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/reiserfs_fs_sb.h>
13 #include <linux/reiserfs_fs_i.h>
14 #include <linux/quotaops.h>
16 #define PREALLOCATION_SIZE 9
18 /* different reiserfs block allocator options */
20 #define SB_ALLOC_OPTS(s) (REISERFS_SB(s)->s_alloc_options.bits)
22 #define _ALLOC_concentrating_formatted_nodes 0
23 #define _ALLOC_displacing_large_files 1
24 #define _ALLOC_displacing_new_packing_localities 2
25 #define _ALLOC_old_hashed_relocation 3
26 #define _ALLOC_new_hashed_relocation 4
27 #define _ALLOC_skip_busy 5
28 #define _ALLOC_displace_based_on_dirid 6
29 #define _ALLOC_hashed_formatted_nodes 7
30 #define _ALLOC_old_way 8
31 #define _ALLOC_hundredth_slices 9
32 #define _ALLOC_dirid_groups 10
33 #define _ALLOC_oid_groups 11
34 #define _ALLOC_packing_groups 12
36 #define concentrating_formatted_nodes(s) test_bit(_ALLOC_concentrating_formatted_nodes, &SB_ALLOC_OPTS(s))
37 #define displacing_large_files(s) test_bit(_ALLOC_displacing_large_files, &SB_ALLOC_OPTS(s))
38 #define displacing_new_packing_localities(s) test_bit(_ALLOC_displacing_new_packing_localities, &SB_ALLOC_OPTS(s))
40 #define SET_OPTION(optname) \
42 reiserfs_warning(s, "reiserfs: option \"%s\" is set", #optname); \
43 set_bit(_ALLOC_ ## optname , &SB_ALLOC_OPTS(s)); \
45 #define TEST_OPTION(optname, s) \
46 test_bit(_ALLOC_ ## optname , &SB_ALLOC_OPTS(s))
48 static inline void get_bit_address(struct super_block *s,
49 b_blocknr_t block, int *bmap_nr, int *offset)
51 /* It is in the bitmap block number equal to the block
52 * number divided by the number of bits in a block. */
53 *bmap_nr = block >> (s->s_blocksize_bits + 3);
54 /* Within that bitmap block it is located at bit offset *offset. */
55 *offset = block & ((s->s_blocksize << 3) - 1);
58 #ifdef CONFIG_REISERFS_CHECK
59 int is_reusable(struct super_block *s, b_blocknr_t block, int bit_value)
62 struct buffer_head *bh;
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 bh = SB_AP_BITMAP(s)[bmap].bh;
103 if ((bit_value == 0 && reiserfs_test_le_bit(offset, bh->b_data)) ||
104 (bit_value == 1 && reiserfs_test_le_bit(offset, bh->b_data) == 0)) {
106 "vs-4040: is_reusable: corresponding bit of block %lu does not "
107 "match required value (bmap==%d, offset==%d) test_bit==%d",
109 reiserfs_test_le_bit(offset, bh->b_data));
116 if (bit_value == 0 && block == SB_ROOT_BLOCK(s)) {
118 "vs-4050: is_reusable: this is root block (%u), "
119 "it must be busy", SB_ROOT_BLOCK(s));
125 #endif /* CONFIG_REISERFS_CHECK */
127 /* searches in journal structures for a given block number (bmap, off). If block
128 is found in reiserfs journal it suggests next free block candidate to test. */
129 static inline int is_block_in_journal(struct super_block *s, int bmap, int
134 if (reiserfs_in_journal(s, bmap, off, 1, &tmp)) {
135 if (tmp) { /* hint supplied */
137 PROC_INFO_INC(s, scan_bitmap.in_journal_hint);
139 (*next) = off + 1; /* inc offset to avoid looping. */
140 PROC_INFO_INC(s, scan_bitmap.in_journal_nohint);
142 PROC_INFO_INC(s, scan_bitmap.retry);
148 /* it searches for a window of zero bits with given minimum and maximum lengths in one bitmap
150 static int scan_bitmap_block(struct reiserfs_transaction_handle *th,
151 int bmap_n, int *beg, int boundary, int min,
154 struct super_block *s = th->t_super;
155 struct reiserfs_bitmap_info *bi = &SB_AP_BITMAP(s)[bmap_n];
156 struct buffer_head *bh;
160 BUG_ON(!th->t_trans_id);
162 RFALSE(bmap_n >= SB_BMAP_NR(s), "Bitmap %d is out of range (0..%d)",
163 bmap_n, SB_BMAP_NR(s) - 1);
164 PROC_INFO_INC(s, scan_bitmap.bmap);
165 /* this is unclear and lacks comments, explain how journal bitmaps
166 work here for the reader. Convey a sense of the design here. What
168 /* - I mean `a window of zero bits' as in description of this function - Zam. */
171 reiserfs_warning(s, "NULL bitmap info pointer for bitmap %d",
178 if (buffer_locked(bh)) {
179 PROC_INFO_INC(s, scan_bitmap.wait);
180 __wait_on_buffer(bh);
185 if (bi->free_count < min) {
187 return 0; // No free blocks in this bitmap
190 /* search for a first zero bit -- beggining of a window */
191 *beg = reiserfs_find_next_zero_le_bit
192 ((unsigned long *)(bh->b_data), boundary, *beg);
194 if (*beg + min > boundary) { /* search for a zero bit fails or the rest of bitmap block
195 * cannot contain a zero window of minimum size */
200 if (unfm && is_block_in_journal(s, bmap_n, *beg, beg))
202 /* first zero bit found; we check next bits */
203 for (end = *beg + 1;; end++) {
204 if (end >= *beg + max || end >= boundary
205 || reiserfs_test_le_bit(end, bh->b_data)) {
209 /* finding the other end of zero bit window requires looking into journal structures (in
210 * case of searching for free blocks for unformatted nodes) */
211 if (unfm && is_block_in_journal(s, bmap_n, end, &next))
215 /* now (*beg) points to beginning of zero bits window,
216 * (end) points to one bit after the window end */
217 if (end - *beg >= min) { /* it seems we have found window of proper size */
219 reiserfs_prepare_for_journal(s, bh, 1);
220 /* try to set all blocks used checking are they still free */
221 for (i = *beg; i < end; i++) {
222 /* It seems that we should not check in journal again. */
223 if (reiserfs_test_and_set_le_bit
225 /* bit was set by another process
226 * while we slept in prepare_for_journal() */
227 PROC_INFO_INC(s, scan_bitmap.stolen);
228 if (i >= *beg + min) { /* we can continue with smaller set of allocated blocks,
229 * if length of this set is more or equal to `min' */
233 /* otherwise we clear all bit were set ... */
235 reiserfs_test_and_clear_le_bit
237 reiserfs_restore_prepared_buffer(s, bh);
239 /* ... and search again in current block from beginning */
243 bi->free_count -= (end - *beg);
244 journal_mark_dirty(th, s, bh);
247 /* free block count calculation */
248 reiserfs_prepare_for_journal(s, SB_BUFFER_WITH_SB(s),
250 PUT_SB_FREE_BLOCKS(s, SB_FREE_BLOCKS(s) - (end - *beg));
251 journal_mark_dirty(th, s, SB_BUFFER_WITH_SB(s));
260 static int bmap_hash_id(struct super_block *s, u32 id)
262 char *hash_in = NULL;
269 hash_in = (char *)(&id);
270 hash = keyed_hash(hash_in, 4);
271 bm = hash % SB_BMAP_NR(s);
275 /* this can only be true when SB_BMAP_NR = 1 */
276 if (bm >= SB_BMAP_NR(s))
282 * hashes the id and then returns > 0 if the block group for the
283 * corresponding hash is full
285 static inline int block_group_used(struct super_block *s, u32 id)
288 bm = bmap_hash_id(s, id);
289 if (SB_AP_BITMAP(s)[bm].free_count > ((s->s_blocksize << 3) * 60 / 100)) {
296 * the packing is returned in disk byte order
298 __le32 reiserfs_choose_packing(struct inode * dir)
301 if (TEST_OPTION(packing_groups, dir->i_sb)) {
302 u32 parent_dir = le32_to_cpu(INODE_PKEY(dir)->k_dir_id);
304 * some versions of reiserfsck expect packing locality 1 to be
307 if (parent_dir == 1 || block_group_used(dir->i_sb, parent_dir))
308 packing = INODE_PKEY(dir)->k_objectid;
310 packing = INODE_PKEY(dir)->k_dir_id;
312 packing = INODE_PKEY(dir)->k_objectid;
316 /* Tries to find contiguous zero bit window (given size) in given region of
317 * bitmap and place new blocks there. Returns number of allocated blocks. */
318 static int scan_bitmap(struct reiserfs_transaction_handle *th,
319 b_blocknr_t * start, b_blocknr_t finish,
320 int min, int max, int unfm, unsigned long file_block)
322 int nr_allocated = 0;
323 struct super_block *s = th->t_super;
324 /* find every bm and bmap and bmap_nr in this file, and change them all to bitmap_blocknr
325 * - Hans, it is not a block number - Zam. */
329 int off_max = s->s_blocksize << 3;
331 BUG_ON(!th->t_trans_id);
333 PROC_INFO_INC(s, scan_bitmap.call);
334 if (SB_FREE_BLOCKS(s) <= 0)
335 return 0; // No point in looking for more free blocks
337 get_bit_address(s, *start, &bm, &off);
338 get_bit_address(s, finish, &end_bm, &end_off);
339 if (bm > SB_BMAP_NR(s))
341 if (end_bm > SB_BMAP_NR(s))
342 end_bm = SB_BMAP_NR(s);
344 /* When the bitmap is more than 10% free, anyone can allocate.
345 * When it's less than 10% free, only files that already use the
346 * bitmap are allowed. Once we pass 80% full, this restriction
349 * We do this so that files that grow later still have space close to
350 * their original allocation. This improves locality, and presumably
351 * performance as a result.
353 * This is only an allocation policy and does not make up for getting a
354 * bad hint. Decent hinting must be implemented for this to work well.
356 if (TEST_OPTION(skip_busy, s)
357 && SB_FREE_BLOCKS(s) > SB_BLOCK_COUNT(s) / 20) {
358 for (; bm < end_bm; bm++, off = 0) {
359 if ((off && (!unfm || (file_block != 0)))
360 || SB_AP_BITMAP(s)[bm].free_count >
361 (s->s_blocksize << 3) / 10)
363 scan_bitmap_block(th, bm, &off, off_max,
368 /* we know from above that start is a reasonable number */
369 get_bit_address(s, *start, &bm, &off);
372 for (; bm < end_bm; bm++, off = 0) {
374 scan_bitmap_block(th, bm, &off, off_max, min, max, unfm);
380 scan_bitmap_block(th, bm, &off, end_off + 1, min, max, unfm);
383 *start = bm * off_max + off;
388 static void _reiserfs_free_block(struct reiserfs_transaction_handle *th,
389 struct inode *inode, b_blocknr_t block,
392 struct super_block *s = th->t_super;
393 struct reiserfs_super_block *rs;
394 struct buffer_head *sbh, *bmbh;
395 struct reiserfs_bitmap_info *apbi;
398 BUG_ON(!th->t_trans_id);
400 PROC_INFO_INC(s, free_block);
402 rs = SB_DISK_SUPER_BLOCK(s);
403 sbh = SB_BUFFER_WITH_SB(s);
404 apbi = SB_AP_BITMAP(s);
406 get_bit_address(s, block, &nr, &offset);
408 if (nr >= sb_bmap_nr(rs)) {
409 reiserfs_warning(s, "vs-4075: reiserfs_free_block: "
410 "block %lu is out of range on %s",
411 block, reiserfs_bdevname(s));
418 reiserfs_prepare_for_journal(s, bmbh, 1);
420 /* clear bit for the given block in bit map */
421 if (!reiserfs_test_and_clear_le_bit(offset, bmbh->b_data)) {
422 reiserfs_warning(s, "vs-4080: reiserfs_free_block: "
423 "free_block (%s:%lu)[dev:blocknr]: bit already cleared",
424 reiserfs_bdevname(s), block);
426 apbi[nr].free_count++;
427 journal_mark_dirty(th, s, bmbh);
430 reiserfs_prepare_for_journal(s, sbh, 1);
431 /* update super block */
432 set_sb_free_blocks(rs, sb_free_blocks(rs) + 1);
434 journal_mark_dirty(th, s, sbh);
436 DQUOT_FREE_BLOCK_NODIRTY(inode, 1);
439 void reiserfs_free_block(struct reiserfs_transaction_handle *th,
440 struct inode *inode, b_blocknr_t block,
443 struct super_block *s = th->t_super;
445 BUG_ON(!th->t_trans_id);
447 RFALSE(!s, "vs-4061: trying to free block on nonexistent device");
448 RFALSE(is_reusable(s, block, 1) == 0,
449 "vs-4071: can not free such block");
450 /* mark it before we clear it, just in case */
451 journal_mark_freed(th, s, block);
452 _reiserfs_free_block(th, inode, block, for_unformatted);
455 /* preallocated blocks don't need to be run through journal_mark_freed */
456 static void reiserfs_free_prealloc_block(struct reiserfs_transaction_handle *th,
457 struct inode *inode, b_blocknr_t block)
460 "vs-4060: trying to free block on nonexistent device");
461 RFALSE(is_reusable(th->t_super, block, 1) == 0,
462 "vs-4070: can not free such block");
463 BUG_ON(!th->t_trans_id);
464 _reiserfs_free_block(th, inode, block, 1);
467 static void __discard_prealloc(struct reiserfs_transaction_handle *th,
468 struct reiserfs_inode_info *ei)
470 unsigned long save = ei->i_prealloc_block;
472 struct inode *inode = &ei->vfs_inode;
473 BUG_ON(!th->t_trans_id);
474 #ifdef CONFIG_REISERFS_CHECK
475 if (ei->i_prealloc_count < 0)
476 reiserfs_warning(th->t_super,
477 "zam-4001:%s: inode has negative prealloc blocks count.",
480 while (ei->i_prealloc_count > 0) {
481 reiserfs_free_prealloc_block(th, inode, ei->i_prealloc_block);
482 ei->i_prealloc_block++;
483 ei->i_prealloc_count--;
487 reiserfs_update_sd(th, inode);
488 ei->i_prealloc_block = save;
489 list_del_init(&(ei->i_prealloc_list));
492 /* FIXME: It should be inline function */
493 void reiserfs_discard_prealloc(struct reiserfs_transaction_handle *th,
496 struct reiserfs_inode_info *ei = REISERFS_I(inode);
497 BUG_ON(!th->t_trans_id);
498 if (ei->i_prealloc_count)
499 __discard_prealloc(th, ei);
502 void reiserfs_discard_all_prealloc(struct reiserfs_transaction_handle *th)
504 struct list_head *plist = &SB_JOURNAL(th->t_super)->j_prealloc_list;
506 BUG_ON(!th->t_trans_id);
508 while (!list_empty(plist)) {
509 struct reiserfs_inode_info *ei;
510 ei = list_entry(plist->next, struct reiserfs_inode_info,
512 #ifdef CONFIG_REISERFS_CHECK
513 if (!ei->i_prealloc_count) {
514 reiserfs_warning(th->t_super,
515 "zam-4001:%s: inode is in prealloc list but has no preallocated blocks.",
519 __discard_prealloc(th, ei);
523 void reiserfs_init_alloc_options(struct super_block *s)
525 set_bit(_ALLOC_skip_busy, &SB_ALLOC_OPTS(s));
526 set_bit(_ALLOC_dirid_groups, &SB_ALLOC_OPTS(s));
527 set_bit(_ALLOC_packing_groups, &SB_ALLOC_OPTS(s));
530 /* block allocator related options are parsed here */
531 int reiserfs_parse_alloc_options(struct super_block *s, char *options)
533 char *this_char, *value;
535 REISERFS_SB(s)->s_alloc_options.bits = 0; /* clear default settings */
537 while ((this_char = strsep(&options, ":")) != NULL) {
538 if ((value = strchr(this_char, '=')) != NULL)
541 if (!strcmp(this_char, "concentrating_formatted_nodes")) {
543 SET_OPTION(concentrating_formatted_nodes);
545 && *value) ? simple_strtoul(value, &value,
547 if (temp <= 0 || temp > 100) {
548 REISERFS_SB(s)->s_alloc_options.border = 10;
550 REISERFS_SB(s)->s_alloc_options.border =
555 if (!strcmp(this_char, "displacing_large_files")) {
556 SET_OPTION(displacing_large_files);
557 REISERFS_SB(s)->s_alloc_options.large_file_size =
559 && *value) ? simple_strtoul(value, &value, 0) : 16;
562 if (!strcmp(this_char, "displacing_new_packing_localities")) {
563 SET_OPTION(displacing_new_packing_localities);
567 if (!strcmp(this_char, "old_hashed_relocation")) {
568 SET_OPTION(old_hashed_relocation);
572 if (!strcmp(this_char, "new_hashed_relocation")) {
573 SET_OPTION(new_hashed_relocation);
577 if (!strcmp(this_char, "dirid_groups")) {
578 SET_OPTION(dirid_groups);
581 if (!strcmp(this_char, "oid_groups")) {
582 SET_OPTION(oid_groups);
585 if (!strcmp(this_char, "packing_groups")) {
586 SET_OPTION(packing_groups);
589 if (!strcmp(this_char, "hashed_formatted_nodes")) {
590 SET_OPTION(hashed_formatted_nodes);
594 if (!strcmp(this_char, "skip_busy")) {
595 SET_OPTION(skip_busy);
599 if (!strcmp(this_char, "hundredth_slices")) {
600 SET_OPTION(hundredth_slices);
604 if (!strcmp(this_char, "old_way")) {
609 if (!strcmp(this_char, "displace_based_on_dirid")) {
610 SET_OPTION(displace_based_on_dirid);
614 if (!strcmp(this_char, "preallocmin")) {
615 REISERFS_SB(s)->s_alloc_options.preallocmin =
617 && *value) ? simple_strtoul(value, &value, 0) : 4;
621 if (!strcmp(this_char, "preallocsize")) {
622 REISERFS_SB(s)->s_alloc_options.preallocsize =
624 && *value) ? simple_strtoul(value, &value,
630 reiserfs_warning(s, "zam-4001: %s : unknown option - %s",
631 __FUNCTION__, this_char);
635 reiserfs_warning(s, "allocator options = [%08x]\n", SB_ALLOC_OPTS(s));
639 static inline void new_hashed_relocation(reiserfs_blocknr_hint_t * hint)
642 if (hint->formatted_node) {
643 hash_in = (char *)&hint->key.k_dir_id;
646 //hint->search_start = hint->beg;
647 hash_in = (char *)&hint->key.k_dir_id;
649 if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
650 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_dir_id);
653 (char *)(&INODE_PKEY(hint->inode)->k_objectid);
657 hint->beg + keyed_hash(hash_in, 4) % (hint->end - hint->beg);
661 * Relocation based on dirid, hashing them into a given bitmap block
662 * files. Formatted nodes are unaffected, a seperate policy covers them
664 static void dirid_groups(reiserfs_blocknr_hint_t * hint)
669 struct super_block *sb = hint->th->t_super;
671 dirid = le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id);
672 else if (hint->formatted_node)
673 dirid = hint->key.k_dir_id;
676 bm = bmap_hash_id(sb, dirid);
677 hash = bm * (sb->s_blocksize << 3);
678 /* give a portion of the block group to metadata */
680 hash += sb->s_blocksize / 2;
681 hint->search_start = hash;
686 * Relocation based on oid, hashing them into a given bitmap block
687 * files. Formatted nodes are unaffected, a seperate policy covers them
689 static void oid_groups(reiserfs_blocknr_hint_t * hint)
697 dirid = le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id);
699 /* keep the root dir and it's first set of subdirs close to
700 * the start of the disk
703 hash = (hint->inode->i_sb->s_blocksize << 3);
705 oid = le32_to_cpu(INODE_PKEY(hint->inode)->k_objectid);
706 bm = bmap_hash_id(hint->inode->i_sb, oid);
707 hash = bm * (hint->inode->i_sb->s_blocksize << 3);
709 hint->search_start = hash;
713 /* returns 1 if it finds an indirect item and gets valid hint info
714 * from it, otherwise 0
716 static int get_left_neighbor(reiserfs_blocknr_hint_t * hint)
719 struct buffer_head *bh;
720 struct item_head *ih;
725 if (!hint->path) /* reiserfs code can call this function w/o pointer to path
726 * structure supplied; then we rely on supplied search_start */
730 bh = get_last_bh(path);
731 RFALSE(!bh, "green-4002: Illegal path specified to get_left_neighbor");
733 pos_in_item = path->pos_in_item;
734 item = get_item(path);
736 hint->search_start = bh->b_blocknr;
738 if (!hint->formatted_node && is_indirect_le_ih(ih)) {
739 /* for indirect item: go to left and look for the first non-hole entry
740 in the indirect item */
741 if (pos_in_item == I_UNFM_NUM(ih))
743 // pos_in_item = I_UNFM_NUM (ih) - 1;
744 while (pos_in_item >= 0) {
745 int t = get_block_num(item, pos_in_item);
747 hint->search_start = t;
755 /* does result value fit into specified region? */
759 /* should be, if formatted node, then try to put on first part of the device
760 specified as number of percent with mount option device, else try to put
761 on last of device. This is not to say it is good code to do so,
762 but the effect should be measured. */
763 static inline void set_border_in_hint(struct super_block *s,
764 reiserfs_blocknr_hint_t * hint)
767 SB_BLOCK_COUNT(s) / REISERFS_SB(s)->s_alloc_options.border;
769 if (hint->formatted_node)
770 hint->end = border - 1;
775 static inline void displace_large_file(reiserfs_blocknr_hint_t * hint)
777 if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
780 keyed_hash((char *)(&INODE_PKEY(hint->inode)->k_dir_id),
781 4) % (hint->end - hint->beg);
785 keyed_hash((char *)(&INODE_PKEY(hint->inode)->k_objectid),
786 4) % (hint->end - hint->beg);
789 static inline void hash_formatted_node(reiserfs_blocknr_hint_t * hint)
794 hash_in = (char *)&hint->key.k_dir_id;
795 else if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
796 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_dir_id);
798 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_objectid);
801 hint->beg + keyed_hash(hash_in, 4) % (hint->end - hint->beg);
805 this_blocknr_allocation_would_make_it_a_large_file(reiserfs_blocknr_hint_t *
808 return hint->block ==
809 REISERFS_SB(hint->th->t_super)->s_alloc_options.large_file_size;
812 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
813 static inline void displace_new_packing_locality(reiserfs_blocknr_hint_t * hint)
815 struct in_core_key *key = &hint->key;
817 hint->th->displace_new_blocks = 0;
819 hint->beg + keyed_hash((char *)(&key->k_objectid),
820 4) % (hint->end - hint->beg);
824 static inline int old_hashed_relocation(reiserfs_blocknr_hint_t * hint)
829 if (hint->formatted_node || hint->inode == NULL) {
833 hash_in = le32_to_cpu((INODE_PKEY(hint->inode))->k_dir_id);
835 hint->beg + (u32) keyed_hash(((char *)(&hash_in)),
836 4) % (hint->end - hint->beg - 1);
837 if (border > hint->search_start)
838 hint->search_start = border;
843 static inline int old_way(reiserfs_blocknr_hint_t * hint)
847 if (hint->formatted_node || hint->inode == NULL) {
853 le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id) % (hint->end -
855 if (border > hint->search_start)
856 hint->search_start = border;
861 static inline void hundredth_slices(reiserfs_blocknr_hint_t * hint)
863 struct in_core_key *key = &hint->key;
864 b_blocknr_t slice_start;
867 (keyed_hash((char *)(&key->k_dir_id), 4) % 100) * (hint->end / 100);
868 if (slice_start > hint->search_start
869 || slice_start + (hint->end / 100) <= hint->search_start) {
870 hint->search_start = slice_start;
874 static void determine_search_start(reiserfs_blocknr_hint_t * hint,
877 struct super_block *s = hint->th->t_super;
881 hint->end = SB_BLOCK_COUNT(s) - 1;
883 /* This is former border algorithm. Now with tunable border offset */
884 if (concentrating_formatted_nodes(s))
885 set_border_in_hint(s, hint);
887 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
888 /* whenever we create a new directory, we displace it. At first we will
889 hash for location, later we might look for a moderately empty place for
891 if (displacing_new_packing_localities(s)
892 && hint->th->displace_new_blocks) {
893 displace_new_packing_locality(hint);
895 /* we do not continue determine_search_start,
896 * if new packing locality is being displaced */
901 /* all persons should feel encouraged to add more special cases here and
904 if (displacing_large_files(s) && !hint->formatted_node
905 && this_blocknr_allocation_would_make_it_a_large_file(hint)) {
906 displace_large_file(hint);
910 /* if none of our special cases is relevant, use the left neighbor in the
911 tree order of the new node we are allocating for */
912 if (hint->formatted_node && TEST_OPTION(hashed_formatted_nodes, s)) {
913 hash_formatted_node(hint);
917 unfm_hint = get_left_neighbor(hint);
919 /* Mimic old block allocator behaviour, that is if VFS allowed for preallocation,
920 new blocks are displaced based on directory ID. Also, if suggested search_start
921 is less than last preallocated block, we start searching from it, assuming that
922 HDD dataflow is faster in forward direction */
923 if (TEST_OPTION(old_way, s)) {
924 if (!hint->formatted_node) {
925 if (!reiserfs_hashed_relocation(s))
927 else if (!reiserfs_no_unhashed_relocation(s))
928 old_hashed_relocation(hint);
931 && hint->search_start <
932 REISERFS_I(hint->inode)->i_prealloc_block)
934 REISERFS_I(hint->inode)->i_prealloc_block;
939 /* This is an approach proposed by Hans */
940 if (TEST_OPTION(hundredth_slices, s)
941 && !(displacing_large_files(s) && !hint->formatted_node)) {
942 hundredth_slices(hint);
946 /* old_hashed_relocation only works on unformatted */
947 if (!unfm_hint && !hint->formatted_node &&
948 TEST_OPTION(old_hashed_relocation, s)) {
949 old_hashed_relocation(hint);
951 /* new_hashed_relocation works with both formatted/unformatted nodes */
952 if ((!unfm_hint || hint->formatted_node) &&
953 TEST_OPTION(new_hashed_relocation, s)) {
954 new_hashed_relocation(hint);
956 /* dirid grouping works only on unformatted nodes */
957 if (!unfm_hint && !hint->formatted_node && TEST_OPTION(dirid_groups, s)) {
960 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
961 if (hint->formatted_node && TEST_OPTION(dirid_groups, s)) {
966 /* oid grouping works only on unformatted nodes */
967 if (!unfm_hint && !hint->formatted_node && TEST_OPTION(oid_groups, s)) {
973 static int determine_prealloc_size(reiserfs_blocknr_hint_t * hint)
975 /* make minimum size a mount option and benchmark both ways */
976 /* we preallocate blocks only for regular files, specific size */
977 /* benchmark preallocating always and see what happens */
979 hint->prealloc_size = 0;
981 if (!hint->formatted_node && hint->preallocate) {
982 if (S_ISREG(hint->inode->i_mode)
983 && hint->inode->i_size >=
984 REISERFS_SB(hint->th->t_super)->s_alloc_options.
985 preallocmin * hint->inode->i_sb->s_blocksize)
986 hint->prealloc_size =
987 REISERFS_SB(hint->th->t_super)->s_alloc_options.
993 /* XXX I know it could be merged with upper-level function;
994 but may be result function would be too complex. */
995 static inline int allocate_without_wrapping_disk(reiserfs_blocknr_hint_t * hint,
996 b_blocknr_t * new_blocknrs,
998 b_blocknr_t finish, int min,
1002 int rest = amount_needed;
1005 while (rest > 0 && start <= finish) {
1006 nr_allocated = scan_bitmap(hint->th, &start, finish, min,
1007 rest + prealloc_size,
1008 !hint->formatted_node, hint->block);
1010 if (nr_allocated == 0) /* no new blocks allocated, return */
1013 /* fill free_blocknrs array first */
1014 while (rest > 0 && nr_allocated > 0) {
1015 *new_blocknrs++ = start++;
1020 /* do we have something to fill prealloc. array also ? */
1021 if (nr_allocated > 0) {
1022 /* it means prealloc_size was greater that 0 and we do preallocation */
1023 list_add(&REISERFS_I(hint->inode)->i_prealloc_list,
1024 &SB_JOURNAL(hint->th->t_super)->
1026 REISERFS_I(hint->inode)->i_prealloc_block = start;
1027 REISERFS_I(hint->inode)->i_prealloc_count =
1033 return (amount_needed - rest);
1036 static inline int blocknrs_and_prealloc_arrays_from_search_start
1037 (reiserfs_blocknr_hint_t * hint, b_blocknr_t * new_blocknrs,
1038 int amount_needed) {
1039 struct super_block *s = hint->th->t_super;
1040 b_blocknr_t start = hint->search_start;
1041 b_blocknr_t finish = SB_BLOCK_COUNT(s) - 1;
1043 int nr_allocated = 0;
1046 determine_prealloc_size(hint);
1047 if (!hint->formatted_node) {
1049 #ifdef REISERQUOTA_DEBUG
1050 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1051 "reiserquota: allocating %d blocks id=%u",
1052 amount_needed, hint->inode->i_uid);
1055 DQUOT_ALLOC_BLOCK_NODIRTY(hint->inode, amount_needed);
1056 if (quota_ret) /* Quota exceeded? */
1057 return QUOTA_EXCEEDED;
1058 if (hint->preallocate && hint->prealloc_size) {
1059 #ifdef REISERQUOTA_DEBUG
1060 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1061 "reiserquota: allocating (prealloc) %d blocks id=%u",
1062 hint->prealloc_size, hint->inode->i_uid);
1065 DQUOT_PREALLOC_BLOCK_NODIRTY(hint->inode,
1066 hint->prealloc_size);
1068 hint->preallocate = hint->prealloc_size = 0;
1070 /* for unformatted nodes, force large allocations */
1071 bigalloc = amount_needed;
1075 /* in bigalloc mode, nr_allocated should stay zero until
1076 * the entire allocation is filled
1078 if (unlikely(bigalloc && nr_allocated)) {
1079 reiserfs_warning(s, "bigalloc is %d, nr_allocated %d\n",
1080 bigalloc, nr_allocated);
1081 /* reset things to a sane value */
1082 bigalloc = amount_needed - nr_allocated;
1085 * try pass 0 and pass 1 looking for a nice big
1086 * contiguous allocation. Then reset and look
1087 * for anything you can find.
1089 if (passno == 2 && bigalloc) {
1094 case 0: /* Search from hint->search_start to end of disk */
1095 start = hint->search_start;
1096 finish = SB_BLOCK_COUNT(s) - 1;
1098 case 1: /* Search from hint->beg to hint->search_start */
1100 finish = hint->search_start;
1102 case 2: /* Last chance: Search from 0 to hint->beg */
1106 default: /* We've tried searching everywhere, not enough space */
1107 /* Free the blocks */
1108 if (!hint->formatted_node) {
1109 #ifdef REISERQUOTA_DEBUG
1110 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1111 "reiserquota: freeing (nospace) %d blocks id=%u",
1113 hint->prealloc_size -
1115 hint->inode->i_uid);
1117 DQUOT_FREE_BLOCK_NODIRTY(hint->inode, amount_needed + hint->prealloc_size - nr_allocated); /* Free not allocated blocks */
1119 while (nr_allocated--)
1120 reiserfs_free_block(hint->th, hint->inode,
1121 new_blocknrs[nr_allocated],
1122 !hint->formatted_node);
1124 return NO_DISK_SPACE;
1126 } while ((nr_allocated += allocate_without_wrapping_disk(hint,
1137 if (!hint->formatted_node &&
1138 amount_needed + hint->prealloc_size >
1139 nr_allocated + REISERFS_I(hint->inode)->i_prealloc_count) {
1140 /* Some of preallocation blocks were not allocated */
1141 #ifdef REISERQUOTA_DEBUG
1142 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1143 "reiserquota: freeing (failed prealloc) %d blocks id=%u",
1144 amount_needed + hint->prealloc_size -
1146 REISERFS_I(hint->inode)->i_prealloc_count,
1147 hint->inode->i_uid);
1149 DQUOT_FREE_BLOCK_NODIRTY(hint->inode, amount_needed +
1150 hint->prealloc_size - nr_allocated -
1151 REISERFS_I(hint->inode)->
1158 /* grab new blocknrs from preallocated list */
1159 /* return amount still needed after using them */
1160 static int use_preallocated_list_if_available(reiserfs_blocknr_hint_t * hint,
1161 b_blocknr_t * new_blocknrs,
1164 struct inode *inode = hint->inode;
1166 if (REISERFS_I(inode)->i_prealloc_count > 0) {
1167 while (amount_needed) {
1169 *new_blocknrs++ = REISERFS_I(inode)->i_prealloc_block++;
1170 REISERFS_I(inode)->i_prealloc_count--;
1174 if (REISERFS_I(inode)->i_prealloc_count <= 0) {
1175 list_del(&REISERFS_I(inode)->i_prealloc_list);
1180 /* return amount still needed after using preallocated blocks */
1181 return amount_needed;
1184 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
1185 already reserved */ )
1187 int initial_amount_needed = amount_needed;
1189 struct super_block *s = hint->th->t_super;
1191 /* Check if there is enough space, taking into account reserved space */
1192 if (SB_FREE_BLOCKS(s) - REISERFS_SB(s)->reserved_blocks <
1193 amount_needed - reserved_by_us)
1194 return NO_DISK_SPACE;
1195 /* should this be if !hint->inode && hint->preallocate? */
1196 /* do you mean hint->formatted_node can be removed ? - Zam */
1197 /* hint->formatted_node cannot be removed because we try to access
1198 inode information here, and there is often no inode assotiated with
1199 metadata allocations - green */
1201 if (!hint->formatted_node && hint->preallocate) {
1202 amount_needed = use_preallocated_list_if_available
1203 (hint, new_blocknrs, amount_needed);
1204 if (amount_needed == 0) /* all blocknrs we need we got from
1207 new_blocknrs += (initial_amount_needed - amount_needed);
1210 /* find search start and save it in hint structure */
1211 determine_search_start(hint, amount_needed);
1212 if (hint->search_start >= SB_BLOCK_COUNT(s))
1213 hint->search_start = SB_BLOCK_COUNT(s) - 1;
1215 /* allocation itself; fill new_blocknrs and preallocation arrays */
1216 ret = blocknrs_and_prealloc_arrays_from_search_start
1217 (hint, new_blocknrs, amount_needed);
1219 /* we used prealloc. list to fill (partially) new_blocknrs array. If final allocation fails we
1220 * need to return blocks back to prealloc. list or just free them. -- Zam (I chose second
1223 if (ret != CARRY_ON) {
1224 while (amount_needed++ < initial_amount_needed) {
1225 reiserfs_free_block(hint->th, hint->inode,
1226 *(--new_blocknrs), 1);
1232 /* These 2 functions are here to provide blocks reservation to the rest of kernel */
1233 /* Reserve @blocks amount of blocks in fs pointed by @sb. Caller must make sure
1234 there are actually this much blocks on the FS available */
1235 void reiserfs_claim_blocks_to_be_allocated(struct super_block *sb, /* super block of
1239 int blocks /* How much to reserve */
1243 /* Fast case, if reservation is zero - exit immediately. */
1247 spin_lock(&REISERFS_SB(sb)->bitmap_lock);
1248 REISERFS_SB(sb)->reserved_blocks += blocks;
1249 spin_unlock(&REISERFS_SB(sb)->bitmap_lock);
1252 /* Unreserve @blocks amount of blocks in fs pointed by @sb */
1253 void reiserfs_release_claimed_blocks(struct super_block *sb, /* super block of
1257 int blocks /* How much to unreserve */
1261 /* Fast case, if unreservation is zero - exit immediately. */
1265 spin_lock(&REISERFS_SB(sb)->bitmap_lock);
1266 REISERFS_SB(sb)->reserved_blocks -= blocks;
1267 spin_unlock(&REISERFS_SB(sb)->bitmap_lock);
1268 RFALSE(REISERFS_SB(sb)->reserved_blocks < 0,
1269 "amount of blocks reserved became zero?");
1272 /* This function estimates how much pages we will be able to write to FS
1273 used for reiserfs_file_write() purposes for now. */
1274 int reiserfs_can_fit_pages(struct super_block *sb /* superblock of filesystem
1275 to estimate space */ )
1279 spin_lock(&REISERFS_SB(sb)->bitmap_lock);
1281 (SB_FREE_BLOCKS(sb) -
1282 REISERFS_SB(sb)->reserved_blocks) >> (PAGE_CACHE_SHIFT -
1283 sb->s_blocksize_bits);
1284 spin_unlock(&REISERFS_SB(sb)->bitmap_lock);
1286 return space > 0 ? space : 0;