2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
4 /* Reiserfs block (de)allocator, bitmap-based. */
6 #include <linux/time.h>
7 #include <linux/reiserfs_fs.h>
8 #include <linux/errno.h>
9 #include <linux/buffer_head.h>
10 #include <linux/kernel.h>
11 #include <linux/pagemap.h>
12 #include <linux/vmalloc.h>
13 #include <linux/reiserfs_fs_sb.h>
14 #include <linux/reiserfs_fs_i.h>
15 #include <linux/quotaops.h>
17 #define PREALLOCATION_SIZE 9
19 /* different reiserfs block allocator options */
21 #define SB_ALLOC_OPTS(s) (REISERFS_SB(s)->s_alloc_options.bits)
23 #define _ALLOC_concentrating_formatted_nodes 0
24 #define _ALLOC_displacing_large_files 1
25 #define _ALLOC_displacing_new_packing_localities 2
26 #define _ALLOC_old_hashed_relocation 3
27 #define _ALLOC_new_hashed_relocation 4
28 #define _ALLOC_skip_busy 5
29 #define _ALLOC_displace_based_on_dirid 6
30 #define _ALLOC_hashed_formatted_nodes 7
31 #define _ALLOC_old_way 8
32 #define _ALLOC_hundredth_slices 9
33 #define _ALLOC_dirid_groups 10
34 #define _ALLOC_oid_groups 11
35 #define _ALLOC_packing_groups 12
37 #define concentrating_formatted_nodes(s) test_bit(_ALLOC_concentrating_formatted_nodes, &SB_ALLOC_OPTS(s))
38 #define displacing_large_files(s) test_bit(_ALLOC_displacing_large_files, &SB_ALLOC_OPTS(s))
39 #define displacing_new_packing_localities(s) test_bit(_ALLOC_displacing_new_packing_localities, &SB_ALLOC_OPTS(s))
41 #define SET_OPTION(optname) \
43 reiserfs_warning(s, "reiserfs: option \"%s\" is set", #optname); \
44 set_bit(_ALLOC_ ## optname , &SB_ALLOC_OPTS(s)); \
46 #define TEST_OPTION(optname, s) \
47 test_bit(_ALLOC_ ## optname , &SB_ALLOC_OPTS(s))
49 static inline void get_bit_address(struct super_block *s,
50 b_blocknr_t block, int *bmap_nr, int *offset)
52 /* It is in the bitmap block number equal to the block
53 * number divided by the number of bits in a block. */
54 *bmap_nr = block >> (s->s_blocksize_bits + 3);
55 /* Within that bitmap block it is located at bit offset *offset. */
56 *offset = block & ((s->s_blocksize << 3) - 1);
59 #ifdef CONFIG_REISERFS_CHECK
60 int is_reusable(struct super_block *s, b_blocknr_t block, int bit_value)
64 if (block == 0 || block >= SB_BLOCK_COUNT(s)) {
66 "vs-4010: is_reusable: block number is out of range %lu (%u)",
67 block, SB_BLOCK_COUNT(s));
71 get_bit_address(s, block, &bmap, &offset);
73 /* Old format filesystem? Unlikely, but the bitmaps are all up front so
74 * we need to account for it. */
75 if (unlikely(test_bit(REISERFS_OLD_FORMAT,
76 &(REISERFS_SB(s)->s_properties)))) {
77 b_blocknr_t bmap1 = REISERFS_SB(s)->s_sbh->b_blocknr + 1;
78 if (block >= bmap1 && block <= bmap1 + SB_BMAP_NR(s)) {
79 reiserfs_warning(s, "vs: 4019: is_reusable: "
80 "bitmap block %lu(%u) can't be freed or reused",
81 block, SB_BMAP_NR(s));
86 reiserfs_warning(s, "vs: 4020: is_reusable: "
87 "bitmap block %lu(%u) can't be freed or reused",
88 block, SB_BMAP_NR(s));
93 if (bmap >= SB_BMAP_NR(s)) {
95 "vs-4030: is_reusable: there is no so many bitmap blocks: "
96 "block=%lu, bitmap_nr=%d", block, bmap);
100 if (bit_value == 0 && block == SB_ROOT_BLOCK(s)) {
102 "vs-4050: is_reusable: this is root block (%u), "
103 "it must be busy", SB_ROOT_BLOCK(s));
109 #endif /* CONFIG_REISERFS_CHECK */
111 /* searches in journal structures for a given block number (bmap, off). If block
112 is found in reiserfs journal it suggests next free block candidate to test. */
113 static inline int is_block_in_journal(struct super_block *s, int bmap, int
118 if (reiserfs_in_journal(s, bmap, off, 1, &tmp)) {
119 if (tmp) { /* hint supplied */
121 PROC_INFO_INC(s, scan_bitmap.in_journal_hint);
123 (*next) = off + 1; /* inc offset to avoid looping. */
124 PROC_INFO_INC(s, scan_bitmap.in_journal_nohint);
126 PROC_INFO_INC(s, scan_bitmap.retry);
132 /* it searches for a window of zero bits with given minimum and maximum lengths in one bitmap
134 static int scan_bitmap_block(struct reiserfs_transaction_handle *th,
135 int bmap_n, int *beg, int boundary, int min,
138 struct super_block *s = th->t_super;
139 struct reiserfs_bitmap_info *bi = &SB_AP_BITMAP(s)[bmap_n];
140 struct buffer_head *bh;
144 BUG_ON(!th->t_trans_id);
146 RFALSE(bmap_n >= SB_BMAP_NR(s), "Bitmap %d is out of range (0..%d)",
147 bmap_n, SB_BMAP_NR(s) - 1);
148 PROC_INFO_INC(s, scan_bitmap.bmap);
149 /* this is unclear and lacks comments, explain how journal bitmaps
150 work here for the reader. Convey a sense of the design here. What
152 /* - I mean `a window of zero bits' as in description of this function - Zam. */
155 reiserfs_warning(s, "NULL bitmap info pointer for bitmap %d",
160 bh = reiserfs_read_bitmap_block(s, bmap_n);
166 if (bi->free_count < min) {
168 return 0; // No free blocks in this bitmap
171 /* search for a first zero bit -- beggining of a window */
172 *beg = reiserfs_find_next_zero_le_bit
173 ((unsigned long *)(bh->b_data), boundary, *beg);
175 if (*beg + min > boundary) { /* search for a zero bit fails or the rest of bitmap block
176 * cannot contain a zero window of minimum size */
181 if (unfm && is_block_in_journal(s, bmap_n, *beg, beg))
183 /* first zero bit found; we check next bits */
184 for (end = *beg + 1;; end++) {
185 if (end >= *beg + max || end >= boundary
186 || reiserfs_test_le_bit(end, bh->b_data)) {
190 /* finding the other end of zero bit window requires looking into journal structures (in
191 * case of searching for free blocks for unformatted nodes) */
192 if (unfm && is_block_in_journal(s, bmap_n, end, &next))
196 /* now (*beg) points to beginning of zero bits window,
197 * (end) points to one bit after the window end */
198 if (end - *beg >= min) { /* it seems we have found window of proper size */
200 reiserfs_prepare_for_journal(s, bh, 1);
201 /* try to set all blocks used checking are they still free */
202 for (i = *beg; i < end; i++) {
203 /* It seems that we should not check in journal again. */
204 if (reiserfs_test_and_set_le_bit
206 /* bit was set by another process
207 * while we slept in prepare_for_journal() */
208 PROC_INFO_INC(s, scan_bitmap.stolen);
209 if (i >= *beg + min) { /* we can continue with smaller set of allocated blocks,
210 * if length of this set is more or equal to `min' */
214 /* otherwise we clear all bit were set ... */
216 reiserfs_test_and_clear_le_bit
218 reiserfs_restore_prepared_buffer(s, bh);
220 /* ... and search again in current block from beginning */
224 bi->free_count -= (end - *beg);
225 journal_mark_dirty(th, s, bh);
228 /* free block count calculation */
229 reiserfs_prepare_for_journal(s, SB_BUFFER_WITH_SB(s),
231 PUT_SB_FREE_BLOCKS(s, SB_FREE_BLOCKS(s) - (end - *beg));
232 journal_mark_dirty(th, s, SB_BUFFER_WITH_SB(s));
241 static int bmap_hash_id(struct super_block *s, u32 id)
243 char *hash_in = NULL;
250 hash_in = (char *)(&id);
251 hash = keyed_hash(hash_in, 4);
252 bm = hash % SB_BMAP_NR(s);
256 /* this can only be true when SB_BMAP_NR = 1 */
257 if (bm >= SB_BMAP_NR(s))
263 * hashes the id and then returns > 0 if the block group for the
264 * corresponding hash is full
266 static inline int block_group_used(struct super_block *s, u32 id)
268 int bm = bmap_hash_id(s, id);
269 struct reiserfs_bitmap_info *info = &SB_AP_BITMAP(s)[bm];
271 /* If we don't have cached information on this bitmap block, we're
272 * going to have to load it later anyway. Loading it here allows us
273 * to make a better decision. This favors long-term performace gain
274 * with a better on-disk layout vs. a short term gain of skipping the
275 * read and potentially having a bad placement. */
276 if (info->first_zero_hint == 0) {
277 struct buffer_head *bh = reiserfs_read_bitmap_block(s, bm);
281 if (info->free_count > ((s->s_blocksize << 3) * 60 / 100)) {
288 * the packing is returned in disk byte order
290 __le32 reiserfs_choose_packing(struct inode * dir)
293 if (TEST_OPTION(packing_groups, dir->i_sb)) {
294 u32 parent_dir = le32_to_cpu(INODE_PKEY(dir)->k_dir_id);
296 * some versions of reiserfsck expect packing locality 1 to be
299 if (parent_dir == 1 || block_group_used(dir->i_sb, parent_dir))
300 packing = INODE_PKEY(dir)->k_objectid;
302 packing = INODE_PKEY(dir)->k_dir_id;
304 packing = INODE_PKEY(dir)->k_objectid;
308 /* Tries to find contiguous zero bit window (given size) in given region of
309 * bitmap and place new blocks there. Returns number of allocated blocks. */
310 static int scan_bitmap(struct reiserfs_transaction_handle *th,
311 b_blocknr_t * start, b_blocknr_t finish,
312 int min, int max, int unfm, unsigned long file_block)
314 int nr_allocated = 0;
315 struct super_block *s = th->t_super;
316 /* find every bm and bmap and bmap_nr in this file, and change them all to bitmap_blocknr
317 * - Hans, it is not a block number - Zam. */
321 int off_max = s->s_blocksize << 3;
323 BUG_ON(!th->t_trans_id);
325 PROC_INFO_INC(s, scan_bitmap.call);
326 if (SB_FREE_BLOCKS(s) <= 0)
327 return 0; // No point in looking for more free blocks
329 get_bit_address(s, *start, &bm, &off);
330 get_bit_address(s, finish, &end_bm, &end_off);
331 if (bm > SB_BMAP_NR(s))
333 if (end_bm > SB_BMAP_NR(s))
334 end_bm = SB_BMAP_NR(s);
336 /* When the bitmap is more than 10% free, anyone can allocate.
337 * When it's less than 10% free, only files that already use the
338 * bitmap are allowed. Once we pass 80% full, this restriction
341 * We do this so that files that grow later still have space close to
342 * their original allocation. This improves locality, and presumably
343 * performance as a result.
345 * This is only an allocation policy and does not make up for getting a
346 * bad hint. Decent hinting must be implemented for this to work well.
348 if (TEST_OPTION(skip_busy, s)
349 && SB_FREE_BLOCKS(s) > SB_BLOCK_COUNT(s) / 20) {
350 for (; bm < end_bm; bm++, off = 0) {
351 if ((off && (!unfm || (file_block != 0)))
352 || SB_AP_BITMAP(s)[bm].free_count >
353 (s->s_blocksize << 3) / 10)
355 scan_bitmap_block(th, bm, &off, off_max,
360 /* we know from above that start is a reasonable number */
361 get_bit_address(s, *start, &bm, &off);
364 for (; bm < end_bm; bm++, off = 0) {
366 scan_bitmap_block(th, bm, &off, off_max, min, max, unfm);
372 scan_bitmap_block(th, bm, &off, end_off + 1, min, max, unfm);
375 *start = bm * off_max + off;
380 static void _reiserfs_free_block(struct reiserfs_transaction_handle *th,
381 struct inode *inode, b_blocknr_t block,
384 struct super_block *s = th->t_super;
385 struct reiserfs_super_block *rs;
386 struct buffer_head *sbh, *bmbh;
387 struct reiserfs_bitmap_info *apbi;
390 BUG_ON(!th->t_trans_id);
392 PROC_INFO_INC(s, free_block);
394 rs = SB_DISK_SUPER_BLOCK(s);
395 sbh = SB_BUFFER_WITH_SB(s);
396 apbi = SB_AP_BITMAP(s);
398 get_bit_address(s, block, &nr, &offset);
400 if (nr >= sb_bmap_nr(rs)) {
401 reiserfs_warning(s, "vs-4075: reiserfs_free_block: "
402 "block %lu is out of range on %s",
403 block, reiserfs_bdevname(s));
407 bmbh = reiserfs_read_bitmap_block(s, nr);
411 reiserfs_prepare_for_journal(s, bmbh, 1);
413 /* clear bit for the given block in bit map */
414 if (!reiserfs_test_and_clear_le_bit(offset, bmbh->b_data)) {
415 reiserfs_warning(s, "vs-4080: reiserfs_free_block: "
416 "free_block (%s:%lu)[dev:blocknr]: bit already cleared",
417 reiserfs_bdevname(s), block);
419 apbi[nr].free_count++;
420 journal_mark_dirty(th, s, bmbh);
423 reiserfs_prepare_for_journal(s, sbh, 1);
424 /* update super block */
425 set_sb_free_blocks(rs, sb_free_blocks(rs) + 1);
427 journal_mark_dirty(th, s, sbh);
429 DQUOT_FREE_BLOCK_NODIRTY(inode, 1);
432 void reiserfs_free_block(struct reiserfs_transaction_handle *th,
433 struct inode *inode, b_blocknr_t block,
436 struct super_block *s = th->t_super;
438 BUG_ON(!th->t_trans_id);
440 RFALSE(!s, "vs-4061: trying to free block on nonexistent device");
441 RFALSE(is_reusable(s, block, 1) == 0,
442 "vs-4071: can not free such block");
443 /* mark it before we clear it, just in case */
444 journal_mark_freed(th, s, block);
445 _reiserfs_free_block(th, inode, block, for_unformatted);
448 /* preallocated blocks don't need to be run through journal_mark_freed */
449 static void reiserfs_free_prealloc_block(struct reiserfs_transaction_handle *th,
450 struct inode *inode, b_blocknr_t block)
453 "vs-4060: trying to free block on nonexistent device");
454 RFALSE(is_reusable(th->t_super, block, 1) == 0,
455 "vs-4070: can not free such block");
456 BUG_ON(!th->t_trans_id);
457 _reiserfs_free_block(th, inode, block, 1);
460 static void __discard_prealloc(struct reiserfs_transaction_handle *th,
461 struct reiserfs_inode_info *ei)
463 unsigned long save = ei->i_prealloc_block;
465 struct inode *inode = &ei->vfs_inode;
466 BUG_ON(!th->t_trans_id);
467 #ifdef CONFIG_REISERFS_CHECK
468 if (ei->i_prealloc_count < 0)
469 reiserfs_warning(th->t_super,
470 "zam-4001:%s: inode has negative prealloc blocks count.",
473 while (ei->i_prealloc_count > 0) {
474 reiserfs_free_prealloc_block(th, inode, ei->i_prealloc_block);
475 ei->i_prealloc_block++;
476 ei->i_prealloc_count--;
480 reiserfs_update_sd(th, inode);
481 ei->i_prealloc_block = save;
482 list_del_init(&(ei->i_prealloc_list));
485 /* FIXME: It should be inline function */
486 void reiserfs_discard_prealloc(struct reiserfs_transaction_handle *th,
489 struct reiserfs_inode_info *ei = REISERFS_I(inode);
490 BUG_ON(!th->t_trans_id);
491 if (ei->i_prealloc_count)
492 __discard_prealloc(th, ei);
495 void reiserfs_discard_all_prealloc(struct reiserfs_transaction_handle *th)
497 struct list_head *plist = &SB_JOURNAL(th->t_super)->j_prealloc_list;
499 BUG_ON(!th->t_trans_id);
501 while (!list_empty(plist)) {
502 struct reiserfs_inode_info *ei;
503 ei = list_entry(plist->next, struct reiserfs_inode_info,
505 #ifdef CONFIG_REISERFS_CHECK
506 if (!ei->i_prealloc_count) {
507 reiserfs_warning(th->t_super,
508 "zam-4001:%s: inode is in prealloc list but has no preallocated blocks.",
512 __discard_prealloc(th, ei);
516 void reiserfs_init_alloc_options(struct super_block *s)
518 set_bit(_ALLOC_skip_busy, &SB_ALLOC_OPTS(s));
519 set_bit(_ALLOC_dirid_groups, &SB_ALLOC_OPTS(s));
520 set_bit(_ALLOC_packing_groups, &SB_ALLOC_OPTS(s));
523 /* block allocator related options are parsed here */
524 int reiserfs_parse_alloc_options(struct super_block *s, char *options)
526 char *this_char, *value;
528 REISERFS_SB(s)->s_alloc_options.bits = 0; /* clear default settings */
530 while ((this_char = strsep(&options, ":")) != NULL) {
531 if ((value = strchr(this_char, '=')) != NULL)
534 if (!strcmp(this_char, "concentrating_formatted_nodes")) {
536 SET_OPTION(concentrating_formatted_nodes);
538 && *value) ? simple_strtoul(value, &value,
540 if (temp <= 0 || temp > 100) {
541 REISERFS_SB(s)->s_alloc_options.border = 10;
543 REISERFS_SB(s)->s_alloc_options.border =
548 if (!strcmp(this_char, "displacing_large_files")) {
549 SET_OPTION(displacing_large_files);
550 REISERFS_SB(s)->s_alloc_options.large_file_size =
552 && *value) ? simple_strtoul(value, &value, 0) : 16;
555 if (!strcmp(this_char, "displacing_new_packing_localities")) {
556 SET_OPTION(displacing_new_packing_localities);
560 if (!strcmp(this_char, "old_hashed_relocation")) {
561 SET_OPTION(old_hashed_relocation);
565 if (!strcmp(this_char, "new_hashed_relocation")) {
566 SET_OPTION(new_hashed_relocation);
570 if (!strcmp(this_char, "dirid_groups")) {
571 SET_OPTION(dirid_groups);
574 if (!strcmp(this_char, "oid_groups")) {
575 SET_OPTION(oid_groups);
578 if (!strcmp(this_char, "packing_groups")) {
579 SET_OPTION(packing_groups);
582 if (!strcmp(this_char, "hashed_formatted_nodes")) {
583 SET_OPTION(hashed_formatted_nodes);
587 if (!strcmp(this_char, "skip_busy")) {
588 SET_OPTION(skip_busy);
592 if (!strcmp(this_char, "hundredth_slices")) {
593 SET_OPTION(hundredth_slices);
597 if (!strcmp(this_char, "old_way")) {
602 if (!strcmp(this_char, "displace_based_on_dirid")) {
603 SET_OPTION(displace_based_on_dirid);
607 if (!strcmp(this_char, "preallocmin")) {
608 REISERFS_SB(s)->s_alloc_options.preallocmin =
610 && *value) ? simple_strtoul(value, &value, 0) : 4;
614 if (!strcmp(this_char, "preallocsize")) {
615 REISERFS_SB(s)->s_alloc_options.preallocsize =
617 && *value) ? simple_strtoul(value, &value,
623 reiserfs_warning(s, "zam-4001: %s : unknown option - %s",
624 __FUNCTION__, this_char);
628 reiserfs_warning(s, "allocator options = [%08x]\n", SB_ALLOC_OPTS(s));
632 static inline void new_hashed_relocation(reiserfs_blocknr_hint_t * hint)
635 if (hint->formatted_node) {
636 hash_in = (char *)&hint->key.k_dir_id;
639 //hint->search_start = hint->beg;
640 hash_in = (char *)&hint->key.k_dir_id;
642 if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
643 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_dir_id);
646 (char *)(&INODE_PKEY(hint->inode)->k_objectid);
650 hint->beg + keyed_hash(hash_in, 4) % (hint->end - hint->beg);
654 * Relocation based on dirid, hashing them into a given bitmap block
655 * files. Formatted nodes are unaffected, a seperate policy covers them
657 static void dirid_groups(reiserfs_blocknr_hint_t * hint)
662 struct super_block *sb = hint->th->t_super;
664 dirid = le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id);
665 else if (hint->formatted_node)
666 dirid = hint->key.k_dir_id;
669 bm = bmap_hash_id(sb, dirid);
670 hash = bm * (sb->s_blocksize << 3);
671 /* give a portion of the block group to metadata */
673 hash += sb->s_blocksize / 2;
674 hint->search_start = hash;
679 * Relocation based on oid, hashing them into a given bitmap block
680 * files. Formatted nodes are unaffected, a seperate policy covers them
682 static void oid_groups(reiserfs_blocknr_hint_t * hint)
690 dirid = le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id);
692 /* keep the root dir and it's first set of subdirs close to
693 * the start of the disk
696 hash = (hint->inode->i_sb->s_blocksize << 3);
698 oid = le32_to_cpu(INODE_PKEY(hint->inode)->k_objectid);
699 bm = bmap_hash_id(hint->inode->i_sb, oid);
700 hash = bm * (hint->inode->i_sb->s_blocksize << 3);
702 hint->search_start = hash;
706 /* returns 1 if it finds an indirect item and gets valid hint info
707 * from it, otherwise 0
709 static int get_left_neighbor(reiserfs_blocknr_hint_t * hint)
711 struct treepath *path;
712 struct buffer_head *bh;
713 struct item_head *ih;
718 if (!hint->path) /* reiserfs code can call this function w/o pointer to path
719 * structure supplied; then we rely on supplied search_start */
723 bh = get_last_bh(path);
724 RFALSE(!bh, "green-4002: Illegal path specified to get_left_neighbor");
726 pos_in_item = path->pos_in_item;
727 item = get_item(path);
729 hint->search_start = bh->b_blocknr;
731 if (!hint->formatted_node && is_indirect_le_ih(ih)) {
732 /* for indirect item: go to left and look for the first non-hole entry
733 in the indirect item */
734 if (pos_in_item == I_UNFM_NUM(ih))
736 // pos_in_item = I_UNFM_NUM (ih) - 1;
737 while (pos_in_item >= 0) {
738 int t = get_block_num(item, pos_in_item);
740 hint->search_start = t;
748 /* does result value fit into specified region? */
752 /* should be, if formatted node, then try to put on first part of the device
753 specified as number of percent with mount option device, else try to put
754 on last of device. This is not to say it is good code to do so,
755 but the effect should be measured. */
756 static inline void set_border_in_hint(struct super_block *s,
757 reiserfs_blocknr_hint_t * hint)
760 SB_BLOCK_COUNT(s) / REISERFS_SB(s)->s_alloc_options.border;
762 if (hint->formatted_node)
763 hint->end = border - 1;
768 static inline void displace_large_file(reiserfs_blocknr_hint_t * hint)
770 if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
773 keyed_hash((char *)(&INODE_PKEY(hint->inode)->k_dir_id),
774 4) % (hint->end - hint->beg);
778 keyed_hash((char *)(&INODE_PKEY(hint->inode)->k_objectid),
779 4) % (hint->end - hint->beg);
782 static inline void hash_formatted_node(reiserfs_blocknr_hint_t * hint)
787 hash_in = (char *)&hint->key.k_dir_id;
788 else if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
789 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_dir_id);
791 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_objectid);
794 hint->beg + keyed_hash(hash_in, 4) % (hint->end - hint->beg);
798 this_blocknr_allocation_would_make_it_a_large_file(reiserfs_blocknr_hint_t *
801 return hint->block ==
802 REISERFS_SB(hint->th->t_super)->s_alloc_options.large_file_size;
805 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
806 static inline void displace_new_packing_locality(reiserfs_blocknr_hint_t * hint)
808 struct in_core_key *key = &hint->key;
810 hint->th->displace_new_blocks = 0;
812 hint->beg + keyed_hash((char *)(&key->k_objectid),
813 4) % (hint->end - hint->beg);
817 static inline int old_hashed_relocation(reiserfs_blocknr_hint_t * hint)
822 if (hint->formatted_node || hint->inode == NULL) {
826 hash_in = le32_to_cpu((INODE_PKEY(hint->inode))->k_dir_id);
828 hint->beg + (u32) keyed_hash(((char *)(&hash_in)),
829 4) % (hint->end - hint->beg - 1);
830 if (border > hint->search_start)
831 hint->search_start = border;
836 static inline int old_way(reiserfs_blocknr_hint_t * hint)
840 if (hint->formatted_node || hint->inode == NULL) {
846 le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id) % (hint->end -
848 if (border > hint->search_start)
849 hint->search_start = border;
854 static inline void hundredth_slices(reiserfs_blocknr_hint_t * hint)
856 struct in_core_key *key = &hint->key;
857 b_blocknr_t slice_start;
860 (keyed_hash((char *)(&key->k_dir_id), 4) % 100) * (hint->end / 100);
861 if (slice_start > hint->search_start
862 || slice_start + (hint->end / 100) <= hint->search_start) {
863 hint->search_start = slice_start;
867 static void determine_search_start(reiserfs_blocknr_hint_t * hint,
870 struct super_block *s = hint->th->t_super;
874 hint->end = SB_BLOCK_COUNT(s) - 1;
876 /* This is former border algorithm. Now with tunable border offset */
877 if (concentrating_formatted_nodes(s))
878 set_border_in_hint(s, hint);
880 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
881 /* whenever we create a new directory, we displace it. At first we will
882 hash for location, later we might look for a moderately empty place for
884 if (displacing_new_packing_localities(s)
885 && hint->th->displace_new_blocks) {
886 displace_new_packing_locality(hint);
888 /* we do not continue determine_search_start,
889 * if new packing locality is being displaced */
894 /* all persons should feel encouraged to add more special cases here and
897 if (displacing_large_files(s) && !hint->formatted_node
898 && this_blocknr_allocation_would_make_it_a_large_file(hint)) {
899 displace_large_file(hint);
903 /* if none of our special cases is relevant, use the left neighbor in the
904 tree order of the new node we are allocating for */
905 if (hint->formatted_node && TEST_OPTION(hashed_formatted_nodes, s)) {
906 hash_formatted_node(hint);
910 unfm_hint = get_left_neighbor(hint);
912 /* Mimic old block allocator behaviour, that is if VFS allowed for preallocation,
913 new blocks are displaced based on directory ID. Also, if suggested search_start
914 is less than last preallocated block, we start searching from it, assuming that
915 HDD dataflow is faster in forward direction */
916 if (TEST_OPTION(old_way, s)) {
917 if (!hint->formatted_node) {
918 if (!reiserfs_hashed_relocation(s))
920 else if (!reiserfs_no_unhashed_relocation(s))
921 old_hashed_relocation(hint);
924 && hint->search_start <
925 REISERFS_I(hint->inode)->i_prealloc_block)
927 REISERFS_I(hint->inode)->i_prealloc_block;
932 /* This is an approach proposed by Hans */
933 if (TEST_OPTION(hundredth_slices, s)
934 && !(displacing_large_files(s) && !hint->formatted_node)) {
935 hundredth_slices(hint);
939 /* old_hashed_relocation only works on unformatted */
940 if (!unfm_hint && !hint->formatted_node &&
941 TEST_OPTION(old_hashed_relocation, s)) {
942 old_hashed_relocation(hint);
944 /* new_hashed_relocation works with both formatted/unformatted nodes */
945 if ((!unfm_hint || hint->formatted_node) &&
946 TEST_OPTION(new_hashed_relocation, s)) {
947 new_hashed_relocation(hint);
949 /* dirid grouping works only on unformatted nodes */
950 if (!unfm_hint && !hint->formatted_node && TEST_OPTION(dirid_groups, s)) {
953 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
954 if (hint->formatted_node && TEST_OPTION(dirid_groups, s)) {
959 /* oid grouping works only on unformatted nodes */
960 if (!unfm_hint && !hint->formatted_node && TEST_OPTION(oid_groups, s)) {
966 static int determine_prealloc_size(reiserfs_blocknr_hint_t * hint)
968 /* make minimum size a mount option and benchmark both ways */
969 /* we preallocate blocks only for regular files, specific size */
970 /* benchmark preallocating always and see what happens */
972 hint->prealloc_size = 0;
974 if (!hint->formatted_node && hint->preallocate) {
975 if (S_ISREG(hint->inode->i_mode)
976 && hint->inode->i_size >=
977 REISERFS_SB(hint->th->t_super)->s_alloc_options.
978 preallocmin * hint->inode->i_sb->s_blocksize)
979 hint->prealloc_size =
980 REISERFS_SB(hint->th->t_super)->s_alloc_options.
986 /* XXX I know it could be merged with upper-level function;
987 but may be result function would be too complex. */
988 static inline int allocate_without_wrapping_disk(reiserfs_blocknr_hint_t * hint,
989 b_blocknr_t * new_blocknrs,
991 b_blocknr_t finish, int min,
995 int rest = amount_needed;
998 while (rest > 0 && start <= finish) {
999 nr_allocated = scan_bitmap(hint->th, &start, finish, min,
1000 rest + prealloc_size,
1001 !hint->formatted_node, hint->block);
1003 if (nr_allocated == 0) /* no new blocks allocated, return */
1006 /* fill free_blocknrs array first */
1007 while (rest > 0 && nr_allocated > 0) {
1008 *new_blocknrs++ = start++;
1013 /* do we have something to fill prealloc. array also ? */
1014 if (nr_allocated > 0) {
1015 /* it means prealloc_size was greater that 0 and we do preallocation */
1016 list_add(&REISERFS_I(hint->inode)->i_prealloc_list,
1017 &SB_JOURNAL(hint->th->t_super)->
1019 REISERFS_I(hint->inode)->i_prealloc_block = start;
1020 REISERFS_I(hint->inode)->i_prealloc_count =
1026 return (amount_needed - rest);
1029 static inline int blocknrs_and_prealloc_arrays_from_search_start
1030 (reiserfs_blocknr_hint_t * hint, b_blocknr_t * new_blocknrs,
1031 int amount_needed) {
1032 struct super_block *s = hint->th->t_super;
1033 b_blocknr_t start = hint->search_start;
1034 b_blocknr_t finish = SB_BLOCK_COUNT(s) - 1;
1036 int nr_allocated = 0;
1038 determine_prealloc_size(hint);
1039 if (!hint->formatted_node) {
1041 #ifdef REISERQUOTA_DEBUG
1042 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1043 "reiserquota: allocating %d blocks id=%u",
1044 amount_needed, hint->inode->i_uid);
1047 DQUOT_ALLOC_BLOCK_NODIRTY(hint->inode, amount_needed);
1048 if (quota_ret) /* Quota exceeded? */
1049 return QUOTA_EXCEEDED;
1050 if (hint->preallocate && hint->prealloc_size) {
1051 #ifdef REISERQUOTA_DEBUG
1052 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1053 "reiserquota: allocating (prealloc) %d blocks id=%u",
1054 hint->prealloc_size, hint->inode->i_uid);
1057 DQUOT_PREALLOC_BLOCK_NODIRTY(hint->inode,
1058 hint->prealloc_size);
1060 hint->preallocate = hint->prealloc_size = 0;
1062 /* for unformatted nodes, force large allocations */
1067 case 0: /* Search from hint->search_start to end of disk */
1068 start = hint->search_start;
1069 finish = SB_BLOCK_COUNT(s) - 1;
1071 case 1: /* Search from hint->beg to hint->search_start */
1073 finish = hint->search_start;
1075 case 2: /* Last chance: Search from 0 to hint->beg */
1079 default: /* We've tried searching everywhere, not enough space */
1080 /* Free the blocks */
1081 if (!hint->formatted_node) {
1082 #ifdef REISERQUOTA_DEBUG
1083 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1084 "reiserquota: freeing (nospace) %d blocks id=%u",
1086 hint->prealloc_size -
1088 hint->inode->i_uid);
1090 DQUOT_FREE_BLOCK_NODIRTY(hint->inode, amount_needed + hint->prealloc_size - nr_allocated); /* Free not allocated blocks */
1092 while (nr_allocated--)
1093 reiserfs_free_block(hint->th, hint->inode,
1094 new_blocknrs[nr_allocated],
1095 !hint->formatted_node);
1097 return NO_DISK_SPACE;
1099 } while ((nr_allocated += allocate_without_wrapping_disk(hint,
1109 if (!hint->formatted_node &&
1110 amount_needed + hint->prealloc_size >
1111 nr_allocated + REISERFS_I(hint->inode)->i_prealloc_count) {
1112 /* Some of preallocation blocks were not allocated */
1113 #ifdef REISERQUOTA_DEBUG
1114 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1115 "reiserquota: freeing (failed prealloc) %d blocks id=%u",
1116 amount_needed + hint->prealloc_size -
1118 REISERFS_I(hint->inode)->i_prealloc_count,
1119 hint->inode->i_uid);
1121 DQUOT_FREE_BLOCK_NODIRTY(hint->inode, amount_needed +
1122 hint->prealloc_size - nr_allocated -
1123 REISERFS_I(hint->inode)->
1130 /* grab new blocknrs from preallocated list */
1131 /* return amount still needed after using them */
1132 static int use_preallocated_list_if_available(reiserfs_blocknr_hint_t * hint,
1133 b_blocknr_t * new_blocknrs,
1136 struct inode *inode = hint->inode;
1138 if (REISERFS_I(inode)->i_prealloc_count > 0) {
1139 while (amount_needed) {
1141 *new_blocknrs++ = REISERFS_I(inode)->i_prealloc_block++;
1142 REISERFS_I(inode)->i_prealloc_count--;
1146 if (REISERFS_I(inode)->i_prealloc_count <= 0) {
1147 list_del(&REISERFS_I(inode)->i_prealloc_list);
1152 /* return amount still needed after using preallocated blocks */
1153 return amount_needed;
1156 int reiserfs_allocate_blocknrs(reiserfs_blocknr_hint_t * hint, b_blocknr_t * new_blocknrs, int amount_needed, int reserved_by_us /* Amount of blocks we have
1157 already reserved */ )
1159 int initial_amount_needed = amount_needed;
1161 struct super_block *s = hint->th->t_super;
1163 /* Check if there is enough space, taking into account reserved space */
1164 if (SB_FREE_BLOCKS(s) - REISERFS_SB(s)->reserved_blocks <
1165 amount_needed - reserved_by_us)
1166 return NO_DISK_SPACE;
1167 /* should this be if !hint->inode && hint->preallocate? */
1168 /* do you mean hint->formatted_node can be removed ? - Zam */
1169 /* hint->formatted_node cannot be removed because we try to access
1170 inode information here, and there is often no inode assotiated with
1171 metadata allocations - green */
1173 if (!hint->formatted_node && hint->preallocate) {
1174 amount_needed = use_preallocated_list_if_available
1175 (hint, new_blocknrs, amount_needed);
1176 if (amount_needed == 0) /* all blocknrs we need we got from
1179 new_blocknrs += (initial_amount_needed - amount_needed);
1182 /* find search start and save it in hint structure */
1183 determine_search_start(hint, amount_needed);
1184 if (hint->search_start >= SB_BLOCK_COUNT(s))
1185 hint->search_start = SB_BLOCK_COUNT(s) - 1;
1187 /* allocation itself; fill new_blocknrs and preallocation arrays */
1188 ret = blocknrs_and_prealloc_arrays_from_search_start
1189 (hint, new_blocknrs, amount_needed);
1191 /* we used prealloc. list to fill (partially) new_blocknrs array. If final allocation fails we
1192 * need to return blocks back to prealloc. list or just free them. -- Zam (I chose second
1195 if (ret != CARRY_ON) {
1196 while (amount_needed++ < initial_amount_needed) {
1197 reiserfs_free_block(hint->th, hint->inode,
1198 *(--new_blocknrs), 1);
1204 void reiserfs_cache_bitmap_metadata(struct super_block *sb,
1205 struct buffer_head *bh,
1206 struct reiserfs_bitmap_info *info)
1208 unsigned long *cur = (unsigned long *)(bh->b_data + bh->b_size);
1210 info->first_zero_hint = 1 << (sb->s_blocksize_bits + 3);
1212 while (--cur >= (unsigned long *)bh->b_data) {
1213 int base = ((char *)cur - bh->b_data) << 3;
1215 /* 0 and ~0 are special, we can optimize for them */
1217 info->first_zero_hint = base;
1218 info->free_count += BITS_PER_LONG;
1219 } else if (*cur != ~0L) { /* A mix, investigate */
1221 for (b = BITS_PER_LONG - 1; b >= 0; b--) {
1222 if (!reiserfs_test_le_bit(b, cur)) {
1223 info->first_zero_hint = base + b;
1229 /* The first bit must ALWAYS be 1 */
1230 BUG_ON(info->first_zero_hint == 0);
1233 struct buffer_head *reiserfs_read_bitmap_block(struct super_block *sb,
1234 unsigned int bitmap)
1236 b_blocknr_t block = (sb->s_blocksize << 3) * bitmap;
1237 struct reiserfs_bitmap_info *info = SB_AP_BITMAP(sb) + bitmap;
1238 struct buffer_head *bh;
1240 /* Way old format filesystems had the bitmaps packed up front.
1241 * I doubt there are any of these left, but just in case... */
1242 if (unlikely(test_bit(REISERFS_OLD_FORMAT,
1243 &(REISERFS_SB(sb)->s_properties))))
1244 block = REISERFS_SB(sb)->s_sbh->b_blocknr + 1 + bitmap;
1245 else if (bitmap == 0)
1246 block = (REISERFS_DISK_OFFSET_IN_BYTES >> sb->s_blocksize_bits) + 1;
1248 bh = sb_bread(sb, block);
1250 reiserfs_warning(sb, "sh-2029: %s: bitmap block (#%u) "
1251 "reading failed", __FUNCTION__, block);
1253 if (buffer_locked(bh)) {
1254 PROC_INFO_INC(sb, scan_bitmap.wait);
1255 __wait_on_buffer(bh);
1257 BUG_ON(!buffer_uptodate(bh));
1258 BUG_ON(atomic_read(&bh->b_count) == 0);
1260 if (info->first_zero_hint == 0)
1261 reiserfs_cache_bitmap_metadata(sb, bh, info);
1267 int reiserfs_init_bitmap_cache(struct super_block *sb)
1269 struct reiserfs_bitmap_info *bitmap;
1271 bitmap = vmalloc(sizeof (*bitmap) * SB_BMAP_NR(sb));
1275 memset(bitmap, 0, sizeof (*bitmap) * SB_BMAP_NR(sb));
1277 SB_AP_BITMAP(sb) = bitmap;
1282 void reiserfs_free_bitmap_cache(struct super_block *sb)
1284 if (SB_AP_BITMAP(sb)) {
1285 vfree(SB_AP_BITMAP(sb));
1286 SB_AP_BITMAP(sb) = NULL;