenced,
298 + *     nobody can re-allocate that block
299 + *  3) we work on bitmaps and '+' actually means 'set bits'. if on-disk has
300 + *     bit set and PA claims same block, it's OK. IOW, one can set bit in
301 + *     on-disk bitmap if buddy has same bit set or/and PA covers corresponded
302 + *     block
303 + *
304 + * so, now we're building a concurrency table:
305 + *  - init buddy vs.
306 + *    - new PA
307 + *      blocks for PA are allocated in the buddy, buddy must be referenced
308 + *      until PA is linked to allocation group to avoid concurrent buddy init
309 + *    - use inode PA
310 + *      we need to make sure that either on-disk bitmap or PA has uptodate data
311 + *      given (3) we care that PA-=N operation doesn't interfere with init
312 + *    - discard inode PA
313 + *      the simplest way would be to have buddy initialized by the discard
314 + *    - use locality group PA
315 + *      again PA-=N must be serialized with init
316 + *    - discard locality group PA
317 + *      the simplest way would be to have buddy initialized by the discard
318 + *  - new PA vs.
319 + *    - use inode PA
320 + *      i_truncate_mutex serializes them
321 + *    - discard inode PA
322 + *      discard process must wait until PA isn't used by another process
323 + *    - use locality group PA
324 + *      some mutex should serialize them
325 + *    - discard locality group PA
326 + *      discard process must wait until PA isn't used by another process
327 + *  - use inode PA
328 + *    - use inode PA
329 + *      i_truncate_mutex or another mutex should serializes them
330 + *    - discard inode PA
331 + *      discard process must wait until PA isn't used by another process
332 + *    - use locality group PA
333 + *      nothing wrong here -- they're different PAs covering different blocks
334 + *    - discard locality group PA
335 + *      discard process must wait until PA isn't used by another process
336 + *
337 + * now we're ready to make few consequences:
338 + *  - PA is referenced and while it is no discard is possible
339 + *  - PA is referenced until block isn't marked in on-disk bitmap
340 + *  - PA changes only after on-disk bitmap
341 + *  - discard must not compete with init. either init is done before
342 + *    any discard or they're serialized somehow
343 + *  - buddy init as sum of on-disk bitmap and PAs is done atomically
344 + *
345 + * a special case when we've used PA to emptiness. no need to modify buddy
346 + * in this case, but we should care about concurrent init
347 + *
348 + */
349 +
350 + /*
351 + * Logic in few words:
352 + *
353 + *  - allocation:
354 + *    load group
355 + *    find blocks
356 + *    mark bits in on-disk bitmap
357 + *    release group
358 + *
359 + *  - use preallocation:
360 + *    find proper PA (per-inode or group)
361 + *    load group
362 + *    mark bits in on-disk bitmap
363 + *    release group
364 + *    release PA
365 + *
366 + *  - free:
367 + *    load group
368 + *    mark bits in on-disk bitmap
369 + *    release group
370 + *
371 + *  - discard preallocations in group:
372 + *    mark PAs deleted
373 + *    move them onto local list
374 + *    load on-disk bitmap
375 + *    load group
376 + *    remove PA from object (inode or locality group)
377 + *    mark free blocks in-core
378 + *
379 + *  - discard inode's preallocations:
380 + */
381 +
382 +/*
383 + * Locking rules
384 + *
385 + * Locks:
386 + *  - bitlock on a group       (group)
387 + *  - object (inode/locality)  (object)
388 + *  - per-pa lock              (pa)
389 + *
390 + * Paths:
391 + *  - new pa
392 + *    object
393 + *    group
394 + *
395 + *  - find and use pa:
396 + *    pa
397 + *
398 + *  - release consumed pa:
399 + *    pa
400 + *    group
401 + *    object
402 + *
403 + *  - generate in-core bitmap:
404 + *    group
405 + *        pa
406 + *
407 + *  - discard all for given object (inode, locality group):
408 + *    object
409 + *        pa
410 + *    group
411 + *
412 + *  - discard all for given group:
413 + *    group
414 + *        pa
415 + *    group
416 + *        object
417 + *
418 + */
419 +
420 +/*
421 + * with AGGRESSIVE_CHECK allocator runs consistency checks over
422 + * structures. these checks slow things down a lot
423 + */
424 +#define AGGRESSIVE_CHECK__
425 +
426 +/*
427 + * with DOUBLE_CHECK defined mballoc creates persistent in-core
428 + * bitmaps, maintains and uses them to check for double allocations
429 + */
430 +#define DOUBLE_CHECK__
431 +
432 +/*
433 + */
434 +#define MB_DEBUG__
435 +#ifdef MB_DEBUG
436 +#define mb_debug(fmt,a...)     printk(fmt, ##a)
437 +#else
438 +#define mb_debug(fmt,a...)
439 +#endif
440 +
441 +/*
442 + * with EXT3_MB_HISTORY mballoc stores last N allocations in memory
443 + * and you can monitor it in /proc/fs/ext3/<dev>/mb_history
444 + */
445 +#define EXT3_MB_HISTORY
446 +#define EXT3_MB_HISTORY_ALLOC          1       /* allocation */
447 +#define EXT3_MB_HISTORY_PREALLOC       2       /* preallocated blocks used */
448 +#define EXT3_MB_HISTORY_DISCARD                4       /* preallocation discarded */
449 +#define EXT3_MB_HISTORY_FREE           8       /* free */
450 +
451 +#define EXT3_MB_HISTORY_DEFAULT                (EXT3_MB_HISTORY_ALLOC | \
452 +                                        EXT3_MB_HISTORY_PREALLOC | \
453 +                                        EXT3_MB_HISTORY_DISCARD | \
454 +                                        EXT3_MB_HISTORY_FREE)
455 +
456 +/*
457 + * How long mballoc can look for a best extent (in found extents)
458 + */
459 +#define MB_DEFAULT_MAX_TO_SCAN         200
460 +
461 +/*
462 + * How long mballoc must look for a best extent
463 + */
464 +#define MB_DEFAULT_MIN_TO_SCAN         10
465 +
466 +/*
467 + * How many groups mballoc will scan looking for the best chunk
468 + */
469 +#define MB_DEFAULT_MAX_GROUPS_TO_SCAN  5
470 +
471 +/*
472 + * with 'ext3_mb_stats' allocator will collect stats that will be
473 + * shown at umount. The collecting costs though!
474 + */
475 +#define MB_DEFAULT_STATS               1
476 +
477 +/*
478 + * files smaller than MB_DEFAULT_STREAM_THRESHOLD are served
479 + * by the stream allocator, which purpose is to pack requests
480 + * as close each to other as possible to produce smooth I/O traffic
481 + */
482 +#define MB_DEFAULT_STREAM_THRESHOLD    16      /* 64K */
483 +
484 +/*
485 + * for which requests use 2^N search using buddies
486 + */
487 +#define MB_DEFAULT_ORDER2_REQS         8
488 +
489 +/*
490 + * default stripe size = 1MB
491 + */
492 +#define MB_DEFAULT_STRIPE              256
493 +
494 +static kmem_cache_t *ext3_pspace_cachep = NULL;
495 +
496 +#ifdef EXT3_BB_MAX_BLOCKS
497 +#undef EXT3_BB_MAX_BLOCKS
498 +#endif
499 +#define EXT3_BB_MAX_BLOCKS     30
500 +
501 +struct ext3_free_metadata {
502 +       unsigned short group;
503 +       unsigned short num;
504 +       unsigned short blocks[EXT3_BB_MAX_BLOCKS];
505 +       struct list_head list;
506 +};
507 +
508 +struct ext3_group_info {
509 +       unsigned long   bb_state;
510 +       unsigned long   bb_tid;
511 +       struct ext3_free_metadata *bb_md_cur;
512 +       unsigned short  bb_first_free;
513 +       unsigned short  bb_free;
514 +       unsigned short  bb_fragments;
515 +       struct          list_head bb_prealloc_list;
516 +#ifdef DOUBLE_CHECK
517 +       void            *bb_bitmap;
518 +#endif
519 +       unsigned short  bb_counters[];
520 +};
521 +
522 +#define EXT3_GROUP_INFO_NEED_INIT_BIT  0
523 +#define EXT3_GROUP_INFO_LOCKED_BIT     1
524 +
525 +#define EXT3_MB_GRP_NEED_INIT(grp)     \
526 +       (test_bit(EXT3_GROUP_INFO_NEED_INIT_BIT, &(grp)->bb_state))
527 +
528 +
529 +struct ext3_prealloc_space {
530 +       struct list_head        pa_inode_list;
531 +       struct list_head        pa_group_list;
532 +       union {
533 +               struct list_head pa_tmp_list;
534 +               struct rcu_head pa_rcu;
535 +       } u;
536 +       spinlock_t              pa_lock;
537 +       atomic_t                pa_count;
538 +       unsigned                pa_deleted;
539 +       unsigned long           pa_pstart;      /* phys. block */
540 +       unsigned long           pa_lstart;      /* log. block */
541 +       unsigned short          pa_len;         /* len of preallocated chunk */
542 +       unsigned short          pa_free;        /* how many blocks are free */
543 +       unsigned short          pa_linear;      /* consumed in one direction
544 +                                                * strictly, for group prealloc */
545 +       spinlock_t              *pa_obj_lock;
546 +       struct inode            *pa_inode;      /* hack, for history only */
547 +};
548 +
549 +
550 +struct ext3_free_extent {
551 +       unsigned long fe_logical;
552 +       unsigned long fe_start;
553 +       unsigned long fe_group;
554 +       unsigned long fe_len;
555 +};
556 +
557 +/*
558 + * Locality group:
559 + *   we try to group all related changes together
560 + *   so that writeback can flush/allocate them together as well
561 + */
562 +struct ext3_locality_group {
563 +       /* for allocator */
564 +       struct semaphore        lg_sem;         /* to serialize allocates */
565 +       struct list_head        lg_prealloc_list;/* list of preallocations */
566 +       spinlock_t              lg_prealloc_lock;
567 +};
568 +
569 +struct ext3_allocation_context {
570 +       struct inode *ac_inode;
571 +       struct super_block *ac_sb;
572 +
573 +       /* original request */
574 +       struct ext3_free_extent ac_o_ex;
575 +
576 +       /* goal request (after normalization) */
577 +       struct ext3_free_extent ac_g_ex;
578 +
579 +       /* the best found extent */
580 +       struct ext3_free_extent ac_b_ex;
581 +
582 +       /* copy of the bext found extent taken before preallocation efforts */
583 +       struct ext3_free_extent ac_f_ex;
584 +
585 +       /* number of iterations done. we have to track to limit searching */
586 +       unsigned long ac_ex_scanned;
587 +       __u16 ac_groups_scanned;
588 +       __u16 ac_found;
589 +       __u16 ac_tail;
590 +       __u16 ac_buddy;
591 +       __u16 ac_flags;         /* allocation hints */
592 +       __u8 ac_status;
593 +       __u8 ac_criteria;
594 +       __u8 ac_repeats;
595 +       __u8 ac_2order;         /* if request is to allocate 2^N blocks and
596 +                                * N > 0, the field stores N, otherwise 0 */
597 +       __u8 ac_op;             /* operation, for history only */
598 +       struct page *ac_bitmap_page;
599 +       struct page *ac_buddy_page;
600 +       struct ext3_prealloc_space *ac_pa;
601 +       struct ext3_locality_group *ac_lg;
602 +};
603 +
604 +#define AC_STATUS_CONTINUE     1
605 +#define AC_STATUS_FOUND                2
606 +#define AC_STATUS_BREAK                3
607 +
608 +struct ext3_mb_history {
609 +       struct ext3_free_extent orig;   /* orig allocation */
610 +       struct ext3_free_extent goal;   /* goal allocation */
611 +       struct ext3_free_extent result; /* result allocation */
612 +       unsigned pid;
613 +       unsigned ino;
614 +       __u16 found;    /* how many extents have been found */
615 +       __u16 groups;   /* how many groups have been scanned */
616 +       __u16 tail;     /* what tail broke some buddy */
617 +       __u16 buddy;    /* buddy the tail ^^^ broke */
618 +       __u16 flags;
619 +       __u8 cr:3;      /* which phase the result extent was found at */
620 +       __u8 op:4;
621 +       __u8 merged:1;
622 +};
623 +
624 +struct ext3_buddy {
625 +       struct page *bd_buddy_page;
626 +       void *bd_buddy;
627 +       struct page *bd_bitmap_page;
628 +       void *bd_bitmap;
629 +       struct ext3_group_info *bd_info;
630 +       struct super_block *bd_sb;
631 +       __u16 bd_blkbits;
632 +       __u16 bd_group;
633 +};
634 +#define EXT3_MB_BITMAP(e3b)    ((e3b)->bd_bitmap)
635 +#define EXT3_MB_BUDDY(e3b)     ((e3b)->bd_buddy)
636 +
637 +#ifndef EXT3_MB_HISTORY
638 +#define ext3_mb_store_history(ac)
639 +#else
640 +static void ext3_mb_store_history(struct ext3_allocation_context *ac);
641 +#endif
642 +
643 +#define in_range(b, first, len)        ((b) >= (first) && (b) <= (first) + (len) - 1)
644 +
645 +static struct proc_dir_entry *proc_root_ext3;
646 +
647 +int ext3_create (struct inode *, struct dentry *, int, struct nameidata *);
648 +struct buffer_head * read_block_bitmap(struct super_block *, unsigned int);
649 +unsigned long ext3_new_blocks_old(handle_t *handle, struct inode *inode,
650 +                       unsigned long goal, unsigned long *count, int *errp);
651 +void ext3_mb_release_blocks(struct super_block *, int);
652 +void ext3_mb_poll_new_transaction(struct super_block *, handle_t *);
653 +void ext3_mb_free_committed_blocks(struct super_block *);
654 +void ext3_mb_generate_from_pa(struct super_block *sb, void *bitmap, int group);
655 +void ext3_mb_free_consumed_preallocations(struct ext3_allocation_context *ac);
656 +void ext3_mb_return_to_preallocation(struct inode *inode, struct ext3_buddy *e3b,
657 +                                       sector_t block, int count);
658 +void ext3_mb_show_ac(struct ext3_allocation_context *ac);
659 +void ext3_mb_check_with_pa(struct ext3_buddy *e3b, int first, int count);
660 +void ext3_mb_put_pa(struct ext3_allocation_context *, struct super_block *, struct ext3_prealloc_space *pa);
661 +int ext3_mb_init_per_dev_proc(struct super_block *sb);
662 +int ext3_mb_destroy_per_dev_proc(struct super_block *sb);
663 +
664 +/*
665 + * Calculate the block group number and offset, given a block number
666 + */
667 +static void ext3_get_group_no_and_offset(struct super_block *sb,
668 +                                       unsigned long blocknr,
669 +                                       unsigned long *blockgrpp,
670 +                                       unsigned long *offsetp)
671 +{
672 +       struct ext3_super_block *es = EXT3_SB(sb)->s_es;
673 +       unsigned long offset;
674 +
675 +       blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
676 +       offset = blocknr % EXT3_BLOCKS_PER_GROUP(sb);
677 +       blocknr = blocknr / EXT3_BLOCKS_PER_GROUP(sb);
678 +       if (offsetp)
679 +               *offsetp = offset;
680 +       if (blockgrpp)
681 +               *blockgrpp = blocknr;
682 +
683 +}
684 +
685 +static inline void
686 +ext3_lock_group(struct super_block *sb, int group)
687 +{
688 +       bit_spin_lock(EXT3_GROUP_INFO_LOCKED_BIT,
689 +                     &EXT3_GROUP_INFO(sb, group)->bb_state);
690 +}
691 +
692 +static inline void
693 +ext3_unlock_group(struct super_block *sb, int group)
694 +{
695 +       bit_spin_unlock(EXT3_GROUP_INFO_LOCKED_BIT,
696 +                       &EXT3_GROUP_INFO(sb, group)->bb_state);
697 +}
698 +
699 +static inline int
700 +ext3_is_group_locked(struct super_block *sb, int group)
701 +{
702 +       return bit_spin_is_locked(EXT3_GROUP_INFO_LOCKED_BIT,
703 +                                       &EXT3_GROUP_INFO(sb, group)->bb_state);
704 +}
705 +
706 +unsigned long ext3_grp_offs_to_block(struct super_block *sb,
707 +                                       struct ext3_free_extent *fex)
708 +{
709 +       unsigned long block;
710 +
711 +       block = (unsigned long) fex->fe_group * EXT3_BLOCKS_PER_GROUP(sb)
712 +                       + fex->fe_start
713 +                       + le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block);
714 +       return block;
715 +}
716 +
717 +#if BITS_PER_LONG == 64
718 +#define mb_correct_addr_and_bit(bit,addr)              \
719 +{                                                      \
720 +       bit += ((unsigned long) addr & 7UL) << 3;       \
721 +       addr = (void *) ((unsigned long) addr & ~7UL);  \
722 +}
723 +#elif BITS_PER_LONG == 32
724 +#define mb_correct_addr_and_bit(bit,addr)              \
725 +{                                                      \
726 +       bit += ((unsigned long) addr & 3UL) << 3;       \
727 +       addr = (void *) ((unsigned long) addr & ~3UL);  \
728 +}
729 +#else
730 +#error "how many bits you are?!"
731 +#endif
732 +
733 +static inline int mb_test_bit(int bit, void *addr)
734 +{
735 +       mb_correct_addr_and_bit(bit,addr);
736 +       return ext2_test_bit(bit, addr);
737 +}
738 +
739 +static inline void mb_set_bit(int bit, void *addr)
740 +{
741 +       mb_correct_addr_and_bit(bit,addr);
742 +       ext2_set_bit(bit, addr);
743 +}
744 +
745 +static inline void mb_set_bit_atomic(int bit, void *addr)
746 +{
747 +       mb_correct_addr_and_bit(bit,addr);
748 +       ext2_set_bit_atomic(NULL, bit, addr);
749 +}
750 +
751 +static inline void mb_clear_bit(int bit, void *addr)
752 +{
753 +       mb_correct_addr_and_bit(bit,addr);
754 +       ext2_clear_bit(bit, addr);
755 +}
756 +
757 +static inline void mb_clear_bit_atomic(int bit, void *addr)
758 +{
759 +       mb_correct_addr_and_bit(bit,addr);
760 +       ext2_clear_bit_atomic(NULL, bit, addr);
761 +}
762 +
763 +static inline int mb_find_next_zero_bit(void *addr, int max, int start)
764 +{
765 +       int fix;
766 +#if BITS_PER_LONG == 64
767 +       fix = ((unsigned long) addr & 7UL) << 3;
768 +       addr = (void *) ((unsigned long) addr & ~7UL);
769 +#elif BITS_PER_LONG == 32
770 +       fix = ((unsigned long) addr & 3UL) << 3;
771 +       addr = (void *) ((unsigned long) addr & ~3UL);
772 +#else
773 +#error "how many bits you are?!"
774 +#endif
775 +       max += fix;
776 +       start += fix;
777 +       return ext2_find_next_zero_bit(addr, max, start) - fix;
778 +}
779 +
780 +static inline int mb_find_next_bit(void *addr, int max, int start)
781 +{
782 +       int fix;
783 +#if BITS_PER_LONG == 64
784 +       fix = ((unsigned long) addr & 7UL) << 3;
785 +       addr = (void *) ((unsigned long) addr & ~7UL);
786 +#elif BITS_PER_LONG == 32
787 +       fix = ((unsigned long) addr & 3UL) << 3;
788 +       addr = (void *) ((unsigned long) addr & ~3UL);
789 +#else
790 +#error "how many bits you are?!"
791 +#endif
792 +       max += fix;
793 +       start += fix;
794 +
795 +#ifdef __BIG_ENDIAN
796 +#else
797 +       return find_next_bit(addr, max, start) - fix;
798 +#endif
799 +}
800 +
801 +static inline void *mb_find_buddy(struct ext3_buddy *e3b, int order, int *max)
802 +{
803 +       char *bb;
804 +
805 +       BUG_ON(EXT3_MB_BITMAP(e3b) == EXT3_MB_BUDDY(e3b));
806 +       BUG_ON(max == NULL);
807 +
808 +       if (order > e3b->bd_blkbits + 1) {
809 +               *max = 0;
810 +               return NULL;
811 +       }
812 +
813 +       /* at order 0 we see each particular block */
814 +       *max = 1 << (e3b->bd_blkbits + 3);
815 +       if (order == 0)
816 +               return EXT3_MB_BITMAP(e3b);
817 +
818 +       bb = EXT3_MB_BUDDY(e3b) + EXT3_SB(e3b->bd_sb)->s_mb_offsets[order];
819 +       *max = EXT3_SB(e3b->bd_sb)->s_mb_maxs[order];
820 +
821 +       return bb;
822 +}
823 +
824 +#ifdef DOUBLE_CHECK
825 +void mb_free_blocks_double(struct inode *inode, struct ext3_buddy *e3b,
826 +                          int first, int count)
827 +{
828 +       int i;
829 +       struct super_block *sb = e3b->bd_sb;
830 +
831 +       if (unlikely(e3b->bd_info->bb_bitmap == NULL))
832 +               return;
833 +       BUG_ON(!ext3_is_group_locked(sb, e3b->bd_group));
834 +       for (i = 0; i < count; i++) {
835 +               if (!mb_test_bit(first + i, e3b->bd_info->bb_bitmap)) {
836 +                       unsigned long blocknr;
837 +                       blocknr = e3b->bd_group * EXT3_BLOCKS_PER_GROUP(sb);
838 +                       blocknr += first + i;
839 +                       blocknr +=
840 +                           le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block);
841 +
842 +                       ext3_error(sb, __FUNCTION__, "double-free of inode"
843 +                                  " %lu's block %lu(bit %u in group %u)\n",
844 +                                  inode ? inode->i_ino : 0, blocknr,
845 +                                  first + i, e3b->bd_group);
846 +               }
847 +               mb_clear_bit(first + i, e3b->bd_info->bb_bitmap);
848 +       }
849 +}
850 +
851 +void mb_mark_used_double(struct ext3_buddy *e3b, int first, int count)
852 +{
853 +       int i;
854 +       if (unlikely(e3b->bd_info->bb_bitmap == NULL))
855 +               return;
856 +       BUG_ON(!ext3_is_group_locked(e3b->bd_sb, e3b->bd_group));
857 +       for (i = 0; i < count; i++) {
858 +               BUG_ON(mb_test_bit(first + i, e3b->bd_info->bb_bitmap));
859 +               mb_set_bit(first + i, e3b->bd_info->bb_bitmap);
860 +       }
861 +}
862 +
863 +void mb_cmp_bitmaps(struct ext3_buddy *e3b, void *bitmap)
864 +{
865 +       if (memcmp(e3b->bd_info->bb_bitmap, bitmap, e3b->bd_sb->s_blocksize)) {
866 +               unsigned char *b1, *b2;
867 +               int i;
868 +               b1 = (unsigned char *) e3b->bd_info->bb_bitmap;
869 +               b2 = (unsigned char *) bitmap;
870 +               for (i = 0; i < e3b->bd_sb->s_blocksize; i++) {
871 +                       if (b1[i] != b2[i]) {
872 +                               printk("corruption in group %u at byte %u(%u): "
873 +                                      "%x in copy != %x on disk/prealloc\n",
874 +                                       e3b->bd_group, i, i * 8, b1[i], b2[i]);
875 +                               BUG();
876 +                       }
877 +               }
878 +       }
879 +}
880 +
881 +#else
882 +#define mb_free_blocks_double(a,b,c,d)
883 +#define mb_mark_used_double(a,b,c)
884 +#define mb_cmp_bitmaps(a,b)
885 +#endif
886 +
887 +#ifdef AGGRESSIVE_CHECK
888 +
889 +#define MB_CHECK_ASSERT(assert)                                                \
890 +do {                                                                   \
891 +       if (!(assert)) {                                                \
892 +               printk (KERN_EMERG                                      \
893 +                       "Assertion failure in %s() at %s:%d: \"%s\"\n", \
894 +                       function, file, line, # assert);                \
895 +               BUG();                                                  \
896 +       }                                                               \
897 +} while (0)
898 +
899 +static int __mb_check_buddy(struct ext3_buddy *e3b, char *file,
900 +                               const char *function, int line)
901 +{
902 +       struct super_block *sb = e3b->bd_sb;
903 +       int order = e3b->bd_blkbits + 1;
904 +       int max, max2, i, j, k, count;
905 +       struct ext3_group_info *grp;
906 +       int fragments = 0, fstart;
907 +       struct list_head *cur;
908 +       void *buddy, *buddy2;
909 +
910 +       if (!test_opt(sb, MBALLOC))
911 +               return 0;
912 +
913 +       {
914 +               static int mb_check_counter = 0;
915 +               if (mb_check_counter++ % 100 != 0)
916 +                       return 0;
917 +       }
918 +
919 +       while (order > 1) {
920 +               buddy = mb_find_buddy(e3b, order, &max);
921 +               MB_CHECK_ASSERT(buddy);
922 +               buddy2 = mb_find_buddy(e3b, order - 1, &max2);
923 +               MB_CHECK_ASSERT(buddy2);
924 +               MB_CHECK_ASSERT(buddy != buddy2);
925 +               MB_CHECK_ASSERT(max * 2 == max2);
926 +
927 +               count = 0;
928 +               for (i = 0; i < max; i++) {
929 +
930 +                       if (mb_test_bit(i, buddy)) {
931 +                               /* only single bit in buddy2 may be 1 */
932 +                               if (!mb_test_bit(i << 1, buddy2))
933 +                                       MB_CHECK_ASSERT(mb_test_bit((i<<1)+1, buddy2));
934 +                               else if (!mb_test_bit((i << 1) + 1, buddy2))
935 +                                       MB_CHECK_ASSERT(mb_test_bit(i << 1, buddy2));
936 +                               continue;
937 +                       }
938 +
939 +                       /* both bits in buddy2 must be 0 */
940 +                       MB_CHECK_ASSERT(mb_test_bit(i << 1, buddy2));
941 +                       MB_CHECK_ASSERT(mb_test_bit((i << 1) + 1, buddy2));
942 +
943 +                       for (j = 0; j < (1 << order); j++) {
944 +                               k = (i * (1 << order)) + j;
945 +                               MB_CHECK_ASSERT(!mb_test_bit(k, EXT3_MB_BITMAP(e3b)));
946 +                       }
947 +                       count++;
948 +               }
949 +               MB_CHECK_ASSERT(e3b->bd_info->bb_counters[order] == count);
950 +               order--;
951 +       }
952 +
953 +       fstart = -1;
954 +       buddy = mb_find_buddy(e3b, 0, &max);
955 +       for (i = 0; i < max; i++) {
956 +               if (!mb_test_bit(i, buddy)) {
957 +                       MB_CHECK_ASSERT(i >= e3b->bd_info->bb_first_free);
958 +                       if (fstart == -1) {
959 +                               fragments++;
960 +                               fstart = i;
961 +                       }
962 +                       continue;
963 +               }
964 +               fstart = -1;
965 +               /* check used bits only */
966 +               for (j = 0; j < e3b->bd_blkbits + 1; j++) {
967 +                       buddy2 = mb_find_buddy(e3b, j, &max2);
968 +                       k = i >> j;
969 +                       MB_CHECK_ASSERT(k < max2);
970 +                       MB_CHECK_ASSERT(mb_test_bit(k, buddy2));
971 +               }
972 +       }
973 +       MB_CHECK_ASSERT(!EXT3_MB_GRP_NEED_INIT(e3b->bd_info));
974 +       MB_CHECK_ASSERT(e3b->bd_info->bb_fragments == fragments);
975 +
976 +       grp = EXT3_GROUP_INFO(sb, e3b->bd_group);
977 +       buddy = mb_find_buddy(e3b, 0, &max);
978 +       list_for_each(cur, &grp->bb_prealloc_list) {
979 +               unsigned long groupnr;
980 +               struct ext3_prealloc_space *pa;
981 +               pa = list_entry(cur, struct ext3_prealloc_space, group_list);
982 +               ext3_get_group_no_and_offset(sb, pa->pstart, &groupnr, &k);
983 +               MB_CHECK_ASSERT(groupnr == e3b->bd_group);
984 +               for (i = 0; i < pa->len; i++)
985 +                       MB_CHECK_ASSERT(mb_test_bit(k + i, buddy));
986 +       }
987 +       return 0;
988 +}
989 +#undef MB_CHECK_ASSERT
990 +#define mb_check_buddy(e3b) __mb_check_buddy(e3b,__FILE__,__FUNCTION__,__LINE__)
991 +#else
992 +#define mb_check_buddy(e3b)
993 +#endif
994 +
995 +/* find most significant bit */
996 +static int inline fmsb(unsigned short word)
997 +{
998 +       int order;
999 +
1000 +       if (word > 255) {
1001 +               order = 7;
1002 +               word >>= 8;
1003 +       } else {
1004 +               order = -1;
1005 +       }
1006 +
1007 +       do {
1008 +               order++;
1009 +               word >>= 1;
1010 +       } while (word != 0);
1011 +
1012 +       return order;
1013 +}
1014 +
1015 +static void inline
1016 +ext3_mb_mark_free_simple(struct super_block *sb, void *buddy, unsigned first,
1017 +                               int len, struct ext3_group_info *grp)
1018 +{
1019 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
1020 +       unsigned short min, max, chunk, border;
1021 +
1022 +       BUG_ON(len >= EXT3_BLOCKS_PER_GROUP(sb));
1023 +
1024 +       border = 2 << sb->s_blocksize_bits;
1025 +
1026 +       while (len > 0) {
1027 +               /* find how many blocks can be covered since this position */
1028 +               max = ffs(first | border) - 1;
1029 +
1030 +               /* find how many blocks of power 2 we need to mark */
1031 +               min = fmsb(len);
1032 +
1033 +               if (max < min)
1034 +                       min = max;
1035 +               chunk = 1 << min;
1036 +
1037 +               /* mark multiblock chunks only */
1038 +               grp->bb_counters[min]++;
1039 +               if (min > 0)
1040 +                       mb_clear_bit(first >> min, buddy + sbi->s_mb_offsets[min]);
1041 +
1042 +               len -= chunk;
1043 +               first += chunk;
1044 +       }
1045 +}
1046 +
1047 +static void
1048 +ext3_mb_generate_buddy(struct super_block *sb, void *buddy, void *bitmap,
1049 +                       int group)
1050 +{
1051 +       struct ext3_group_info *grp = EXT3_GROUP_INFO(sb, group);
1052 +       unsigned short max = EXT3_BLOCKS_PER_GROUP(sb);
1053 +       unsigned short i = 0, first, len;
1054 +       unsigned free = 0, fragments = 0;
1055 +       unsigned long long period = get_cycles();
1056 +
1057 +       /* initialize buddy from bitmap which is aggregation
1058 +        * of on-disk bitmap and preallocations */
1059 +       i = mb_find_next_zero_bit(bitmap, max, 0);
1060 +       grp->bb_first_free = i;
1061 +       while (i < max) {
1062 +               fragments++;
1063 +               first = i;
1064 +               i = ext2_find_next_le_bit(bitmap, max, i);
1065 +               len = i - first;
1066 +               free += len;
1067 +               if (len > 1)
1068 +                       ext3_mb_mark_free_simple(sb, buddy, first, len, grp);
1069 +               else
1070 +                       grp->bb_counters[0]++;
1071 +               if (i < max)
1072 +                       i = mb_find_next_zero_bit(bitmap, max, i);
1073 +       }
1074 +       grp->bb_fragments = fragments;
1075 +
1076 +       if (free != grp->bb_free) {
1077 +               printk("EXT3-fs: group %u: %u blocks in bitmap, %u in gd\n",
1078 +                       group, free, grp->bb_free);
1079 +               grp->bb_free = free;
1080 +       }
1081 +
1082 +       clear_bit(EXT3_GROUP_INFO_NEED_INIT_BIT, &grp->bb_state);
1083 +
1084 +       period = get_cycles() - period;
1085 +       spin_lock(&EXT3_SB(sb)->s_bal_lock);
1086 +       EXT3_SB(sb)->s_mb_buddies_generated++;
1087 +       EXT3_SB(sb)->s_mb_generation_time += period;
1088 +       spin_unlock(&EXT3_SB(sb)->s_bal_lock);
1089 +}
1090 +
1091 +static int ext3_mb_init_cache(struct page *page, char *incore)
1092 +{
1093 +       int blocksize, blocks_per_page, groups_per_page;
1094 +       int err = 0, i, first_group, first_block;
1095 +       struct super_block *sb;
1096 +       struct buffer_head *bhs;
1097 +       struct buffer_head **bh;
1098 +       struct inode *inode;
1099 +       char *data, *bitmap;
1100 +
1101 +       mb_debug("init page %lu\n", page->index);
1102 +
1103 +       inode = page->mapping->host;
1104 +       sb = inode->i_sb;
1105 +       blocksize = 1 << inode->i_blkbits;
1106 +       blocks_per_page = PAGE_CACHE_SIZE / blocksize;
1107 +
1108 +       groups_per_page = blocks_per_page >> 1;
1109 +       if (groups_per_page == 0)
1110 +               groups_per_page = 1;
1111 +
1112 +       /* allocate buffer_heads to read bitmaps */
1113 +       if (groups_per_page > 1) {
1114 +               err = -ENOMEM;
1115 +               i = sizeof(struct buffer_head *) * groups_per_page;
1116 +               bh = kmalloc(i, GFP_NOFS);
1117 +               if (bh == NULL)
1118 +                       goto out;
1119 +               memset(bh, 0, i);
1120 +       } else
1121 +               bh = &bhs;
1122 +
1123 +       first_group = page->index * blocks_per_page / 2;
1124 +
1125 +       /* read all groups the page covers into the cache */
1126 +       for (i = 0; i < groups_per_page; i++) {
1127 +               struct ext3_group_desc * desc;
1128 +
1129 +               if (first_group + i >= EXT3_SB(sb)->s_groups_count)
1130 +                       break;
1131 +
1132 +               err = -EIO;
1133 +               desc = ext3_get_group_desc(sb, first_group + i, NULL);
1134 +               if (desc == NULL)
1135 +                       goto out;
1136 +
1137 +               err = -ENOMEM;
1138 +               bh[i] = sb_getblk(sb, le32_to_cpu(desc->bg_block_bitmap));
1139 +               if (bh[i] == NULL)
1140 +                       goto out;
1141 +
1142 +               if (buffer_uptodate(bh[i]))
1143 +                       continue;
1144 +
1145 +               lock_buffer(bh[i]);
1146 +               if (buffer_uptodate(bh[i])) {
1147 +                       unlock_buffer(bh[i]);
1148 +                       continue;
1149 +               }
1150 +
1151 +               get_bh(bh[i]);
1152 +               bh[i]->b_end_io = end_buffer_read_sync;
1153 +               submit_bh(READ, bh[i]);
1154 +               mb_debug("read bitmap for group %u\n", first_group + i);
1155 +       }
1156 +
1157 +       /* wait for I/O completion */
1158 +       for (i = 0; i < groups_per_page && bh[i]; i++)
1159 +               wait_on_buffer(bh[i]);
1160 +
1161 +       err = -EIO;
1162 +       for (i = 0; i < groups_per_page && bh[i]; i++)
1163 +               if (!buffer_uptodate(bh[i]))
1164 +                       goto out;
1165 +
1166 +       first_block = page->index * blocks_per_page;
1167 +       for (i = 0; i < blocks_per_page; i++) {
1168 +               int group;
1169 +
1170 +               group = (first_block + i) >> 1;
1171 +               if (group >= EXT3_SB(sb)->s_groups_count)
1172 +                       break;
1173 +
1174 +               data = page_address(page) + (i * blocksize);
1175 +               bitmap = bh[group - first_group]->b_data;
1176 +
1177 +               if ((first_block + i) & 1) {
1178 +                       /* this is block of buddy */
1179 +                       BUG_ON(incore == NULL);
1180 +                       mb_debug("put buddy for group %u in page %lu/%x\n",
1181 +                               group, page->index, i * blocksize);
1182 +                       memset(data, 0xff, blocksize);
1183 +                       EXT3_GROUP_INFO(sb, group)->bb_fragments = 0;
1184 +                       memset(EXT3_GROUP_INFO(sb, group)->bb_counters, 0,
1185 +                              sizeof(unsigned short)*(sb->s_blocksize_bits+2));
1186 +                       ext3_mb_generate_buddy(sb, data, incore, group);
1187 +                       incore = NULL;
1188 +               } else {
1189 +                       /* this is block of bitmap */
1190 +                       BUG_ON(incore != NULL);
1191 +                       mb_debug("put bitmap for group %u in page %lu/%x\n",
1192 +                               group, page->index, i * blocksize);
1193 +
1194 +                       /* see comments in ext3_mb_put_pa() */
1195 +                       ext3_lock_group(sb, group);
1196 +                       memcpy(data, bitmap, blocksize);
1197 +
1198 +                       /* mark all preallocated blocks used in in-core bitmap */
1199 +                       ext3_mb_generate_from_pa(sb, data, group);
1200 +                       ext3_unlock_group(sb, group);
1201 +
1202 +                       incore = data;
1203 +               }
1204 +       }
1205 +       SetPageUptodate(page);
1206 +
1207 +out:
1208 +       if (bh) {
1209 +               for (i = 0; i < groups_per_page && bh[i]; i++)
1210 +                       brelse(bh[i]);
1211 +               if (bh != &bhs)
1212 +                       kfree(bh);
1213 +       }
1214 +       return err;
1215 +}
1216 +
1217 +static int ext3_mb_load_buddy(struct super_block *sb, int group,
1218 +               struct ext3_buddy *e3b)
1219 +{
1220 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
1221 +       struct inode *inode = sbi->s_buddy_cache;
1222 +       int blocks_per_page, block, pnum, poff;
1223 +       struct page *page;
1224 +
1225 +       mb_debug("load group %u\n", group);
1226 +
1227 +       blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
1228 +
1229 +       e3b->bd_blkbits = sb->s_blocksize_bits;
1230 +       e3b->bd_info = EXT3_GROUP_INFO(sb, group);
1231 +       e3b->bd_sb = sb;
1232 +       e3b->bd_group = group;
1233 +       e3b->bd_buddy_page = NULL;
1234 +       e3b->bd_bitmap_page = NULL;
1235 +
1236 +       block = group * 2;
1237 +       pnum = block / blocks_per_page;
1238 +       poff = block % blocks_per_page;
1239 +
1240 +       /* we could use find_or_create_page(), but it locks page
1241 +        * what we'd like to avoid in fast path ... */
1242 +       page = find_get_page(inode->i_mapping, pnum);
1243 +       if (page == NULL || !PageUptodate(page)) {
1244 +               if (page)
1245 +                       page_cache_release(page);
1246 +               page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
1247 +               if (page) {
1248 +                       BUG_ON(page->mapping != inode->i_mapping);
1249 +                       if (!PageUptodate(page)) {
1250 +                               ext3_mb_init_cache(page, NULL);
1251 +                               mb_cmp_bitmaps(e3b, page_address(page) +
1252 +                                              (poff * sb->s_blocksize));
1253 +                       }
1254 +                       unlock_page(page);
1255 +               }
1256 +       }
1257 +       if (page == NULL || !PageUptodate(page))
1258 +               goto err;
1259 +       e3b->bd_bitmap_page = page;
1260 +       e3b->bd_bitmap = page_address(page) + (poff * sb->s_blocksize);
1261 +       mark_page_accessed(page);
1262 +
1263 +       block++;
1264 +       pnum = block / blocks_per_page;
1265 +       poff = block % blocks_per_page;
1266 +
1267 +       page = find_get_page(inode->i_mapping, pnum);
1268 +       if (page == NULL || !PageUptodate(page)) {
1269 +               if (page)
1270 +                       page_cache_release(page);
1271 +               page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
1272 +               if (page) {
1273 +                       BUG_ON(page->mapping != inode->i_mapping);
1274 +                       if (!PageUptodate(page))
1275 +                               ext3_mb_init_cache(page, e3b->bd_bitmap);
1276 +
1277 +                       unlock_page(page);
1278 +               }
1279 +       }
1280 +       if (page == NULL || !PageUptodate(page))
1281 +               goto err;
1282 +       e3b->bd_buddy_page = page;
1283 +       e3b->bd_buddy = page_address(page) + (poff * sb->s_blocksize);
1284 +       mark_page_accessed(page);
1285 +
1286 +       BUG_ON(e3b->bd_bitmap_page == NULL);
1287 +       BUG_ON(e3b->bd_buddy_page == NULL);
1288 +
1289 +       return 0;
1290 +
1291 +err:
1292 +       if (e3b->bd_bitmap_page)
1293 +               page_cache_release(e3b->bd_bitmap_page);
1294 +       if (e3b->bd_buddy_page)
1295 +               page_cache_release(e3b->bd_buddy_page);
1296 +       e3b->bd_buddy = NULL;
1297 +       e3b->bd_bitmap = NULL;
1298 +       return -EIO;
1299 +}
1300 +
1301 +static void ext3_mb_release_desc(struct ext3_buddy *e3b)
1302 +{
1303 +       if (e3b->bd_bitmap_page)
1304 +               page_cache_release(e3b->bd_bitmap_page);
1305 +       if (e3b->bd_buddy_page)
1306 +               page_cache_release(e3b->bd_buddy_page);
1307 +}
1308 +
1309 +
1310 +static int mb_find_order_for_block(struct ext3_buddy *e3b, int block)
1311 +{
1312 +       int order = 1;
1313 +       void *bb;
1314 +
1315 +       BUG_ON(EXT3_MB_BITMAP(e3b) == EXT3_MB_BUDDY(e3b));
1316 +       BUG_ON(block >= (1 << (e3b->bd_blkbits + 3)));
1317 +
1318 +       bb = EXT3_MB_BUDDY(e3b);
1319 +       while (order <= e3b->bd_blkbits + 1) {
1320 +               block = block >> 1;
1321 +               if (!mb_test_bit(block, bb)) {
1322 +                       /* this block is part of buddy of order 'order' */
1323 +                       return order;
1324 +               }
1325 +               bb += 1 << (e3b->bd_blkbits - order);
1326 +               order++;
1327 +       }
1328 +       return 0;
1329 +}
1330 +
1331 +static inline void mb_clear_bits(void *bm, int cur, int len)
1332 +{
1333 +       __u32 *addr;
1334 +
1335 +       len = cur + len;
1336 +       while (cur < len) {
1337 +               if ((cur & 31) == 0 && (len - cur) >= 32) {
1338 +                       /* fast path: clear whole word at once */
1339 +                       addr = bm + (cur >> 3);
1340 +                       *addr = 0;
1341 +                       cur += 32;
1342 +                       continue;
1343 +               }
1344 +               mb_clear_bit_atomic(cur, bm);
1345 +               cur++;
1346 +       }
1347 +}
1348 +
1349 +static inline void mb_set_bits(void *bm, int cur, int len)
1350 +{
1351 +       __u32 *addr;
1352 +
1353 +       len = cur + len;
1354 +       while (cur < len) {
1355 +               if ((cur & 31) == 0 && (len - cur) >= 32) {
1356 +                       /* fast path: clear whole word at once */
1357 +                       addr = bm + (cur >> 3);
1358 +                       *addr = 0xffffffff;
1359 +                       cur += 32;
1360 +                       continue;
1361 +               }
1362 +               mb_set_bit_atomic(cur, bm);
1363 +               cur++;
1364 +       }
1365 +}
1366 +
1367 +static int mb_free_blocks(struct inode *inode, struct ext3_buddy *e3b,
1368 +                         int first, int count)
1369 +{
1370 +       int block = 0, max = 0, order;
1371 +       void *buddy, *buddy2;
1372 +       struct super_block *sb = e3b->bd_sb;
1373 +
1374 +       BUG_ON(first + count > (sb->s_blocksize << 3));
1375 +       BUG_ON(!ext3_is_group_locked(sb, e3b->bd_group));
1376 +       mb_check_buddy(e3b);
1377 +       mb_free_blocks_double(inode, e3b, first, count);
1378 +
1379 +       e3b->bd_info->bb_free += count;
1380 +       if (first < e3b->bd_info->bb_first_free)
1381 +               e3b->bd_info->bb_first_free = first;
1382 +
1383 +       /* let's maintain fragments counter */
1384 +       if (first != 0)
1385 +               block = !mb_test_bit(first - 1, EXT3_MB_BITMAP(e3b));
1386 +       if (first + count < EXT3_SB(sb)->s_mb_maxs[0])
1387 +               max = !mb_test_bit(first + count, EXT3_MB_BITMAP(e3b));
1388 +       if (block && max)
1389 +               e3b->bd_info->bb_fragments--;
1390 +       else if (!block && !max)
1391 +               e3b->bd_info->bb_fragments++;
1392 +
1393 +       /* let's maintain buddy itself */
1394 +       while (count-- > 0) {
1395 +               block = first++;
1396 +               order = 0;
1397 +
1398 +               if (!mb_test_bit(block, EXT3_MB_BITMAP(e3b))) {
1399 +                       unsigned long blocknr;
1400 +                       blocknr = e3b->bd_group * EXT3_BLOCKS_PER_GROUP(sb);
1401 +                       blocknr += block;
1402 +                       blocknr +=
1403 +                           le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block);
1404 +
1405 +                       ext3_error(sb, __FUNCTION__, "double-free of inode"
1406 +                                  " %lu's block %lu(bit %u in group %u)\n",
1407 +                                  inode ? inode->i_ino : 0, blocknr, block,
1408 +                                  e3b->bd_group);
1409 +               }
1410 +               mb_clear_bit(block, EXT3_MB_BITMAP(e3b));
1411 +               e3b->bd_info->bb_counters[order]++;
1412 +
1413 +               /* start of the buddy */
1414 +               buddy = mb_find_buddy(e3b, order, &max);
1415 +
1416 +               do {
1417 +                       block &= ~1UL;
1418 +                       if (mb_test_bit(block, buddy) ||
1419 +                                       mb_test_bit(block + 1, buddy))
1420 +                               break;
1421 +
1422 +                       /* both the buddies are free, try to coalesce them */
1423 +                       buddy2 = mb_find_buddy(e3b, order + 1, &max);
1424 +
1425 +                       if (!buddy2)
1426 +                               break;
1427 +
1428 +                       if (order > 0) {
1429 +                               /* for special purposes, we don't set
1430 +                                * free bits in bitmap */
1431 +                               mb_set_bit(block, buddy);
1432 +                               mb_set_bit(block + 1, buddy);
1433 +                       }
1434 +                       e3b->bd_info->bb_counters[order]--;
1435 +                       e3b->bd_info->bb_counters[order]--;
1436 +
1437 +                       block = block >> 1;
1438 +                       order++;
1439 +                       e3b->bd_info->bb_counters[order]++;
1440 +
1441 +                       mb_clear_bit(block, buddy2);
1442 +                       buddy = buddy2;
1443 +               } while (1);
1444 +       }
1445 +       mb_check_buddy(e3b);
1446 +
1447 +       return 0;
1448 +}
1449 +
1450 +static int mb_find_extent(struct ext3_buddy *e3b, int order, int block,
1451 +                               int needed, struct ext3_free_extent *ex)
1452 +{
1453 +       int next = block, max, ord;
1454 +       void *buddy;
1455 +
1456 +       BUG_ON(!ext3_is_group_locked(e3b->bd_sb, e3b->bd_group));
1457 +       BUG_ON(ex == NULL);
1458 +
1459 +       buddy = mb_find_buddy(e3b, order, &max);
1460 +       BUG_ON(buddy == NULL);
1461 +       BUG_ON(block >= max);
1462 +       if (mb_test_bit(block, buddy)) {
1463 +               ex->fe_len = 0;
1464 +               ex->fe_start = 0;
1465 +               ex->fe_group = 0;
1466 +               return 0;
1467 +       }
1468 +
1469 +       if (likely(order == 0)) {
1470 +               /* find actual order */
1471 +               order = mb_find_order_for_block(e3b, block);
1472 +               block = block >> order;
1473 +       }
1474 +
1475 +       ex->fe_len = 1 << order;
1476 +       ex->fe_start = block << order;
1477 +       ex->fe_group = e3b->bd_group;
1478 +
1479 +       /* calc difference from given start */
1480 +       next = next - ex->fe_start;
1481 +       ex->fe_len -= next;
1482 +       ex->fe_start += next;
1483 +
1484 +       while (needed > ex->fe_len && (buddy = mb_find_buddy(e3b, order, &max))) {
1485 +
1486 +               if (block + 1 >= max)
1487 +                       break;
1488 +
1489 +               next = (block + 1) * (1 << order);
1490 +               if (mb_test_bit(next, EXT3_MB_BITMAP(e3b)))
1491 +                       break;
1492 +
1493 +               ord = mb_find_order_for_block(e3b, next);
1494 +
1495 +               order = ord;
1496 +               block = next >> order;
1497 +               ex->fe_len += 1 << order;
1498 +       }
1499 +
1500 +       BUG_ON(ex->fe_start + ex->fe_len > (1 << (e3b->bd_blkbits + 3)));
1501 +       return ex->fe_len;
1502 +}
1503 +
1504 +static int mb_mark_used(struct ext3_buddy *e3b, struct ext3_free_extent *ex)
1505 +{
1506 +       int ord, mlen = 0, max = 0, cur;
1507 +       int start = ex->fe_start;
1508 +       int len = ex->fe_len;
1509 +       unsigned ret = 0;
1510 +       int len0 = len;
1511 +       void *buddy;
1512 +
1513 +       BUG_ON(start + len > (e3b->bd_sb->s_blocksize << 3));
1514 +       BUG_ON(e3b->bd_group != ex->fe_group);
1515 +       BUG_ON(!ext3_is_group_locked(e3b->bd_sb, e3b->bd_group));
1516 +       mb_check_buddy(e3b);
1517 +       mb_mark_used_double(e3b, start, len);
1518 +
1519 +       e3b->bd_info->bb_free -= len;
1520 +       if (e3b->bd_info->bb_first_free == start)
1521 +               e3b->bd_info->bb_first_free += len;
1522 +
1523 +       /* let's maintain fragments counter */
1524 +       if (start != 0)
1525 +               mlen = !mb_test_bit(start - 1, EXT3_MB_BITMAP(e3b));
1526 +       if (start + len < EXT3_SB(e3b->bd_sb)->s_mb_maxs[0])
1527 +               max = !mb_test_bit(start + len, EXT3_MB_BITMAP(e3b));
1528 +       if (mlen && max)
1529 +               e3b->bd_info->bb_fragments++;
1530 +       else if (!mlen && !max)
1531 +               e3b->bd_info->bb_fragments--;
1532 +
1533 +       /* let's maintain buddy itself */
1534 +       while (len) {
1535 +               ord = mb_find_order_for_block(e3b, start);
1536 +
1537 +               if (((start >> ord) << ord) == start && len >= (1 << ord)) {
1538 +                       /* the whole chunk may be allocated at once! */
1539 +                       mlen = 1 << ord;
1540 +                       buddy = mb_find_buddy(e3b, ord, &max);
1541 +                       BUG_ON((start >> ord) >= max);
1542 +                       mb_set_bit(start >> ord, buddy);
1543 +                       e3b->bd_info->bb_counters[ord]--;
1544 +                       start += mlen;
1545 +                       len -= mlen;
1546 +                       BUG_ON(len < 0);
1547 +                       continue;
1548 +               }
1549 +
1550 +               /* store for history */
1551 +               if (ret == 0)
1552 +                       ret = len | (ord << 16);
1553 +
1554 +               /* we have to split large buddy */
1555 +               BUG_ON(ord <= 0);
1556 +               buddy = mb_find_buddy(e3b, ord, &max);
1557 +               mb_set_bit(start >> ord, buddy);
1558 +               e3b->bd_info->bb_counters[ord]--;
1559 +
1560 +               ord--;
1561 +               cur = (start >> ord) & ~1U;
1562 +               buddy = mb_find_buddy(e3b, ord, &max);
1563 +               mb_clear_bit(cur, buddy);
1564 +               mb_clear_bit(cur + 1, buddy);
1565 +               e3b->bd_info->bb_counters[ord]++;
1566 +               e3b->bd_info->bb_counters[ord]++;
1567 +       }
1568 +
1569 +       mb_set_bits(EXT3_MB_BITMAP(e3b), ex->fe_start, len0);
1570 +       mb_check_buddy(e3b);
1571 +
1572 +       return ret;
1573 +}
1574 +
1575 +/*
1576 + * Must be called under group lock!
1577 + */
1578 +static void ext3_mb_use_best_found(struct ext3_allocation_context *ac,
1579 +                                       struct ext3_buddy *e3b)
1580 +{
1581 +       unsigned long ret;
1582 +
1583 +       BUG_ON(ac->ac_b_ex.fe_group != e3b->bd_group);
1584 +       BUG_ON(ac->ac_status == AC_STATUS_FOUND);
1585 +
1586 +       ac->ac_b_ex.fe_len = min(ac->ac_b_ex.fe_len, ac->ac_g_ex.fe_len);
1587 +       ac->ac_b_ex.fe_logical = ac->ac_g_ex.fe_logical;
1588 +       ret = mb_mark_used(e3b, &ac->ac_b_ex);
1589 +
1590 +       /* preallocation can change ac_b_ex, thus we store actually
1591 +        * allocated blocks for history */
1592 +       ac->ac_f_ex = ac->ac_b_ex;
1593 +
1594 +       ac->ac_status = AC_STATUS_FOUND;
1595 +       ac->ac_tail = ret & 0xffff;
1596 +       ac->ac_buddy = ret >> 16;
1597 +
1598 +       /* XXXXXXX: SUCH A HORRIBLE **CK */
1599 +       ac->ac_bitmap_page = e3b->bd_bitmap_page;
1600 +       get_page(ac->ac_bitmap_page);
1601 +       ac->ac_buddy_page = e3b->bd_buddy_page;
1602 +       get_page(ac->ac_buddy_page);
1603 +}
1604 +
1605 +/*
1606 + * regular allocator, for general purposes allocation
1607 + */
1608 +
1609 +void ext3_mb_check_limits(struct ext3_allocation_context *ac,
1610 +                                       struct ext3_buddy *e3b,
1611 +                                       int finish_group)
1612 +{
1613 +       struct ext3_sb_info *sbi = EXT3_SB(ac->ac_sb);
1614 +       struct ext3_free_extent *bex = &ac->ac_b_ex;
1615 +       struct ext3_free_extent *gex = &ac->ac_g_ex;
1616 +       struct ext3_free_extent ex;
1617 +       int max;
1618 +
1619 +       /*
1620 +        * We don't want to scan for a whole year
1621 +        */
1622 +       if (ac->ac_found > sbi->s_mb_max_to_scan &&
1623 +                       !(ac->ac_flags & EXT3_MB_HINT_FIRST)) {
1624 +               ac->ac_status = AC_STATUS_BREAK;
1625 +               return;
1626 +       }
1627 +
1628 +       /*
1629 +        * Haven't found good chunk so far, let's continue
1630 +        */
1631 +       if (bex->fe_len < gex->fe_len)
1632 +               return;
1633 +
1634 +       if ((finish_group || ac->ac_found > sbi->s_mb_min_to_scan)
1635 +                       && bex->fe_group == e3b->bd_group) {
1636 +               /* recheck chunk's availability - we don't know
1637 +                * when it was found (within this lock-unlock
1638 +                * period or not) */
1639 +               max = mb_find_extent(e3b, 0, bex->fe_start, gex->fe_len, &ex);
1640 +               if (max >= gex->fe_len) {
1641 +                       ext3_mb_use_best_found(ac, e3b);
1642 +                       return;
1643 +               }
1644 +       }
1645 +}
1646 +
1647 +/*
1648 + * The routine checks whether found extent is good enough. If it is,
1649 + * then the extent gets marked used and flag is set to the context
1650 + * to stop scanning. Otherwise, the extent is compared with the
1651 + * previous found extent and if new one is better, then it's stored
1652 + * in the context. Later, the best found extent will be used, if
1653 + * mballoc can't find good enough extent.
1654 + *
1655 + * FIXME: real allocation policy is to be designed yet!
1656 + */
1657 +static void ext3_mb_measure_extent(struct ext3_allocation_context *ac,
1658 +                                       struct ext3_free_extent *ex,
1659 +                                       struct ext3_buddy *e3b)
1660 +{
1661 +       struct ext3_free_extent *bex = &ac->ac_b_ex;
1662 +       struct ext3_free_extent *gex = &ac->ac_g_ex;
1663 +
1664 +       BUG_ON(ex->fe_len <= 0);
1665 +       BUG_ON(ex->fe_len >= (1 << ac->ac_sb->s_blocksize_bits) * 8);
1666 +       BUG_ON(ex->fe_start >= (1 << ac->ac_sb->s_blocksize_bits) * 8);
1667 +       BUG_ON(ac->ac_status != AC_STATUS_CONTINUE);
1668 +
1669 +       ac->ac_found++;
1670 +
1671 +       /*
1672 +        * The special case - take what you catch first
1673 +        */
1674 +       if (unlikely(ac->ac_flags & EXT3_MB_HINT_FIRST)) {
1675 +               *bex = *ex;
1676 +               ext3_mb_use_best_found(ac, e3b);
1677 +               return;
1678 +       }
1679 +
1680 +       /*
1681 +        * Let's check whether the chuck is good enough
1682 +        */
1683 +       if (ex->fe_len == gex->fe_len) {
1684 +               *bex = *ex;
1685 +               ext3_mb_use_best_found(ac, e3b);
1686 +               return;
1687 +       }
1688 +
1689 +       /*
1690 +        * If this is first found extent, just store it in the context
1691 +        */
1692 +       if (bex->fe_len == 0) {
1693 +               *bex = *ex;
1694 +               return;
1695 +       }
1696 +
1697 +       /*
1698 +        * If new found extent is better, store it in the context
1699 +        */
1700 +       if (bex->fe_len < gex->fe_len) {
1701 +               /* if the request isn't satisfied, any found extent
1702 +                * larger than previous best one is better */
1703 +               if (ex->fe_len > bex->fe_len)
1704 +                       *bex = *ex;
1705 +       } else if (ex->fe_len > gex->fe_len) {
1706 +               /* if the request is satisfied, then we try to find
1707 +                * an extent that still satisfy the request, but is
1708 +                * smaller than previous one */
1709 +               *bex = *ex;
1710 +       }
1711 +
1712 +       ext3_mb_check_limits(ac, e3b, 0);
1713 +}
1714 +
1715 +static int ext3_mb_try_best_found(struct ext3_allocation_context *ac,
1716 +                                       struct ext3_buddy *e3b)
1717 +{
1718 +       struct ext3_free_extent ex = ac->ac_b_ex;
1719 +       int group = ex.fe_group, max, err;
1720 +
1721 +       BUG_ON(ex.fe_len <= 0);
1722 +       err = ext3_mb_load_buddy(ac->ac_sb, group, e3b);
1723 +       if (err)
1724 +               return err;
1725 +
1726 +       ext3_lock_group(ac->ac_sb, group);
1727 +       max = mb_find_extent(e3b, 0, ex.fe_start, ex.fe_len, &ex);
1728 +
1729 +       if (max > 0) {
1730 +               ac->ac_b_ex = ex;
1731 +               ext3_mb_use_best_found(ac, e3b);
1732 +       }
1733 +
1734 +       ext3_unlock_group(ac->ac_sb, group);
1735 +       ext3_mb_release_desc(e3b);
1736 +
1737 +       return 0;
1738 +}
1739 +
1740 +static int ext3_mb_find_by_goal(struct ext3_allocation_context *ac,
1741 +                               struct ext3_buddy *e3b)
1742 +{
1743 +       int group = ac->ac_g_ex.fe_group, max, err;
1744 +       struct ext3_sb_info *sbi = EXT3_SB(ac->ac_sb);
1745 +       struct ext3_super_block *es = sbi->s_es;
1746 +       struct ext3_free_extent ex;
1747 +
1748 +       err = ext3_mb_load_buddy(ac->ac_sb, group, e3b);
1749 +       if (err)
1750 +               return err;
1751 +
1752 +       ext3_lock_group(ac->ac_sb, group);
1753 +       max = mb_find_extent(e3b, 0, ac->ac_g_ex.fe_start,
1754 +                            ac->ac_g_ex.fe_len, &ex);
1755 +
1756 +       if (max >= ac->ac_g_ex.fe_len && ac->ac_g_ex.fe_len == sbi->s_stripe) {
1757 +               unsigned long start;
1758 +               start = (e3b->bd_group * EXT3_BLOCKS_PER_GROUP(ac->ac_sb) +
1759 +                       ex.fe_start + le32_to_cpu(es->s_first_data_block));
1760 +               if (start % sbi->s_stripe == 0) {
1761 +                       ac->ac_found++;
1762 +                       ac->ac_b_ex = ex;
1763 +                       ext3_mb_use_best_found(ac, e3b);
1764 +               }
1765 +       } else if (max >= ac->ac_g_ex.fe_len) {
1766 +               BUG_ON(ex.fe_len <= 0);
1767 +               BUG_ON(ex.fe_group != ac->ac_g_ex.fe_group);
1768 +               BUG_ON(ex.fe_start != ac->ac_g_ex.fe_start);
1769 +               ac->ac_found++;
1770 +               ac->ac_b_ex = ex;
1771 +               ext3_mb_use_best_found(ac, e3b);
1772 +       } else if (max > 0 && (ac->ac_flags & EXT3_MB_HINT_MERGE)) {
1773 +               /* Sometimes, caller may want to merge even small
1774 +                * number of blocks to an existing extent */
1775 +               BUG_ON(ex.fe_len <= 0);
1776 +               BUG_ON(ex.fe_group != ac->ac_g_ex.fe_group);
1777 +               BUG_ON(ex.fe_start != ac->ac_g_ex.fe_start);
1778 +               ac->ac_found++;
1779 +               ac->ac_b_ex = ex;
1780 +               ext3_mb_use_best_found(ac, e3b);
1781 +       }
1782 +       ext3_unlock_group(ac->ac_sb, group);
1783 +       ext3_mb_release_desc(e3b);
1784 +
1785 +       return 0;
1786 +}
1787 +
1788 +/*
1789 + * The routine scans buddy structures (not bitmap!) from given order
1790 + * to max order and tries to find big enough chunk to satisfy the req
1791 + */
1792 +static void ext3_mb_simple_scan_group(struct ext3_allocation_context *ac,
1793 +                                       struct ext3_buddy *e3b)
1794 +{
1795 +       struct super_block *sb = ac->ac_sb;
1796 +       struct ext3_group_info *grp = e3b->bd_info;
1797 +       void *buddy;
1798 +       int i, k, max;
1799 +
1800 +       BUG_ON(ac->ac_2order <= 0);
1801 +       for (i = ac->ac_2order; i <= sb->s_blocksize_bits + 1; i++) {
1802 +               if (grp->bb_counters[i] == 0)
1803 +                       continue;
1804 +
1805 +               buddy = mb_find_buddy(e3b, i, &max);
1806 +               BUG_ON(buddy == NULL);
1807 +
1808 +               k = mb_find_next_zero_bit(buddy, max, 0);
1809 +               BUG_ON(k >= max);
1810 +
1811 +               ac->ac_found++;
1812 +
1813 +               ac->ac_b_ex.fe_len = 1 << i;
1814 +               ac->ac_b_ex.fe_start = k << i;
1815 +               ac->ac_b_ex.fe_group = e3b->bd_group;
1816 +
1817 +               ext3_mb_use_best_found(ac, e3b);
1818 +
1819 +               BUG_ON(ac->ac_b_ex.fe_len != ac->ac_g_ex.fe_len);
1820 +
1821 +               if (EXT3_SB(sb)->s_mb_stats)
1822 +                       atomic_inc(&EXT3_SB(sb)->s_bal_2orders);
1823 +
1824 +               break;
1825 +       }
1826 +}
1827 +
1828 +/*
1829 + * The routine scans the group and measures all found extents.
1830 + * In order to optimize scanning, caller must pass number of
1831 + * free blocks in the group, so the routine can know upper limit.
1832 + */
1833 +static void ext3_mb_complex_scan_group(struct ext3_allocation_context *ac,
1834 +                                       struct ext3_buddy *e3b)
1835 +{
1836 +       struct super_block *sb = ac->ac_sb;
1837 +       void *bitmap = EXT3_MB_BITMAP(e3b);
1838 +       struct ext3_free_extent ex;
1839 +       int i, free;
1840 +
1841 +       free = e3b->bd_info->bb_free;
1842 +       BUG_ON(free <= 0);
1843 +
1844 +       i = e3b->bd_info->bb_first_free;
1845 +
1846 +       while (free && ac->ac_status == AC_STATUS_CONTINUE) {
1847 +               i = mb_find_next_zero_bit(bitmap, sb->s_blocksize * 8, i);
1848 +               if (i >= sb->s_blocksize * 8) {
1849 +                       BUG_ON(free != 0);
1850 +                       break;
1851 +               }
1852 +
1853 +               mb_find_extent(e3b, 0, i, ac->ac_g_ex.fe_len, &ex);
1854 +               BUG_ON(ex.fe_len <= 0);
1855 +               BUG_ON(free < ex.fe_len);
1856 +
1857 +               ext3_mb_measure_extent(ac, &ex, e3b);
1858 +
1859 +               i += ex.fe_len;
1860 +               free -= ex.fe_len;
1861 +       }
1862 +
1863 +       ext3_mb_check_limits(ac, e3b, 1);
1864 +}
1865 +
1866 +/*
1867 + * This is a special case for storages like raid5
1868 + * we try to find stripe-aligned chunks for stripe-size requests
1869 + */
1870 +static void ext3_mb_scan_aligned(struct ext3_allocation_context *ac,
1871 +                                struct ext3_buddy *e3b)
1872 +{
1873 +       struct super_block *sb = ac->ac_sb;
1874 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
1875 +       void *bitmap = EXT3_MB_BITMAP(e3b);
1876 +       struct ext3_free_extent ex;
1877 +       unsigned long i, max;
1878 +
1879 +       BUG_ON(sbi->s_stripe == 0);
1880 +
1881 +       /* find first stripe-aligned block */
1882 +       i = e3b->bd_group * EXT3_BLOCKS_PER_GROUP(sb)
1883 +               + le32_to_cpu(sbi->s_es->s_first_data_block);
1884 +       i = ((i + sbi->s_stripe - 1) / sbi->s_stripe) * sbi->s_stripe;
1885 +       i = (i - le32_to_cpu(sbi->s_es->s_first_data_block))
1886 +                       % EXT3_BLOCKS_PER_GROUP(sb);
1887 +
1888 +       while (i < sb->s_blocksize * 8) {
1889 +               if (!mb_test_bit(i, bitmap)) {
1890 +                       max = mb_find_extent(e3b, 0, i, sbi->s_stripe, &ex);
1891 +                       if (max >= sbi->s_stripe) {
1892 +                               ac->ac_found++;
1893 +                               ac->ac_b_ex = ex;
1894 +                               ext3_mb_use_best_found(ac, e3b);
1895 +                               break;
1896 +                       }
1897 +               }
1898 +               i += sbi->s_stripe;
1899 +       }
1900 +}
1901 +
1902 +static int ext3_mb_good_group(struct ext3_allocation_context *ac,
1903 +                               int group, int cr)
1904 +{
1905 +       struct ext3_group_info *grp = EXT3_GROUP_INFO(ac->ac_sb, group);
1906 +       unsigned free, fragments, i, bits;
1907 +
1908 +       BUG_ON(cr < 0 || cr >= 4);
1909 +       BUG_ON(EXT3_MB_GRP_NEED_INIT(grp));
1910 +
1911 +       free = grp->bb_free;
1912 +       fragments = grp->bb_fragments;
1913 +       if (free == 0)
1914 +               return 0;
1915 +       if (fragments == 0)
1916 +               return 0;
1917 +
1918 +       switch (cr) {
1919 +               case 0:
1920 +                       BUG_ON(ac->ac_2order == 0);
1921 +                       bits = ac->ac_sb->s_blocksize_bits + 1;
1922 +                       for (i = ac->ac_2order; i <= bits; i++)
1923 +                               if (grp->bb_counters[i] > 0)
1924 +                                       return 1;
1925 +                       break;
1926 +               case 1:
1927 +                       if ((free / fragments) >= ac->ac_g_ex.fe_len)
1928 +                               return 1;
1929 +                       break;
1930 +               case 2:
1931 +                       if (free >= ac->ac_g_ex.fe_len)
1932 +                               return 1;
1933 +                       break;
1934 +               case 3:
1935 +                       return 1;
1936 +               default:
1937 +                       BUG();
1938 +       }
1939 +
1940 +       return 0;
1941 +}
1942 +
1943 +int ext3_mb_regular_allocator(struct ext3_allocation_context *ac)
1944 +{
1945 +       int group, i, cr, err = 0;
1946 +       struct ext3_sb_info *sbi;
1947 +       struct super_block *sb;
1948 +       struct ext3_buddy e3b;
1949 +
1950 +       sb = ac->ac_sb;
1951 +       sbi = EXT3_SB(sb);
1952 +       BUG_ON(ac->ac_status == AC_STATUS_FOUND);
1953 +
1954 +       /* first, try the goal */
1955 +       err = ext3_mb_find_by_goal(ac, &e3b);
1956 +       if (err || ac->ac_status == AC_STATUS_FOUND)
1957 +               goto out;
1958 +
1959 +       if (unlikely(ac->ac_flags & EXT3_MB_HINT_GOAL_ONLY))
1960 +               goto out;
1961 +
1962 +       i = ffs(ac->ac_g_ex.fe_len);
1963 +       ac->ac_2order = 0;
1964 +       if (i >= sbi->s_mb_order2_reqs) {
1965 +               i--;
1966 +               if ((ac->ac_g_ex.fe_len & (~(1 << i))) == 0)
1967 +                       ac->ac_2order = i;
1968 +       }
1969 +
1970 +       group = ac->ac_g_ex.fe_group;
1971 +
1972 +       /* Let's just scan groups to find more-less suitable blocks */
1973 +       cr = ac->ac_2order ? 0 : 1;
1974 +repeat:
1975 +       for (; cr < 4 && ac->ac_status == AC_STATUS_CONTINUE; cr++) {
1976 +               ac->ac_criteria = cr;
1977 +               for (i = 0; i < EXT3_SB(sb)->s_groups_count; group++, i++) {
1978 +                       struct ext3_group_info *grp;
1979 +
1980 +                       if (group == EXT3_SB(sb)->s_groups_count)
1981 +                               group = 0;
1982 +
1983 +                       /* quick check to skip empty groups */
1984 +                       grp = EXT3_GROUP_INFO(ac->ac_sb, group);
1985 +                       if (grp->bb_free == 0)
1986 +                               continue;
1987 +
1988 +                       if (EXT3_MB_GRP_NEED_INIT(EXT3_GROUP_INFO(sb, group))) {
1989 +                               /* we need full data about the group
1990 +                                * to make a good selection */
1991 +                               err = ext3_mb_load_buddy(sb, group, &e3b);
1992 +                               if (err)
1993 +                                       goto out;
1994 +                               ext3_mb_release_desc(&e3b);
1995 +                       }
1996 +
1997 +                       /* check is group good for our criteries */
1998 +                       if (!ext3_mb_good_group(ac, group, cr))
1999 +                               continue;
2000 +
2001 +                       err = ext3_mb_load_buddy(sb, group, &e3b);
2002 +                       if (err)
2003 +                               goto out;
2004 +
2005 +                       ext3_lock_group(sb, group);
2006 +                       if (!ext3_mb_good_group(ac, group, cr)) {
2007 +                               /* someone did allocation from this group */
2008 +                               ext3_unlock_group(sb, group);
2009 +                               ext3_mb_release_desc(&e3b);
2010 +                               continue;
2011 +                       }
2012 +
2013 +                       ac->ac_groups_scanned++;
2014 +                       if (cr == 0)
2015 +                               ext3_mb_simple_scan_group(ac, &e3b);
2016 +                       else if (cr == 1 && ac->ac_g_ex.fe_len == sbi->s_stripe)
2017 +                               ext3_mb_scan_aligned(ac, &e3b);
2018 +                       else
2019 +                               ext3_mb_complex_scan_group(ac, &e3b);
2020 +
2021 +                       ext3_unlock_group(sb, group);
2022 +                       ext3_mb_release_desc(&e3b);
2023 +
2024 +                       if (ac->ac_status != AC_STATUS_CONTINUE)
2025 +                               break;
2026 +               }
2027 +       }
2028 +
2029 +       if (ac->ac_b_ex.fe_len > 0 && ac->ac_status != AC_STATUS_FOUND &&
2030 +           !(ac->ac_flags & EXT3_MB_HINT_FIRST)) {
2031 +               /*
2032 +                * We've been searching too long. Let's try to allocate
2033 +                * the best chunk we've found so far
2034 +                */
2035 +
2036 +               ext3_mb_try_best_found(ac, &e3b);
2037 +               if (ac->ac_status != AC_STATUS_FOUND) {
2038 +                       /*
2039 +                        * Someone more lucky has already allocated it.
2040 +                        * The only thing we can do is just take first
2041 +                        * found block(s)
2042 +                       printk(KERN_DEBUG "EXT3-fs: someone won our chunk\n");
2043 +                        */
2044 +                       ac->ac_b_ex.fe_group = 0;
2045 +                       ac->ac_b_ex.fe_start = 0;
2046 +                       ac->ac_b_ex.fe_len = 0;
2047 +                       ac->ac_status = AC_STATUS_CONTINUE;
2048 +                       ac->ac_flags |= EXT3_MB_HINT_FIRST;
2049 +                       cr = 3;
2050 +                       atomic_inc(&sbi->s_mb_lost_chunks);
2051 +                       goto repeat;
2052 +               }
2053 +       }
2054 +out:
2055 +       return err;
2056 +}
2057 +
2058 +#ifdef EXT3_MB_HISTORY
2059 +struct ext3_mb_proc_session {
2060 +       struct ext3_mb_history *history;
2061 +       struct super_block *sb;
2062 +       int start;
2063 +       int max;
2064 +};
2065 +
2066 +static void *ext3_mb_history_skip_empty(struct ext3_mb_proc_session *s,
2067 +                                       struct ext3_mb_history *hs,
2068 +                                       int first)
2069 +{
2070 +       if (hs == s->history + s->max)
2071 +               hs = s->history;
2072 +       if (!first && hs == s->history + s->start)
2073 +               return NULL;
2074 +       while (hs->orig.fe_len == 0) {
2075 +               hs++;
2076 +               if (hs == s->history + s->max)
2077 +                       hs = s->history;
2078 +               if (hs == s->history + s->start)
2079 +                       return NULL;
2080 +       }
2081 +       return hs;
2082 +}
2083 +
2084 +static void *ext3_mb_seq_history_start(struct seq_file *seq, loff_t *pos)
2085 +{
2086 +       struct ext3_mb_proc_session *s = seq->private;
2087 +       struct ext3_mb_history *hs;
2088 +       int l = *pos;
2089 +
2090 +       if (l == 0)
2091 +               return SEQ_START_TOKEN;
2092 +       hs = ext3_mb_history_skip_empty(s, s->history + s->start, 1);
2093 +       if (!hs)
2094 +               return NULL;
2095 +       while (--l && (hs = ext3_mb_history_skip_empty(s, ++hs, 0)) != NULL);
2096 +       return hs;
2097 +}
2098 +
2099 +static void *ext3_mb_seq_history_next(struct seq_file *seq, void *v, loff_t *pos)
2100 +{
2101 +       struct ext3_mb_proc_session *s = seq->private;
2102 +       struct ext3_mb_history *hs = v;
2103 +
2104 +       ++*pos;
2105 +       if (v == SEQ_START_TOKEN)
2106 +               return ext3_mb_history_skip_empty(s, s->history + s->start, 1);
2107 +       else
2108 +               return ext3_mb_history_skip_empty(s, ++hs, 0);
2109 +}
2110 +
2111 +static int ext3_mb_seq_history_show(struct seq_file *seq, void *v)
2112 +{
2113 +       char buf[25], buf2[25], buf3[25], *fmt;
2114 +       struct ext3_mb_history *hs = v;
2115 +
2116 +       if (v == SEQ_START_TOKEN) {
2117 +               seq_printf(seq, "%-5s %-8s %-23s %-23s %-23s %-5s "
2118 +                               "%-5s %-2s %-5s %-5s %-5s %-6s\n",
2119 +                          "pid", "inode", "original", "goal", "result","found",
2120 +                          "grps", "cr", "flags", "merge", "tail", "broken");
2121 +               return 0;
2122 +       }
2123 +
2124 +       if (hs->op == EXT3_MB_HISTORY_ALLOC) {
2125 +               fmt = "%-5u %-8u %-23s %-23s %-23s %-5u %-5u %-2u "
2126 +                       "%-5u %-5s %-5u %-6u\n";
2127 +               sprintf(buf2, "%lu/%lu/%lu@%lu", hs->result.fe_group,
2128 +                       hs->result.fe_start, hs->result.fe_len,
2129 +                       hs->result.fe_logical);
2130 +               sprintf(buf, "%lu/%lu/%lu@%lu", hs->orig.fe_group,
2131 +                       hs->orig.fe_start, hs->orig.fe_len,
2132 +                       hs->orig.fe_logical);
2133 +               sprintf(buf3, "%lu/%lu/%lu@%lu", hs->goal.fe_group,
2134 +                       hs->goal.fe_start, hs->goal.fe_len,
2135 +                       hs->goal.fe_logical);
2136 +               seq_printf(seq, fmt, hs->pid, hs->ino, buf, buf3, buf2,
2137 +                               hs->found, hs->groups, hs->cr, hs->flags,
2138 +                               hs->merged ? "M" : "", hs->tail,
2139 +                               hs->buddy ? 1 << hs->buddy : 0);
2140 +       } else if (hs->op == EXT3_MB_HISTORY_PREALLOC) {
2141 +               fmt = "%-5u %-8u %-23s %-23s %-23s\n";
2142 +               sprintf(buf2, "%lu/%lu/%lu@%lu", hs->result.fe_group,
2143 +                       hs->result.fe_start, hs->result.fe_len,
2144 +                       hs->result.fe_logical);
2145 +               sprintf(buf, "%lu/%lu/%lu@%lu", hs->orig.fe_group,
2146 +                       hs->orig.fe_start, hs->orig.fe_len,
2147 +                       hs->orig.fe_logical);
2148 +               seq_printf(seq, fmt, hs->pid, hs->ino, buf, "", buf2);
2149 +       } else if (hs->op == EXT3_MB_HISTORY_DISCARD) {
2150 +               sprintf(buf2, "%lu/%lu/%lu", hs->result.fe_group,
2151 +                       hs->result.fe_start, hs->result.fe_len);
2152 +               seq_printf(seq, "%-5u %-8u %-23s discard\n",
2153 +                               hs->pid, hs->ino, buf2);
2154 +       } else if (hs->op == EXT3_MB_HISTORY_FREE) {
2155 +               sprintf(buf2, "%lu/%lu/%lu", hs->result.fe_group,
2156 +                       hs->result.fe_start, hs->result.fe_len);
2157 +               seq_printf(seq, "%-5u %-8u %-23s free\n",
2158 +                               hs->pid, hs->ino, buf2);
2159 +       }
2160 +       return 0;
2161 +}
2162 +
2163 +static void ext3_mb_seq_history_stop(struct seq_file *seq, void *v)
2164 +{
2165 +}
2166 +
2167 +static struct seq_operations ext3_mb_seq_history_ops = {
2168 +       .start  = ext3_mb_seq_history_start,
2169 +       .next   = ext3_mb_seq_history_next,
2170 +       .stop   = ext3_mb_seq_history_stop,
2171 +       .show   = ext3_mb_seq_history_show,
2172 +};
2173 +
2174 +static int ext3_mb_seq_history_open(struct inode *inode, struct file *file)
2175 +{
2176 +       struct super_block *sb = PDE(inode)->data;
2177 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
2178 +       struct ext3_mb_proc_session *s;
2179 +       int rc, size;
2180 +
2181 +       s = kmalloc(sizeof(*s), GFP_KERNEL);
2182 +       if (s == NULL)
2183 +               return -ENOMEM;
2184 +       s->sb = sb;
2185 +       size = sizeof(struct ext3_mb_history) * sbi->s_mb_history_max;
2186 +       s->history = kmalloc(size, GFP_KERNEL);
2187 +       if (s->history == NULL) {
2188 +               kfree(s);
2189 +               return -ENOMEM;
2190 +       }
2191 +
2192 +       spin_lock(&sbi->s_mb_history_lock);
2193 +       memcpy(s->history, sbi->s_mb_history, size);
2194 +       s->max = sbi->s_mb_history_max;
2195 +       s->start = sbi->s_mb_history_cur % s->max;
2196 +       spin_unlock(&sbi->s_mb_history_lock);
2197 +
2198 +       rc = seq_open(file, &ext3_mb_seq_history_ops);
2199 +       if (rc == 0) {
2200 +               struct seq_file *m = (struct seq_file *)file->private_data;
2201 +               m->private = s;
2202 +       } else {
2203 +               kfree(s->history);
2204 +               kfree(s);
2205 +       }
2206 +       return rc;
2207 +
2208 +}
2209 +
2210 +static int ext3_mb_seq_history_release(struct inode *inode, struct file *file)
2211 +{
2212 +       struct seq_file *seq = (struct seq_file *)file->private_data;
2213 +       struct ext3_mb_proc_session *s = seq->private;
2214 +       kfree(s->history);
2215 +       kfree(s);
2216 +       return seq_release(inode, file);
2217 +}
2218 +
2219 +static ssize_t ext3_mb_seq_history_write(struct file *file,
2220 +                               const char __user *buffer,
2221 +                               size_t count, loff_t *ppos)
2222 +{
2223 +       struct seq_file *seq = (struct seq_file *)file->private_data;
2224 +       struct ext3_mb_proc_session *s = seq->private;
2225 +       struct super_block *sb = s->sb;
2226 +       char str[32];
2227 +       int value;
2228 +
2229 +       if (count >= sizeof(str)) {
2230 +               printk(KERN_ERR "EXT3-fs: %s string too long, max %u bytes\n",
2231 +                               "mb_history", (int)sizeof(str));
2232 +               return -EOVERFLOW;
2233 +       }
2234 +
2235 +       if (copy_from_user(str, buffer, count))
2236 +               return -EFAULT;
2237 +
2238 +       value = simple_strtol(str, NULL, 0);
2239 +       if (value < 0)
2240 +               return -ERANGE;
2241 +       EXT3_SB(sb)->s_mb_history_filter = value;
2242 +
2243 +       return count;
2244 +}
2245 +
2246 +static struct file_operations ext3_mb_seq_history_fops = {
2247 +       .owner          = THIS_MODULE,
2248 +       .open           = ext3_mb_seq_history_open,
2249 +       .read           = seq_read,
2250 +       .write          = ext3_mb_seq_history_write,
2251 +       .llseek         = seq_lseek,
2252 +       .release        = ext3_mb_seq_history_release,
2253 +};
2254 +
2255 +static void *ext3_mb_seq_groups_start(struct seq_file *seq, loff_t *pos)
2256 +{
2257 +       struct super_block *sb = seq->private;
2258 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
2259 +       long group;
2260 +
2261 +       if (*pos < 0 || *pos >= sbi->s_groups_count)
2262 +               return NULL;
2263 +
2264 +       group = *pos + 1;
2265 +       return (void *) group;
2266 +}
2267 +
2268 +static void *ext3_mb_seq_groups_next(struct seq_file *seq, void *v, loff_t *pos)
2269 +{
2270 +       struct super_block *sb = seq->private;
2271 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
2272 +       long group;
2273 +
2274 +       ++*pos;
2275 +       if (*pos < 0 || *pos >= sbi->s_groups_count)
2276 +               return NULL;
2277 +       group = *pos + 1;
2278 +       return (void *) group;;
2279 +}
2280 +
2281 +static int ext3_mb_seq_groups_show(struct seq_file *seq, void *v)
2282 +{
2283 +       struct super_block *sb = seq->private;
2284 +       long group = (long) v;
2285 +       int i, err;
2286 +       struct ext3_buddy e3b;
2287 +       struct sg {
2288 +               struct ext3_group_info info;
2289 +               unsigned short counters[16];
2290 +       } sg;
2291 +
2292 +       group--;
2293 +       if (group == 0)
2294 +               seq_printf(seq, "#%-5s: %-5s %-5s %-5s "
2295 +                               "[ %-5s %-5s %-5s %-5s %-5s %-5s %-5s "
2296 +                                 "%-5s %-5s %-5s %-5s %-5s %-5s %-5s ]\n",
2297 +                          "group", "free", "frags", "first",
2298 +                          "2^0", "2^1", "2^2", "2^3", "2^4", "2^5","2^6",
2299 +                          "2^7", "2^8", "2^9", "2^10", "2^11", "2^12", "2^13");
2300 +
2301 +       i = (sb->s_blocksize_bits + 2) * sizeof(sg.info.bb_counters[0]) +
2302 +               sizeof(struct ext3_group_info);
2303 +       err = ext3_mb_load_buddy(sb, group, &e3b);
2304 +       if (err) {
2305 +               seq_printf(seq, "#%-5lu: I/O error\n", group);
2306 +               return 0;
2307 +       }
2308 +       ext3_lock_group(sb, group);
2309 +       memcpy(&sg, EXT3_GROUP_INFO(sb, group), i);
2310 +       ext3_unlock_group(sb, group);
2311 +       ext3_mb_release_desc(&e3b);
2312 +
2313 +       seq_printf(seq, "#%-5lu: %-5u %-5u %-5u [", group, sg.info.bb_free,
2314 +                       sg.info.bb_fragments, sg.info.bb_first_free);
2315 +       for (i = 0; i <= 13; i++)
2316 +               seq_printf(seq, " %-5u", i <= sb->s_blocksize_bits + 1 ?
2317 +                               sg.info.bb_counters[i] : 0);
2318 +       seq_printf(seq, " ]\n");
2319 +
2320 +       return 0;
2321 +}
2322 +
2323 +static void ext3_mb_seq_groups_stop(struct seq_file *seq, void *v)
2324 +{
2325 +}
2326 +
2327 +static struct seq_operations ext3_mb_seq_groups_ops = {
2328 +       .start  = ext3_mb_seq_groups_start,
2329 +       .next   = ext3_mb_seq_groups_next,
2330 +       .stop   = ext3_mb_seq_groups_stop,
2331 +       .show   = ext3_mb_seq_groups_show,
2332 +};
2333 +
2334 +static int ext3_mb_seq_groups_open(struct inode *inode, struct file *file)
2335 +{
2336 +       struct super_block *sb = PDE(inode)->data;
2337 +       int rc;
2338 +
2339 +       rc = seq_open(file, &ext3_mb_seq_groups_ops);
2340 +       if (rc == 0) {
2341 +               struct seq_file *m = (struct seq_file *)file->private_data;
2342 +               m->private = sb;
2343 +       }
2344 +       return rc;
2345 +
2346 +}
2347 +
2348 +static struct file_operations ext3_mb_seq_groups_fops = {
2349 +       .owner          = THIS_MODULE,
2350 +       .open           = ext3_mb_seq_groups_open,
2351 +       .read           = seq_read,
2352 +       .llseek         = seq_lseek,
2353 +       .release        = seq_release,
2354 +};
2355 +
2356 +static void ext3_mb_history_release(struct super_block *sb)
2357 +{
2358 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
2359 +
2360 +       remove_proc_entry("mb_groups", sbi->s_mb_proc);
2361 +       remove_proc_entry("mb_history", sbi->s_mb_proc);
2362 +
2363 +       if (sbi->s_mb_history)
2364 +               kfree(sbi->s_mb_history);
2365 +}
2366 +
2367 +static void ext3_mb_history_init(struct super_block *sb)
2368 +{
2369 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
2370 +       int i;
2371 +
2372 +       if (sbi->s_mb_proc != NULL) {
2373 +               struct proc_dir_entry *p;
2374 +               p = create_proc_entry("mb_history", S_IRUGO, sbi->s_mb_proc);
2375 +               if (p) {
2376 +                       p->proc_fops = &ext3_mb_seq_history_fops;
2377 +                       p->data = sb;
2378 +               }
2379 +               p = create_proc_entry("mb_groups", S_IRUGO, sbi->s_mb_proc);
2380 +               if (p) {
2381 +                       p->proc_fops = &ext3_mb_seq_groups_fops;
2382 +                       p->data = sb;
2383 +               }
2384 +       }
2385 +
2386 +       sbi->s_mb_history_max = 1000;
2387 +       sbi->s_mb_history_cur = 0;
2388 +       spin_lock_init(&sbi->s_mb_history_lock);
2389 +       i = sbi->s_mb_history_max * sizeof(struct ext3_mb_history);
2390 +       sbi->s_mb_history = kmalloc(i, GFP_KERNEL);
2391 +       memset(sbi->s_mb_history, 0, i);
2392 +       /* if we can't allocate history, then we simple won't use it */
2393 +}
2394 +
2395 +static void
2396 +ext3_mb_store_history(struct ext3_allocation_context *ac)
2397 +{
2398 +       struct ext3_sb_info *sbi = EXT3_SB(ac->ac_sb);
2399 +       struct ext3_mb_history h;
2400 +
2401 +       if (likely(sbi->s_mb_history == NULL))
2402 +               return;
2403 +
2404 +       if (!(ac->ac_op & sbi->s_mb_history_filter))
2405 +               return;
2406 +
2407 +       h.op = ac->ac_op;
2408 +       h.pid = current->pid;
2409 +       h.ino = ac->ac_inode ? ac->ac_inode->i_ino : 0;
2410 +       h.orig = ac->ac_o_ex;
2411 +       h.result = ac->ac_b_ex;
2412 +       h.flags = ac->ac_flags;
2413 +       h.merged = 0;
2414 +       if (ac->ac_op == EXT3_MB_HISTORY_ALLOC) {
2415 +               if (ac->ac_g_ex.fe_start == ac->ac_b_ex.fe_start &&
2416 +                               ac->ac_g_ex.fe_group == ac->ac_b_ex.fe_group)
2417 +                       h.merged = 1;
2418 +               h.goal = ac->ac_g_ex;
2419 +               h.result = ac->ac_f_ex;
2420 +       }
2421 +
2422 +       spin_lock(&sbi->s_mb_history_lock);
2423 +       memcpy(sbi->s_mb_history + sbi->s_mb_history_cur, &h, sizeof(h));
2424 +       if (++sbi->s_mb_history_cur >= sbi->s_mb_history_max)
2425 +               sbi->s_mb_history_cur = 0;
2426 +       spin_unlock(&sbi->s_mb_history_lock);
2427 +}
2428 +
2429 +#else
2430 +#define ext3_mb_history_release(sb)
2431 +#define ext3_mb_history_init(sb)
2432 +#endif
2433 +
2434 +int ext3_mb_init_backend(struct super_block *sb)
2435 +{
2436 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
2437 +       int i, j, len, metalen;
2438 +       int num_meta_group_infos =
2439 +               (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) >>
2440 +                       EXT3_DESC_PER_BLOCK_BITS(sb);
2441 +       struct ext3_group_info **meta_group_info;
2442 +
2443 +       /* An 8TB filesystem with 64-bit pointers requires a 4096 byte
2444 +        * kmalloc. A 128kb malloc should suffice for a 256TB filesystem.
2445 +        * So a two level scheme suffices for now. */
2446 +       sbi->s_group_info = kmalloc(sizeof(*sbi->s_group_info) *
2447 +                                   num_meta_group_infos, GFP_KERNEL);
2448 +       if (sbi->s_group_info == NULL) {
2449 +               printk(KERN_ERR "EXT3-fs: can't allocate buddy meta group\n");
2450 +               return -ENOMEM;
2451 +       }
2452 +       sbi->s_buddy_cache = new_inode(sb);
2453 +       if (sbi->s_buddy_cache == NULL) {
2454 +               printk(KERN_ERR "EXT3-fs: can't get new inode\n");
2455 +               goto err_freesgi;
2456 +       }
2457 +       EXT3_I(sbi->s_buddy_cache)->i_disksize = 0;
2458 +
2459 +       metalen = sizeof(*meta_group_info) << EXT3_DESC_PER_BLOCK_BITS(sb);
2460 +       for (i = 0; i < num_meta_group_infos; i++) {
2461 +               if ((i + 1) == num_meta_group_infos)
2462 +                       metalen = sizeof(*meta_group_info) *
2463 +                               (sbi->s_groups_count -
2464 +                                       (i << EXT3_DESC_PER_BLOCK_BITS(sb)));
2465 +               meta_group_info = kmalloc(metalen, GFP_KERNEL);
2466 +               if (meta_group_info == NULL) {
2467 +                       printk(KERN_ERR "EXT3-fs: can't allocate mem for a "
2468 +                              "buddy group\n");
2469 +                       goto err_freemeta;
2470 +               }
2471 +               sbi->s_group_info[i] = meta_group_info;
2472 +       }
2473 +
2474 +       /*
2475 +        * calculate needed size. if change bb_counters size,
2476 +        * don't forget about ext3_mb_generate_buddy()
2477 +        */
2478 +       len = sizeof(struct ext3_group_info);
2479 +       len += sizeof(unsigned short) * (sb->s_blocksize_bits + 2);
2480 +       for (i = 0; i < sbi->s_groups_count; i++) {
2481 +               struct ext3_group_desc * desc;
2482 +
2483 +               meta_group_info =
2484 +                       sbi->s_group_info[i >> EXT3_DESC_PER_BLOCK_BITS(sb)];
2485 +               j = i & (EXT3_DESC_PER_BLOCK(sb) - 1);
2486 +
2487 +               meta_group_info[j] = kmalloc(len, GFP_KERNEL);
2488 +               if (meta_group_info[j] == NULL) {
2489 +                       printk(KERN_ERR "EXT3-fs: can't allocate buddy mem\n");
2490 +                       i--;
2491 +                       goto err_freebuddy;
2492 +               }
2493 +               desc = ext3_get_group_desc(sb, i, NULL);
2494 +               if (desc == NULL) {
2495 +                       printk(KERN_ERR"EXT3-fs: can't read descriptor %u\n",i);
2496 +                       goto err_freebuddy;
2497 +               }
2498 +               memset(meta_group_info[j], 0, len);
2499 +               set_bit(EXT3_GROUP_INFO_NEED_INIT_BIT,
2500 +                       &meta_group_info[j]->bb_state);
2501 +
2502 +               /* initialize bb_free to be able to skip
2503 +                * empty groups without initialization */
2504 +               meta_group_info[j]->bb_free =
2505 +                       le16_to_cpu(desc->bg_free_blocks_count);
2506 +
2507 +               INIT_LIST_HEAD(&meta_group_info[j]->bb_prealloc_list);
2508 +
2509 +#ifdef DOUBLE_CHECK
2510 +               {
2511 +                       struct buffer_head *bh;
2512 +                       meta_group_info[j]->bb_bitmap =
2513 +                               kmalloc(sb->s_blocksize, GFP_KERNEL);
2514 +                       BUG_ON(meta_group_info[j]->bb_bitmap == NULL);
2515 +                       bh = read_block_bitmap(sb, i);
2516 +                       BUG_ON(bh == NULL);
2517 +                       memcpy(meta_group_info[j]->bb_bitmap, bh->b_data,
2518 +                                       sb->s_blocksize);
2519 +                       brelse(bh);
2520 +               }
2521 +#endif
2522 +
2523 +       }
2524 +
2525 +       return 0;
2526 +
2527 +err_freebuddy:
2528 +       while (i >= 0) {
2529 +               kfree(EXT3_GROUP_INFO(sb, i));
2530 +               i--;
2531 +       }
2532 +       i = num_meta_group_infos;
2533 +err_freemeta:
2534 +       while (--i >= 0)
2535 +               kfree(sbi->s_group_info[i]);
2536 +       iput(sbi->s_buddy_cache);
2537 +err_freesgi:
2538 +       kfree(sbi->s_group_info);
2539 +       return -ENOMEM;
2540 +}
2541 +
2542 +int ext3_mb_init(struct super_block *sb, int needs_recovery)
2543 +{
2544 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
2545 +       unsigned i, offset, max;
2546 +
2547 +       if (!test_opt(sb, MBALLOC))
2548 +               return 0;
2549 +
2550 +       i = (sb->s_blocksize_bits + 2) * sizeof(unsigned short);
2551 +
2552 +       sbi->s_mb_offsets = kmalloc(i, GFP_KERNEL);
2553 +       if (sbi->s_mb_offsets == NULL) {
2554 +               clear_opt(sbi->s_mount_opt, MBALLOC);
2555 +               return -ENOMEM;
2556 +       }
2557 +       sbi->s_mb_maxs = kmalloc(i, GFP_KERNEL);
2558 +       if (sbi->s_mb_maxs == NULL) {
2559 +               clear_opt(sbi->s_mount_opt, MBALLOC);
2560 +               kfree(sbi->s_mb_maxs);
2561 +               return -ENOMEM;
2562 +       }
2563 +
2564 +       /* order 0 is regular bitmap */
2565 +       sbi->s_mb_maxs[0] = sb->s_blocksize << 3;
2566 +       sbi->s_mb_offsets[0] = 0;
2567 +
2568 +       i = 1;
2569 +       offset = 0;
2570 +       max = sb->s_blocksize << 2;
2571 +       do {
2572 +               sbi->s_mb_offsets[i] = offset;
2573 +               sbi->s_mb_maxs[i] = max;
2574 +               offset += 1 << (sb->s_blocksize_bits - i);
2575 +               max = max >> 1;
2576 +               i++;
2577 +       } while (i <= sb->s_blocksize_bits + 1);
2578 +
2579 +       /* init file for buddy data */
2580 +       if ((i = ext3_mb_init_backend(sb))) {
2581 +               clear_opt(sbi->s_mount_opt, MBALLOC);
2582 +               kfree(sbi->s_mb_offsets);
2583 +               kfree(sbi->s_mb_maxs);
2584 +               return i;
2585 +       }
2586 +
2587 +       spin_lock_init(&sbi->s_md_lock);
2588 +       INIT_LIST_HEAD(&sbi->s_active_transaction);
2589 +       INIT_LIST_HEAD(&sbi->s_closed_transaction);
2590 +       INIT_LIST_HEAD(&sbi->s_committed_transaction);
2591 +       spin_lock_init(&sbi->s_bal_lock);
2592 +
2593 +       sbi->s_mb_max_to_scan = MB_DEFAULT_MAX_TO_SCAN;
2594 +       sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN;
2595 +       sbi->s_mb_max_groups_to_scan = MB_DEFAULT_MAX_GROUPS_TO_SCAN;
2596 +       sbi->s_mb_stats = MB_DEFAULT_STATS;
2597 +       sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD;
2598 +       sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
2599 +       sbi->s_mb_history_filter = EXT3_MB_HISTORY_DEFAULT;
2600 +
2601 +       i = sizeof(struct ext3_locality_group) * NR_CPUS;
2602 +       sbi->s_locality_groups = kmalloc(i, GFP_NOFS);
2603 +       if (sbi->s_locality_groups == NULL) {
2604 +               clear_opt(sbi->s_mount_opt, MBALLOC);
2605 +               kfree(sbi->s_mb_offsets);
2606 +               kfree(sbi->s_mb_maxs);
2607 +               return -ENOMEM;
2608 +       }
2609 +       for (i = 0; i < NR_CPUS; i++) {
2610 +               struct ext3_locality_group *lg;
2611 +               lg = &sbi->s_locality_groups[i];
2612 +               sema_init(&lg->lg_sem, 1);
2613 +               INIT_LIST_HEAD(&lg->lg_prealloc_list);
2614 +               spin_lock_init(&lg->lg_prealloc_lock);
2615 +       }
2616 +
2617 +       ext3_mb_init_per_dev_proc(sb);
2618 +       ext3_mb_history_init(sb);
2619 +
2620 +       printk("EXT3-fs: mballoc enabled\n");
2621 +       return 0;
2622 +}
2623 +
2624 +void ext3_mb_cleanup_pa(struct ext3_group_info *grp)
2625 +{
2626 +       struct ext3_prealloc_space *pa;
2627 +       struct list_head *cur, *tmp;
2628 +       int count = 0;
2629 +
2630 +       list_for_each_safe(cur, tmp, &grp->bb_prealloc_list) {
2631 +               pa = list_entry(cur, struct ext3_prealloc_space, pa_group_list);
2632 +               list_del_rcu(&pa->pa_group_list);
2633 +               count++;
2634 +               kfree(pa);
2635 +       }
2636 +       if (count)
2637 +               mb_debug("mballoc: %u PAs left\n", count);
2638 +
2639 +}
2640 +
2641 +int ext3_mb_release(struct super_block *sb)
2642 +{
2643 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
2644 +       int i, num_meta_group_infos;
2645 +
2646 +       if (!test_opt(sb, MBALLOC))
2647 +               return 0;
2648 +
2649 +       /* release freed, non-committed blocks */
2650 +       spin_lock(&sbi->s_md_lock);
2651 +       list_splice_init(&sbi->s_closed_transaction,
2652 +                       &sbi->s_committed_transaction);
2653 +       list_splice_init(&sbi->s_active_transaction,
2654 +                       &sbi->s_committed_transaction);
2655 +       spin_unlock(&sbi->s_md_lock);
2656 +       ext3_mb_free_committed_blocks(sb);
2657 +
2658 +       if (sbi->s_group_info) {
2659 +               for (i = 0; i < sbi->s_groups_count; i++) {
2660 +#ifdef DOUBLE_CHECK
2661 +                       if (EXT3_GROUP_INFO(sb, i)->bb_bitmap)
2662 +                               kfree(EXT3_GROUP_INFO(sb, i)->bb_bitmap);
2663 +#endif
2664 +                       ext3_mb_cleanup_pa(EXT3_GROUP_INFO(sb, i));
2665 +                       kfree(EXT3_GROUP_INFO(sb, i));
2666 +               }
2667 +               num_meta_group_infos = (sbi->s_groups_count +
2668 +                               EXT3_DESC_PER_BLOCK(sb) - 1) >>
2669 +                       EXT3_DESC_PER_BLOCK_BITS(sb);
2670 +               for (i = 0; i < num_meta_group_infos; i++)
2671 +                       kfree(sbi->s_group_info[i]);
2672 +               kfree(sbi->s_group_info);
2673 +       }
2674 +       if (sbi->s_mb_offsets)
2675 +               kfree(sbi->s_mb_offsets);
2676 +       if (sbi->s_mb_maxs)
2677 +               kfree(sbi->s_mb_maxs);
2678 +       if (sbi->s_buddy_cache)
2679 +               iput(sbi->s_buddy_cache);
2680 +       if (sbi->s_mb_stats) {
2681 +               printk("EXT3-fs: mballoc: %u blocks %u reqs (%u success)\n",
2682 +                               atomic_read(&sbi->s_bal_allocated),
2683 +                               atomic_read(&sbi->s_bal_reqs),
2684 +                               atomic_read(&sbi->s_bal_success));
2685 +               printk("EXT3-fs: mballoc: %u extents scanned, %u goal hits, "
2686 +                               "%u 2^N hits, %u breaks, %u lost\n",
2687 +                               atomic_read(&sbi->s_bal_ex_scanned),
2688 +                               atomic_read(&sbi->s_bal_goals),
2689 +                               atomic_read(&sbi->s_bal_2orders),
2690 +                               atomic_read(&sbi->s_bal_breaks),
2691 +                               atomic_read(&sbi->s_mb_lost_chunks));
2692 +               printk("EXT3-fs: mballoc: %lu generated and it took %Lu\n",
2693 +                               sbi->s_mb_buddies_generated++,
2694 +                               sbi->s_mb_generation_time);
2695 +               printk("EXT3-fs: mballoc: %u preallocated, %u discarded\n",
2696 +                               atomic_read(&sbi->s_mb_preallocated),
2697 +                               atomic_read(&sbi->s_mb_discarded));
2698 +       }
2699 +
2700 +       if (sbi->s_locality_groups)
2701 +               kfree(sbi->s_locality_groups);
2702 +
2703 +       ext3_mb_history_release(sb);
2704 +       ext3_mb_destroy_per_dev_proc(sb);
2705 +
2706 +       return 0;
2707 +}
2708 +
2709 +void ext3_mb_free_committed_blocks(struct super_block *sb)
2710 +{
2711 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
2712 +       int err, i, count = 0, count2 = 0;
2713 +       struct ext3_free_metadata *md;
2714 +       struct ext3_buddy e3b;
2715 +
2716 +       if (list_empty(&sbi->s_committed_transaction))
2717 +               return;
2718 +
2719 +       /* there is committed blocks to be freed yet */
2720 +       do {
2721 +               /* get next array of blocks */
2722 +               md = NULL;
2723 +               spin_lock(&sbi->s_md_lock);
2724 +               if (!list_empty(&sbi->s_committed_transaction)) {
2725 +                       md = list_entry(sbi->s_committed_transaction.next,
2726 +                                       struct ext3_free_metadata, list);
2727 +                       list_del(&md->list);
2728 +               }
2729 +               spin_unlock(&sbi->s_md_lock);
2730 +
2731 +               if (md == NULL)
2732 +                       break;
2733 +
2734 +               mb_debug("gonna free %u blocks in group %u (0x%p):",
2735 +                               md->num, md->group, md);
2736 +
2737 +               err = ext3_mb_load_buddy(sb, md->group, &e3b);
2738 +               /* we expect to find existing buddy because it's pinned */
2739 +               BUG_ON(err != 0);
2740 +
2741 +               /* there are blocks to put in buddy to make them really free */
2742 +               count += md->num;
2743 +               count2++;
2744 +               ext3_lock_group(sb, md->group);
2745 +               for (i = 0; i < md->num; i++) {
2746 +                       mb_debug(" %u", md->blocks[i]);
2747 +                       err = mb_free_blocks(NULL, &e3b, md->blocks[i], 1);
2748 +                       BUG_ON(err != 0);
2749 +               }
2750 +               mb_debug("\n");
2751 +               ext3_unlock_group(sb, md->group);
2752 +
2753 +               /* balance refcounts from ext3_mb_free_metadata() */
2754 +               page_cache_release(e3b.bd_buddy_page);
2755 +               page_cache_release(e3b.bd_bitmap_page);
2756 +
2757 +               kfree(md);
2758 +               ext3_mb_release_desc(&e3b);
2759 +
2760 +       } while (md);
2761 +
2762 +       mb_debug("freed %u blocks in %u structures\n", count, count2);
2763 +}
2764 +
2765 +#define EXT3_ROOT                      "ext3"
2766 +#define EXT3_MB_STATS_NAME             "stats"
2767 +#define EXT3_MB_MAX_TO_SCAN_NAME       "max_to_scan"
2768 +#define EXT3_MB_MIN_TO_SCAN_NAME       "min_to_scan"
2769 +#define EXT3_MB_ORDER2_REQ             "order2_req"
2770 +#define EXT3_MB_STREAM_REQ             "stream_req"
2771 +
2772 +static int ext3_mb_stats_read(char *page, char **start, off_t off,
2773 +               int count, int *eof, void *data)
2774 +{
2775 +       struct ext3_sb_info *sbi = data;
2776 +       int len;
2777 +
2778 +       *eof = 1;
2779 +       if (off != 0)
2780 +               return 0;
2781 +
2782 +       len = sprintf(page, "%ld\n", sbi->s_mb_stats);
2783 +       *start = page;
2784 +       return len;
2785 +}
2786 +
2787 +static int ext3_mb_stats_write(struct file *file, const char *buffer,
2788 +               unsigned long count, void *data)
2789 +{
2790 +       struct ext3_sb_info *sbi = data;
2791 +       char str[32];
2792 +
2793 +       if (count >= sizeof(str)) {
2794 +               printk(KERN_ERR "EXT3-fs: %s string too long, max %u bytes\n",
2795 +                               EXT3_MB_STATS_NAME, (int)sizeof(str));
2796 +               return -EOVERFLOW;
2797 +       }
2798 +
2799 +       if (copy_from_user(str, buffer, count))
2800 +               return -EFAULT;
2801 +
2802 +       /* Only set to 0 or 1 respectively; zero->0; non-zero->1 */
2803 +       sbi->s_mb_stats = (simple_strtol(str, NULL, 0) != 0);
2804 +       return count;
2805 +}
2806 +
2807 +static int ext3_mb_max_to_scan_read(char *page, char **start, off_t off,
2808 +               int count, int *eof, void *data)
2809 +{
2810 +       struct ext3_sb_info *sbi = data;
2811 +       int len;
2812 +
2813 +       *eof = 1;
2814 +       if (off != 0)
2815 +               return 0;
2816 +
2817 +       len = sprintf(page, "%ld\n", sbi->s_mb_max_to_scan);
2818 +       *start = page;
2819 +       return len;
2820 +}
2821 +
2822 +static int ext3_mb_max_to_scan_write(struct file *file, const char *buffer,
2823 +               unsigned long count, void *data)
2824 +{
2825 +       struct ext3_sb_info *sbi = data;
2826 +       char str[32];
2827 +       long value;
2828 +
2829 +       if (count >= sizeof(str)) {
2830 +               printk(KERN_ERR "EXT3-fs: %s string too long, max %u bytes\n",
2831 +                               EXT3_MB_MAX_TO_SCAN_NAME, (int)sizeof(str));
2832 +               return -EOVERFLOW;
2833 +       }
2834 +
2835 +       if (copy_from_user(str, buffer, count))
2836 +               return -EFAULT;
2837 +
2838 +       /* Only set to 0 or 1 respectively; zero->0; non-zero->1 */
2839 +       value = simple_strtol(str, NULL, 0);
2840 +       if (value <= 0)
2841 +               return -ERANGE;
2842 +
2843 +       sbi->s_mb_max_to_scan = value;
2844 +
2845 +       return count;
2846 +}
2847 +
2848 +static int ext3_mb_min_to_scan_read(char *page, char **start, off_t off,
2849 +               int count, int *eof, void *data)
2850 +{
2851 +       struct ext3_sb_info *sbi = data;
2852 +       int len;
2853 +
2854 +       *eof = 1;
2855 +       if (off != 0)
2856 +               return 0;
2857 +
2858 +       len = sprintf(page, "%ld\n", sbi->s_mb_min_to_scan);
2859 +       *start = page;
2860 +       return len;
2861 +}
2862 +
2863 +static int ext3_mb_order2_req_write(struct file *file, const char *buffer,
2864 +               unsigned long count, void *data)
2865 +{
2866 +       struct ext3_sb_info *sbi = data;
2867 +       char str[32];
2868 +       long value;
2869 +
2870 +       if (count >= sizeof(str)) {
2871 +               printk(KERN_ERR "EXT3-fs: %s string too long, max %u bytes\n",
2872 +                               EXT3_MB_MIN_TO_SCAN_NAME, (int)sizeof(str));
2873 +               return -EOVERFLOW;
2874 +       }
2875 +
2876 +       if (copy_from_user(str, buffer, count))
2877 +               return -EFAULT;
2878 +
2879 +       /* Only set to 0 or 1 respectively; zero->0; non-zero->1 */
2880 +       value = simple_strtol(str, NULL, 0);
2881 +       if (value <= 0)
2882 +               return -ERANGE;
2883 +
2884 +       sbi->s_mb_order2_reqs = value;
2885 +
2886 +       return count;
2887 +}
2888 +
2889 +static int ext3_mb_order2_req_read(char *page, char **start, off_t off,
2890 +               int count, int *eof, void *data)
2891 +{
2892 +       struct ext3_sb_info *sbi = data;
2893 +       int len;
2894 +
2895 +       *eof = 1;
2896 +       if (off != 0)
2897 +               return 0;
2898 +
2899 +       len = sprintf(page, "%ld\n", sbi->s_mb_order2_reqs);
2900 +       *start = page;
2901 +       return len;
2902 +}
2903 +
2904 +static int ext3_mb_min_to_scan_write(struct file *file, const char *buffer,
2905 +               unsigned long count, void *data)
2906 +{
2907 +       struct ext3_sb_info *sbi = data;
2908 +       char str[32];
2909 +       long value;
2910 +
2911 +       if (count >= sizeof(str)) {
2912 +               printk(KERN_ERR "EXT3-fs: %s string too long, max %u bytes\n",
2913 +                               EXT3_MB_MIN_TO_SCAN_NAME, (int)sizeof(str));
2914 +               return -EOVERFLOW;
2915 +       }
2916 +
2917 +       if (copy_from_user(str, buffer, count))
2918 +               return -EFAULT;
2919 +
2920 +       /* Only set to 0 or 1 respectively; zero->0; non-zero->1 */
2921 +       value = simple_strtol(str, NULL, 0);
2922 +       if (value <= 0)
2923 +               return -ERANGE;
2924 +
2925 +       sbi->s_mb_min_to_scan = value;
2926 +
2927 +       return count;
2928 +}
2929 +
2930 +static int ext3_mb_stream_req_read(char *page, char **start, off_t off,
2931 +               int count, int *eof, void *data)
2932 +{
2933 +       struct ext3_sb_info *sbi = data;
2934 +       int len;
2935 +
2936 +       *eof = 1;
2937 +       if (off != 0)
2938 +               return 0;
2939 +
2940 +       len = sprintf(page, "%ld\n", sbi->s_mb_stream_request);
2941 +       *start = page;
2942 +       return len;
2943 +}
2944 +
2945 +static int ext3_mb_stream_req_write(struct file *file, const char *buffer,
2946 +               unsigned long count, void *data)
2947 +{
2948 +       struct ext3_sb_info *sbi = data;
2949 +       char str[32];
2950 +       long value;
2951 +
2952 +       if (count >= sizeof(str)) {
2953 +               printk(KERN_ERR "EXT3-fs: %s string too long, max %u bytes\n",
2954 +                               EXT3_MB_STREAM_REQ, (int)sizeof(str));
2955 +               return -EOVERFLOW;
2956 +       }
2957 +
2958 +       if (copy_from_user(str, buffer, count))
2959 +               return -EFAULT;
2960 +
2961 +       /* Only set to 0 or 1 respectively; zero->0; non-zero->1 */
2962 +       value = simple_strtol(str, NULL, 0);
2963 +       if (value <= 0)
2964 +               return -ERANGE;
2965 +
2966 +       sbi->s_mb_stream_request = value;
2967 +
2968 +       return count;
2969 +}
2970 +
2971 +int ext3_mb_init_per_dev_proc(struct super_block *sb)
2972 +{
2973 +       mode_t mode = S_IFREG | S_IRUGO | S_IWUSR;
2974 +       struct ext3_sb_info *sbi = EXT3_SB(sb);
2975 +       struct proc_dir_entry *proc;
2976 +       char devname[64], *name;
2977 +
2978 +       snprintf(devname, sizeof(devname) - 1, "%s",
2979 +               bdevname(sb->s_bdev, devname));
2980 +       sbi->s_mb_proc = proc_mkdir(devname, proc_root_ext3);
2981 +
2982 +       name = EXT3_MB_STATS_NAME;
2983 +       proc = create_proc_entry(name, mode, sbi->s_mb_proc);
2984 +       if (proc == NULL)
2985 +               goto err_out;
2986 +       proc->data = sbi;
2987 +       proc->read_proc  = ext3_mb_stats_read;
2988 +       proc->write_proc = ext3_mb_stats_write;
2989 +
2990 +       name = EXT3_MB_MAX_TO_SCAN_NAME;
2991 +       proc = create_proc_entry(name, mode, sbi->s_mb_proc);
2992 +       if (proc == NULL)
2993 +               goto err_out;
2994 +       proc->data = sbi;
2995 +       proc->read_proc  = ext3_mb_max_to_scan_read;
2996 +       proc->write_proc = ext3_mb_max_to_scan_write;
2997 +
2998 +       name = EXT3_MB_MIN_TO_SCAN_NAME;
2999 +       proc = create_proc_entry(name, mode, sbi->s_mb_proc);
3000 +       if (proc == NULL)
3001 +               goto err_out;
3002 +       proc->data = sbi;
3003 +       proc->read_proc  = ext3_mb_min_to_scan_read;
3004 +       proc->write_proc = ext3_mb_min_to_scan_write;
3005 +
3006 +       name = EXT3_MB_ORDER2_REQ;
3007 +       proc = create_proc_entry(name, mode, sbi->s_mb_proc);
3008 +       if (proc == NULL)
3009 +               goto err_out;
3010 +       proc->data = sbi;
3011 +       proc->read_proc  = ext3_mb_order2_req_read;
3012 +       proc->write_proc = ext3_mb_order2_req_write;
3013 +
3014 +       name = EXT3_MB_STREAM_REQ;
3015 +       proc = create_proc_entry(name, mode, sbi->s_mb_proc);
3016 +       if (proc == NULL)
3017 +               goto err_out;
3018 +       proc->data = sbi;
3019 +       proc->read_proc  = ext3_mb_stream_req_read;
3020 +       proc->write_proc = ext3_mb_stream_req_write;
3021 +
3022 +       return 0;
3023 +
3024 +err_out:
3025 +       printk(KERN_ERR "EXT3-fs: Unable to create %s\n", name);
3026 +       remove_proc_entry(EXT3_MB_STREAM_REQ, sbi->s_mb_proc);
3027 +       remove_proc_entry(EXT3_MB_ORDER2_REQ, sbi->s_mb_proc);
3028 +       remove_proc_entry(EXT3_MB_MIN_TO_SCAN_NAME, sbi->s_mb_proc);
3029 +       remove_proc_entry(EXT3_MB_MAX_TO_SCAN_NAME, sbi->s_mb_proc);
3030 +       remove_proc_entry(EXT3_MB_STATS_NAME, sbi->s_mb_proc);
3031 +  &nbs