2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
4 /* Reiserfs block (de)allocator, bitmap-based. */
6 #include <linux/config.h>
7 #include <linux/time.h>
8 #include <linux/reiserfs_fs.h>
9 #include <linux/errno.h>
10 #include <linux/buffer_head.h>
11 #include <linux/kernel.h>
12 #include <linux/pagemap.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 << 3);
55 /* Within that bitmap block it is located at bit offset *offset. */
56 *offset = block & ((s->s_blocksize << 3) - 1);
60 #ifdef CONFIG_REISERFS_CHECK
61 int is_reusable(struct super_block *s, b_blocknr_t block, int bit_value)
65 if (block == 0 || block >= SB_BLOCK_COUNT(s)) {
67 "vs-4010: is_reusable: block number is out of range %lu (%u)",
68 block, SB_BLOCK_COUNT(s));
72 /* it can't be one of the bitmap blocks */
73 for (i = 0; i < SB_BMAP_NR(s); i++)
74 if (block == SB_AP_BITMAP(s)[i].bh->b_blocknr) {
75 reiserfs_warning(s, "vs: 4020: is_reusable: "
76 "bitmap block %lu(%u) can't be freed or reused",
77 block, SB_BMAP_NR(s));
81 get_bit_address(s, block, &i, &j);
83 if (i >= SB_BMAP_NR(s)) {
85 "vs-4030: is_reusable: there is no so many bitmap blocks: "
86 "block=%lu, bitmap_nr=%d", block, i);
90 if ((bit_value == 0 &&
91 reiserfs_test_le_bit(j, SB_AP_BITMAP(s)[i].bh->b_data)) ||
93 reiserfs_test_le_bit(j, SB_AP_BITMAP(s)[i].bh->b_data) == 0)) {
95 "vs-4040: is_reusable: corresponding bit of block %lu does not "
96 "match required value (i==%d, j==%d) test_bit==%d",
97 block, i, j, reiserfs_test_le_bit(j,
105 if (bit_value == 0 && block == SB_ROOT_BLOCK(s)) {
107 "vs-4050: is_reusable: this is root block (%u), "
108 "it must be busy", SB_ROOT_BLOCK(s));
114 #endif /* CONFIG_REISERFS_CHECK */
116 /* searches in journal structures for a given block number (bmap, off). If block
117 is found in reiserfs journal it suggests next free block candidate to test. */
118 static inline int is_block_in_journal(struct super_block *s, int bmap, int
123 if (reiserfs_in_journal(s, bmap, off, 1, &tmp)) {
124 if (tmp) { /* hint supplied */
126 PROC_INFO_INC(s, scan_bitmap.in_journal_hint);
128 (*next) = off + 1; /* inc offset to avoid looping. */
129 PROC_INFO_INC(s, scan_bitmap.in_journal_nohint);
131 PROC_INFO_INC(s, scan_bitmap.retry);
137 /* it searches for a window of zero bits with given minimum and maximum lengths in one bitmap
139 static int scan_bitmap_block(struct reiserfs_transaction_handle *th,
140 int bmap_n, int *beg, int boundary, int min,
143 struct super_block *s = th->t_super;
144 struct reiserfs_bitmap_info *bi = &SB_AP_BITMAP(s)[bmap_n];
148 BUG_ON(!th->t_trans_id);
150 RFALSE(bmap_n >= SB_BMAP_NR(s), "Bitmap %d is out of range (0..%d)",
151 bmap_n, SB_BMAP_NR(s) - 1);
152 PROC_INFO_INC(s, scan_bitmap.bmap);
153 /* this is unclear and lacks comments, explain how journal bitmaps
154 work here for the reader. Convey a sense of the design here. What
156 /* - I mean `a window of zero bits' as in description of this function - Zam. */
159 reiserfs_warning(s, "NULL bitmap info pointer for bitmap %d",
163 if (buffer_locked(bi->bh)) {
164 PROC_INFO_INC(s, scan_bitmap.wait);
165 __wait_on_buffer(bi->bh);
170 if (bi->free_count < min)
171 return 0; // No free blocks in this bitmap
173 /* search for a first zero bit -- beggining of a window */
174 *beg = reiserfs_find_next_zero_le_bit
175 ((unsigned long *)(bi->bh->b_data), boundary, *beg);
177 if (*beg + min > boundary) { /* search for a zero bit fails or the rest of bitmap block
178 * cannot contain a zero window of minimum size */
182 if (unfm && is_block_in_journal(s, bmap_n, *beg, beg))
184 /* first zero bit found; we check next bits */
185 for (end = *beg + 1;; end++) {
186 if (end >= *beg + max || end >= boundary
187 || reiserfs_test_le_bit(end, bi->bh->b_data)) {
191 /* finding the other end of zero bit window requires looking into journal structures (in
192 * case of searching for free blocks for unformatted nodes) */
193 if (unfm && is_block_in_journal(s, bmap_n, end, &next))
197 /* now (*beg) points to beginning of zero bits window,
198 * (end) points to one bit after the window end */
199 if (end - *beg >= min) { /* it seems we have found window of proper size */
201 reiserfs_prepare_for_journal(s, bi->bh, 1);
202 /* try to set all blocks used checking are they still free */
203 for (i = *beg; i < end; i++) {
204 /* It seems that we should not check in journal again. */
205 if (reiserfs_test_and_set_le_bit
206 (i, bi->bh->b_data)) {
207 /* bit was set by another process
208 * while we slept in prepare_for_journal() */
209 PROC_INFO_INC(s, scan_bitmap.stolen);
210 if (i >= *beg + min) { /* we can continue with smaller set of allocated blocks,
211 * if length of this set is more or equal to `min' */
215 /* otherwise we clear all bit were set ... */
217 reiserfs_test_and_clear_le_bit
219 reiserfs_restore_prepared_buffer(s,
223 /* ... and search again in current block from beginning */
227 bi->free_count -= (end - *beg);
228 journal_mark_dirty(th, s, bi->bh);
230 /* free block count calculation */
231 reiserfs_prepare_for_journal(s, SB_BUFFER_WITH_SB(s),
233 PUT_SB_FREE_BLOCKS(s, SB_FREE_BLOCKS(s) - (end - *beg));
234 journal_mark_dirty(th, s, SB_BUFFER_WITH_SB(s));
243 static int bmap_hash_id(struct super_block *s, u32 id)
245 char *hash_in = NULL;
252 hash_in = (char *)(&id);
253 hash = keyed_hash(hash_in, 4);
254 bm = hash % SB_BMAP_NR(s);
258 /* this can only be true when SB_BMAP_NR = 1 */
259 if (bm >= SB_BMAP_NR(s))
265 * hashes the id and then returns > 0 if the block group for the
266 * corresponding hash is full
268 static inline int block_group_used(struct super_block *s, u32 id)
271 bm = bmap_hash_id(s, id);
272 if (SB_AP_BITMAP(s)[bm].free_count > ((s->s_blocksize << 3) * 60 / 100)) {
279 * the packing is returned in disk byte order
281 __le32 reiserfs_choose_packing(struct inode * dir)
284 if (TEST_OPTION(packing_groups, dir->i_sb)) {
285 u32 parent_dir = le32_to_cpu(INODE_PKEY(dir)->k_dir_id);
287 * some versions of reiserfsck expect packing locality 1 to be
290 if (parent_dir == 1 || block_group_used(dir->i_sb, parent_dir))
291 packing = INODE_PKEY(dir)->k_objectid;
293 packing = INODE_PKEY(dir)->k_dir_id;
295 packing = INODE_PKEY(dir)->k_objectid;
299 /* Tries to find contiguous zero bit window (given size) in given region of
300 * bitmap and place new blocks there. Returns number of allocated blocks. */
301 static int scan_bitmap(struct reiserfs_transaction_handle *th,
302 b_blocknr_t * start, b_blocknr_t finish,
303 int min, int max, int unfm, unsigned long file_block)
305 int nr_allocated = 0;
306 struct super_block *s = th->t_super;
307 /* find every bm and bmap and bmap_nr in this file, and change them all to bitmap_blocknr
308 * - Hans, it is not a block number - Zam. */
312 int off_max = s->s_blocksize << 3;
314 BUG_ON(!th->t_trans_id);
316 PROC_INFO_INC(s, scan_bitmap.call);
317 if (SB_FREE_BLOCKS(s) <= 0)
318 return 0; // No point in looking for more free blocks
320 get_bit_address(s, *start, &bm, &off);
321 get_bit_address(s, finish, &end_bm, &end_off);
322 if (bm > SB_BMAP_NR(s))
324 if (end_bm > SB_BMAP_NR(s))
325 end_bm = SB_BMAP_NR(s);
327 /* When the bitmap is more than 10% free, anyone can allocate.
328 * When it's less than 10% free, only files that already use the
329 * bitmap are allowed. Once we pass 80% full, this restriction
332 * We do this so that files that grow later still have space close to
333 * their original allocation. This improves locality, and presumably
334 * performance as a result.
336 * This is only an allocation policy and does not make up for getting a
337 * bad hint. Decent hinting must be implemented for this to work well.
339 if (TEST_OPTION(skip_busy, s)
340 && SB_FREE_BLOCKS(s) > SB_BLOCK_COUNT(s) / 20) {
341 for (; bm < end_bm; bm++, off = 0) {
342 if ((off && (!unfm || (file_block != 0)))
343 || SB_AP_BITMAP(s)[bm].free_count >
344 (s->s_blocksize << 3) / 10)
346 scan_bitmap_block(th, bm, &off, off_max,
351 /* we know from above that start is a reasonable number */
352 get_bit_address(s, *start, &bm, &off);
355 for (; bm < end_bm; bm++, off = 0) {
357 scan_bitmap_block(th, bm, &off, off_max, min, max, unfm);
363 scan_bitmap_block(th, bm, &off, end_off + 1, min, max, unfm);
366 *start = bm * off_max + off;
371 static void _reiserfs_free_block(struct reiserfs_transaction_handle *th,
372 struct inode *inode, b_blocknr_t block,
375 struct super_block *s = th->t_super;
376 struct reiserfs_super_block *rs;
377 struct buffer_head *sbh;
378 struct reiserfs_bitmap_info *apbi;
381 BUG_ON(!th->t_trans_id);
383 PROC_INFO_INC(s, free_block);
385 rs = SB_DISK_SUPER_BLOCK(s);
386 sbh = SB_BUFFER_WITH_SB(s);
387 apbi = SB_AP_BITMAP(s);
389 get_bit_address(s, block, &nr, &offset);
391 if (nr >= sb_bmap_nr(rs)) {
392 reiserfs_warning(s, "vs-4075: reiserfs_free_block: "
393 "block %lu is out of range on %s",
394 block, reiserfs_bdevname(s));
398 reiserfs_prepare_for_journal(s, apbi[nr].bh, 1);
400 /* clear bit for the given block in bit map */
401 if (!reiserfs_test_and_clear_le_bit(offset, apbi[nr].bh->b_data)) {
402 reiserfs_warning(s, "vs-4080: reiserfs_free_block: "
403 "free_block (%s:%lu)[dev:blocknr]: bit already cleared",
404 reiserfs_bdevname(s), block);
406 apbi[nr].free_count++;
407 journal_mark_dirty(th, s, apbi[nr].bh);
409 reiserfs_prepare_for_journal(s, sbh, 1);
410 /* update super block */
411 set_sb_free_blocks(rs, sb_free_blocks(rs) + 1);
413 journal_mark_dirty(th, s, sbh);
415 DQUOT_FREE_BLOCK_NODIRTY(inode, 1);
418 void reiserfs_free_block(struct reiserfs_transaction_handle *th,
419 struct inode *inode, b_blocknr_t block,
422 struct super_block *s = th->t_super;
424 BUG_ON(!th->t_trans_id);
426 RFALSE(!s, "vs-4061: trying to free block on nonexistent device");
427 RFALSE(is_reusable(s, block, 1) == 0,
428 "vs-4071: can not free such block");
429 /* mark it before we clear it, just in case */
430 journal_mark_freed(th, s, block);
431 _reiserfs_free_block(th, inode, block, for_unformatted);
434 /* preallocated blocks don't need to be run through journal_mark_freed */
435 static void reiserfs_free_prealloc_block(struct reiserfs_transaction_handle *th,
436 struct inode *inode, b_blocknr_t block)
439 "vs-4060: trying to free block on nonexistent device");
440 RFALSE(is_reusable(th->t_super, block, 1) == 0,
441 "vs-4070: can not free such block");
442 BUG_ON(!th->t_trans_id);
443 _reiserfs_free_block(th, inode, block, 1);
446 static void __discard_prealloc(struct reiserfs_transaction_handle *th,
447 struct reiserfs_inode_info *ei)
449 unsigned long save = ei->i_prealloc_block;
451 struct inode *inode = &ei->vfs_inode;
452 BUG_ON(!th->t_trans_id);
453 #ifdef CONFIG_REISERFS_CHECK
454 if (ei->i_prealloc_count < 0)
455 reiserfs_warning(th->t_super,
456 "zam-4001:%s: inode has negative prealloc blocks count.",
459 while (ei->i_prealloc_count > 0) {
460 reiserfs_free_prealloc_block(th, inode, ei->i_prealloc_block);
461 ei->i_prealloc_block++;
462 ei->i_prealloc_count--;
466 reiserfs_update_sd(th, inode);
467 ei->i_prealloc_block = save;
468 list_del_init(&(ei->i_prealloc_list));
471 /* FIXME: It should be inline function */
472 void reiserfs_discard_prealloc(struct reiserfs_transaction_handle *th,
475 struct reiserfs_inode_info *ei = REISERFS_I(inode);
476 BUG_ON(!th->t_trans_id);
477 if (ei->i_prealloc_count)
478 __discard_prealloc(th, ei);
481 void reiserfs_discard_all_prealloc(struct reiserfs_transaction_handle *th)
483 struct list_head *plist = &SB_JOURNAL(th->t_super)->j_prealloc_list;
485 BUG_ON(!th->t_trans_id);
487 while (!list_empty(plist)) {
488 struct reiserfs_inode_info *ei;
489 ei = list_entry(plist->next, struct reiserfs_inode_info,
491 #ifdef CONFIG_REISERFS_CHECK
492 if (!ei->i_prealloc_count) {
493 reiserfs_warning(th->t_super,
494 "zam-4001:%s: inode is in prealloc list but has no preallocated blocks.",
498 __discard_prealloc(th, ei);
502 void reiserfs_init_alloc_options(struct super_block *s)
504 set_bit(_ALLOC_skip_busy, &SB_ALLOC_OPTS(s));
505 set_bit(_ALLOC_dirid_groups, &SB_ALLOC_OPTS(s));
506 set_bit(_ALLOC_packing_groups, &SB_ALLOC_OPTS(s));
509 /* block allocator related options are parsed here */
510 int reiserfs_parse_alloc_options(struct super_block *s, char *options)
512 char *this_char, *value;
514 REISERFS_SB(s)->s_alloc_options.bits = 0; /* clear default settings */
516 while ((this_char = strsep(&options, ":")) != NULL) {
517 if ((value = strchr(this_char, '=')) != NULL)
520 if (!strcmp(this_char, "concentrating_formatted_nodes")) {
522 SET_OPTION(concentrating_formatted_nodes);
524 && *value) ? simple_strtoul(value, &value,
526 if (temp <= 0 || temp > 100) {
527 REISERFS_SB(s)->s_alloc_options.border = 10;
529 REISERFS_SB(s)->s_alloc_options.border =
534 if (!strcmp(this_char, "displacing_large_files")) {
535 SET_OPTION(displacing_large_files);
536 REISERFS_SB(s)->s_alloc_options.large_file_size =
538 && *value) ? simple_strtoul(value, &value, 0) : 16;
541 if (!strcmp(this_char, "displacing_new_packing_localities")) {
542 SET_OPTION(displacing_new_packing_localities);
546 if (!strcmp(this_char, "old_hashed_relocation")) {
547 SET_OPTION(old_hashed_relocation);
551 if (!strcmp(this_char, "new_hashed_relocation")) {
552 SET_OPTION(new_hashed_relocation);
556 if (!strcmp(this_char, "dirid_groups")) {
557 SET_OPTION(dirid_groups);
560 if (!strcmp(this_char, "oid_groups")) {
561 SET_OPTION(oid_groups);
564 if (!strcmp(this_char, "packing_groups")) {
565 SET_OPTION(packing_groups);
568 if (!strcmp(this_char, "hashed_formatted_nodes")) {
569 SET_OPTION(hashed_formatted_nodes);
573 if (!strcmp(this_char, "skip_busy")) {
574 SET_OPTION(skip_busy);
578 if (!strcmp(this_char, "hundredth_slices")) {
579 SET_OPTION(hundredth_slices);
583 if (!strcmp(this_char, "old_way")) {
588 if (!strcmp(this_char, "displace_based_on_dirid")) {
589 SET_OPTION(displace_based_on_dirid);
593 if (!strcmp(this_char, "preallocmin")) {
594 REISERFS_SB(s)->s_alloc_options.preallocmin =
596 && *value) ? simple_strtoul(value, &value, 0) : 4;
600 if (!strcmp(this_char, "preallocsize")) {
601 REISERFS_SB(s)->s_alloc_options.preallocsize =
603 && *value) ? simple_strtoul(value, &value,
609 reiserfs_warning(s, "zam-4001: %s : unknown option - %s",
610 __FUNCTION__, this_char);
614 reiserfs_warning(s, "allocator options = [%08x]\n", SB_ALLOC_OPTS(s));
618 static inline void new_hashed_relocation(reiserfs_blocknr_hint_t * hint)
621 if (hint->formatted_node) {
622 hash_in = (char *)&hint->key.k_dir_id;
625 //hint->search_start = hint->beg;
626 hash_in = (char *)&hint->key.k_dir_id;
628 if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
629 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_dir_id);
632 (char *)(&INODE_PKEY(hint->inode)->k_objectid);
636 hint->beg + keyed_hash(hash_in, 4) % (hint->end - hint->beg);
640 * Relocation based on dirid, hashing them into a given bitmap block
641 * files. Formatted nodes are unaffected, a seperate policy covers them
643 static void dirid_groups(reiserfs_blocknr_hint_t * hint)
648 struct super_block *sb = hint->th->t_super;
650 dirid = le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id);
651 else if (hint->formatted_node)
652 dirid = hint->key.k_dir_id;
655 bm = bmap_hash_id(sb, dirid);
656 hash = bm * (sb->s_blocksize << 3);
657 /* give a portion of the block group to metadata */
659 hash += sb->s_blocksize / 2;
660 hint->search_start = hash;
665 * Relocation based on oid, hashing them into a given bitmap block
666 * files. Formatted nodes are unaffected, a seperate policy covers them
668 static void oid_groups(reiserfs_blocknr_hint_t * hint)
676 dirid = le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id);
678 /* keep the root dir and it's first set of subdirs close to
679 * the start of the disk
682 hash = (hint->inode->i_sb->s_blocksize << 3);
684 oid = le32_to_cpu(INODE_PKEY(hint->inode)->k_objectid);
685 bm = bmap_hash_id(hint->inode->i_sb, oid);
686 hash = bm * (hint->inode->i_sb->s_blocksize << 3);
688 hint->search_start = hash;
692 /* returns 1 if it finds an indirect item and gets valid hint info
693 * from it, otherwise 0
695 static int get_left_neighbor(reiserfs_blocknr_hint_t * hint)
698 struct buffer_head *bh;
699 struct item_head *ih;
704 if (!hint->path) /* reiserfs code can call this function w/o pointer to path
705 * structure supplied; then we rely on supplied search_start */
709 bh = get_last_bh(path);
710 RFALSE(!bh, "green-4002: Illegal path specified to get_left_neighbor");
712 pos_in_item = path->pos_in_item;
713 item = get_item(path);
715 hint->search_start = bh->b_blocknr;
717 if (!hint->formatted_node && is_indirect_le_ih(ih)) {
718 /* for indirect item: go to left and look for the first non-hole entry
719 in the indirect item */
720 if (pos_in_item == I_UNFM_NUM(ih))
722 // pos_in_item = I_UNFM_NUM (ih) - 1;
723 while (pos_in_item >= 0) {
724 int t = get_block_num(item, pos_in_item);
726 hint->search_start = t;
734 /* does result value fit into specified region? */
738 /* should be, if formatted node, then try to put on first part of the device
739 specified as number of percent with mount option device, else try to put
740 on last of device. This is not to say it is good code to do so,
741 but the effect should be measured. */
742 static inline void set_border_in_hint(struct super_block *s,
743 reiserfs_blocknr_hint_t * hint)
746 SB_BLOCK_COUNT(s) / REISERFS_SB(s)->s_alloc_options.border;
748 if (hint->formatted_node)
749 hint->end = border - 1;
754 static inline void displace_large_file(reiserfs_blocknr_hint_t * hint)
756 if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
759 keyed_hash((char *)(&INODE_PKEY(hint->inode)->k_dir_id),
760 4) % (hint->end - hint->beg);
764 keyed_hash((char *)(&INODE_PKEY(hint->inode)->k_objectid),
765 4) % (hint->end - hint->beg);
768 static inline void hash_formatted_node(reiserfs_blocknr_hint_t * hint)
773 hash_in = (char *)&hint->key.k_dir_id;
774 else if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
775 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_dir_id);
777 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_objectid);
780 hint->beg + keyed_hash(hash_in, 4) % (hint->end - hint->beg);
784 this_blocknr_allocation_would_make_it_a_large_file(reiserfs_blocknr_hint_t *
787 return hint->block ==
788 REISERFS_SB(hint->th->t_super)->s_alloc_options.large_file_size;
791 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
792 static inline void displace_new_packing_locality(reiserfs_blocknr_hint_t * hint)
794 struct in_core_key *key = &hint->key;
796 hint->th->displace_new_blocks = 0;
798 hint->beg + keyed_hash((char *)(&key->k_objectid),
799 4) % (hint->end - hint->beg);
803 static inline int old_hashed_relocation(reiserfs_blocknr_hint_t * hint)
808 if (hint->formatted_node || hint->inode == NULL) {
812 hash_in = le32_to_cpu((INODE_PKEY(hint->inode))->k_dir_id);
814 hint->beg + (u32) keyed_hash(((char *)(&hash_in)),
815 4) % (hint->end - hint->beg - 1);
816 if (border > hint->search_start)
817 hint->search_start = border;
822 static inline int old_way(reiserfs_blocknr_hint_t * hint)
826 if (hint->formatted_node || hint->inode == NULL) {
832 le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id) % (hint->end -
834 if (border > hint->search_start)
835 hint->search_start = border;
840 static inline void hundredth_slices(reiserfs_blocknr_hint_t * hint)
842 struct in_core_key *key = &hint->key;
843 b_blocknr_t slice_start;
846 (keyed_hash((char *)(&key->k_dir_id), 4) % 100) * (hint->end / 100);
847 if (slice_start > hint->search_start
848 || slice_start + (hint->end / 100) <= hint->search_start) {
849 hint->search_start = slice_start;
853 static void determine_search_start(reiserfs_blocknr_hint_t * hint,
856 struct super_block *s = hint->th->t_super;
860 hint->end = SB_BLOCK_COUNT(s) - 1;
862 /* This is former border algorithm. Now with tunable border offset */
863 if (concentrating_formatted_nodes(s))
864 set_border_in_hint(s, hint);
866 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
867 /* whenever we create a new directory, we displace it. At first we will
868 hash for location, later we might look for a moderately empty place for
870 if (displacing_new_packing_localities(s)
871 && hint->th->displace_new_blocks) {
872 displace_new_packing_locality(hint);
874 /* we do not continue determine_search_start,
875 * if new packing locality is being displaced */
880 /* all persons should feel encouraged to add more special cases here and
883 if (displacing_large_files(s) && !hint->formatted_node
884 && this_blocknr_allocation_would_make_it_a_large_file(hint)) {
885 displace_large_file(hint);
889 /* if none of our special cases is relevant, use the left neighbor in the
890 tree order of the new node we are allocating for */
891 if (hint->formatted_node && TEST_OPTION(hashed_formatted_nodes, s)) {
892 hash_formatted_node(hint);
896 unfm_hint = get_left_neighbor(hint);
898 /* Mimic old block allocator behaviour, that is if VFS allowed for preallocation,
899 new blocks are displaced based on directory ID. Also, if suggested search_start
900 is less than last preallocated block, we start searching from it, assuming that
901 HDD dataflow is faster in forward direction */
902 if (TEST_OPTION(old_way, s)) {
903 if (!hint->formatted_node) {
904 if (!reiserfs_hashed_relocation(s))
906 else if (!reiserfs_no_unhashed_relocation(s))
907 old_hashed_relocation(hint);
910 && hint->search_start <
911 REISERFS_I(hint->inode)->i_prealloc_block)
913 REISERFS_I(hint->inode)->i_prealloc_block;
918 /* This is an approach proposed by Hans */
919 if (TEST_OPTION(hundredth_slices, s)
920 && !(displacing_large_files(s) && !hint->formatted_node)) {
921 hundredth_slices(hint);
925 /* old_hashed_relocation only works on unformatted */
926 if (!unfm_hint && !hint->formatted_node &&
927 TEST_OPTION(old_hashed_relocation, s)) {
928 old_hashed_relocation(hint);
930 /* new_hashed_relocation works with both formatted/unformatted nodes */
931 if ((!unfm_hint || hint->formatted_node) &&
932 TEST_OPTION(new_hashed_relocation, s)) {
933 new_hashed_relocation(hint);
935 /* dirid grouping works only on unformatted nodes */
936 if (!unfm_hint && !hint->formatted_node && TEST_OPTION(dirid_groups, s)) {
939 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
940 if (hint->formatted_node && TEST_OPTION(dirid_groups, s)) {
945 /* oid grouping works only on unformatted nodes */
946 if (!unfm_hint && !hint->formatted_node && TEST_OPTION(oid_groups, s)) {
952 static int determine_prealloc_size(reiserfs_blocknr_hint_t * hint)
954 /* make minimum size a mount option and benchmark both ways */
955 /* we preallocate blocks only for regular files, specific size */
956 /* benchmark preallocating always and see what happens */
958 hint->prealloc_size = 0;
960 if (!hint->formatted_node && hint->preallocate) {
961 if (S_ISREG(hint->inode->i_mode)
962 && hint->inode->i_size >=
963 REISERFS_SB(hint->th->t_super)->s_alloc_options.
964 preallocmin * hint->inode->i_sb->s_blocksize)
965 hint->prealloc_size =
966 REISERFS_SB(hint->th->t_super)->s_alloc_options.
972 /* XXX I know it could be merged with upper-level function;
973 but may be result function would be too complex. */
974 static inline int allocate_without_wrapping_disk(reiserfs_blocknr_hint_t * hint,
975 b_blocknr_t * new_blocknrs,
977 b_blocknr_t finish, int min,
981 int rest = amount_needed;
984 while (rest > 0 && start <= finish) {
985 nr_allocated = scan_bitmap(hint->th, &start, finish, min,
986 rest + prealloc_size,
987 !hint->formatted_node, hint->block);
989 if (nr_allocated == 0) /* no new blocks allocated, return */
992 /* fill free_blocknrs array first */
993 while (rest > 0 && nr_allocated > 0) {
994 *new_blocknrs++ = start++;
999 /* do we have something to fill prealloc. array also ? */
1000 if (nr_allocated > 0) {
1001 /* it means prealloc_size was greater that 0 and we do preallocation */
1002 list_add(&REISERFS_I(hint->inode)->i_prealloc_list,
1003 &SB_JOURNAL(hint->th->t_super)->
1005 REISERFS_I(hint->inode)->i_prealloc_block = start;
1006 REISERFS_I(hint->inode)->i_prealloc_count =
1012 return (amount_needed - rest);
1015 static inline int blocknrs_and_prealloc_arrays_from_search_start
1016 (reiserfs_blocknr_hint_t * hint, b_blocknr_t * new_blocknrs,
1017 int amount_needed) {
1018 struct super_block *s = hint->th->t_super;
1019 b_blocknr_t start = hint->search_start;
1020 b_blocknr_t finish = SB_BLOCK_COUNT(s) - 1;
1022 int nr_allocated = 0;
1025 determine_prealloc_size(hint);
1026 if (!hint->formatted_node) {
1028 #ifdef REISERQUOTA_DEBUG
1029 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1030 "reiserquota: allocating %d blocks id=%u",
1031 amount_needed, hint->inode->i_uid);
1034 DQUOT_ALLOC_BLOCK_NODIRTY(hint->inode, amount_needed);
1035 if (quota_ret) /* Quota exceeded? */
1036 return QUOTA_EXCEEDED;
1037 if (hint->preallocate && hint->prealloc_size) {
1038 #ifdef REISERQUOTA_DEBUG
1039 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1040 "reiserquota: allocating (prealloc) %d blocks id=%u",
1041 hint->prealloc_size, hint->inode->i_uid);
1044 DQUOT_PREALLOC_BLOCK_NODIRTY(hint->inode,
1045 hint->prealloc_size);
1047 hint->preallocate = hint->prealloc_size = 0;
1049 /* for unformatted nodes, force large allocations */
1050 bigalloc = amount_needed;
1054 /* in bigalloc mode, nr_allocated should stay zero until
1055 * the entire allocation is filled
1057 if (unlikely(bigalloc && nr_allocated)) {
1058 reiserfs_warning(s, "bigalloc is %d, nr_allocated %d\n",
1059 bigalloc, nr_allocated);
1060 /* reset things to a sane value */
1061 bigalloc = amount_needed - nr_allocated;
1064 * try pass 0 and pass 1 looking for a nice big
1065 * contiguous allocation. Then reset and look
1066 * for anything you can find.
1068 if (passno == 2 && bigalloc) {
1073 case 0: /* Search from hint->search_start to end of disk */
1074 start = hint->search_start;
1075 finish = SB_BLOCK_COUNT(s) - 1;
1077 case 1: /* Search from hint->beg to hint->search_start */
1079 finish = hint->search_start;
1081 case 2: /* Last chance: Search from 0 to hint->beg */
1085 default: /* We've tried searching everywhere, not enough space */
1086 /* Free the blocks */
1087 if (!hint->formatted_node) {
1088 #ifdef REISERQUOTA_DEBUG
1089 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1090 "reiserquota: freeing (nospace) %d blocks id=%u",
1092 hint->prealloc_size -
1094 hint->inode->i_uid);
1096 DQUOT_FREE_BLOCK_NODIRTY(hint->inode, amount_needed + hint->prealloc_size - nr_allocated); /* Free not allocated blocks */
1098 while (nr_allocated--)
1099 reiserfs_free_block(hint->th, hint->inode,
1100 new_blocknrs[nr_allocated],
1101 !hint->formatted_node);
1103 return NO_DISK_SPACE;
1105 } while ((nr_allocated += allocate_without_wrapping_disk(hint,
1116 if (!hint->formatted_node &&
1117 amount_needed + hint->prealloc_size >
1118 nr_allocated + REISERFS_I(hint->inode)->i_prealloc_count) {
1119 /* Some of preallocation blocks were not allocated */
1120 #ifdef REISERQUOTA_DEBUG
1121 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1122 "reiserquota: freeing (failed prealloc) %d blocks id=%u",
1123 amount_needed + hint->prealloc_size -
1125 REISERFS_I(hint->inode)->i_prealloc_count,
1126 hint->inode->i_uid);
1128 DQUOT_FREE_BLOCK_NODIRTY(hint->inode, amount_needed +
1129 hint->prealloc_size - nr_allocated -
1130 REISERFS_I(hint->inode)->
1137 /* grab new blocknrs from preallocated list */
1138 /* return amount still needed after using them */
1139 static int use_preallocated_list_if_available(reiserfs_blocknr_hint_t * hint,
1140 b_blocknr_t * new_blocknrs,
1143 struct inode *inode = hint->inode;
1145 if (REISERFS_I(inode)->i_prealloc_count > 0) {
1146 while (amount_needed) {
1148 *new_blocknrs++ = REISERFS_I(inode)->i_prealloc_block++;
1149 REISERFS_I(inode)->i_prealloc_count--;
1153 if (REISERFS_I(inode)->i_prealloc_count <= 0) {
1154 list_del(&REISERFS_I(inode)->i_prealloc_list);
1159 /* return amount still needed after using preallocated blocks */
1160 return amount_needed;
1163 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
1164 already reserved */ )
1166 int initial_amount_needed = amount_needed;
1168 struct super_block *s = hint->th->t_super;
1170 /* Check if there is enough space, taking into account reserved space */
1171 if (SB_FREE_BLOCKS(s) - REISERFS_SB(s)->reserved_blocks <
1172 amount_needed - reserved_by_us)
1173 return NO_DISK_SPACE;
1174 /* should this be if !hint->inode && hint->preallocate? */
1175 /* do you mean hint->formatted_node can be removed ? - Zam */
1176 /* hint->formatted_node cannot be removed because we try to access
1177 inode information here, and there is often no inode assotiated with
1178 metadata allocations - green */
1180 if (!hint->formatted_node && hint->preallocate) {
1181 amount_needed = use_preallocated_list_if_available
1182 (hint, new_blocknrs, amount_needed);
1183 if (amount_needed == 0) /* all blocknrs we need we got from
1186 new_blocknrs += (initial_amount_needed - amount_needed);
1189 /* find search start and save it in hint structure */
1190 determine_search_start(hint, amount_needed);
1191 if (hint->search_start >= SB_BLOCK_COUNT(s))
1192 hint->search_start = SB_BLOCK_COUNT(s) - 1;
1194 /* allocation itself; fill new_blocknrs and preallocation arrays */
1195 ret = blocknrs_and_prealloc_arrays_from_search_start
1196 (hint, new_blocknrs, amount_needed);
1198 /* we used prealloc. list to fill (partially) new_blocknrs array. If final allocation fails we
1199 * need to return blocks back to prealloc. list or just free them. -- Zam (I chose second
1202 if (ret != CARRY_ON) {
1203 while (amount_needed++ < initial_amount_needed) {
1204 reiserfs_free_block(hint->th, hint->inode,
1205 *(--new_blocknrs), 1);
1211 /* These 2 functions are here to provide blocks reservation to the rest of kernel */
1212 /* Reserve @blocks amount of blocks in fs pointed by @sb. Caller must make sure
1213 there are actually this much blocks on the FS available */
1214 void reiserfs_claim_blocks_to_be_allocated(struct super_block *sb, /* super block of
1218 int blocks /* How much to reserve */
1222 /* Fast case, if reservation is zero - exit immediately. */
1226 spin_lock(&REISERFS_SB(sb)->bitmap_lock);
1227 REISERFS_SB(sb)->reserved_blocks += blocks;
1228 spin_unlock(&REISERFS_SB(sb)->bitmap_lock);
1231 /* Unreserve @blocks amount of blocks in fs pointed by @sb */
1232 void reiserfs_release_claimed_blocks(struct super_block *sb, /* super block of
1236 int blocks /* How much to unreserve */
1240 /* Fast case, if unreservation is zero - exit immediately. */
1244 spin_lock(&REISERFS_SB(sb)->bitmap_lock);
1245 REISERFS_SB(sb)->reserved_blocks -= blocks;
1246 spin_unlock(&REISERFS_SB(sb)->bitmap_lock);
1247 RFALSE(REISERFS_SB(sb)->reserved_blocks < 0,
1248 "amount of blocks reserved became zero?");
1251 /* This function estimates how much pages we will be able to write to FS
1252 used for reiserfs_file_write() purposes for now. */
1253 int reiserfs_can_fit_pages(struct super_block *sb /* superblock of filesystem
1254 to estimate space */ )
1258 spin_lock(&REISERFS_SB(sb)->bitmap_lock);
1260 (SB_FREE_BLOCKS(sb) -
1261 REISERFS_SB(sb)->reserved_blocks) >> (PAGE_CACHE_SHIFT -
1262 sb->s_blocksize_bits);
1263 spin_unlock(&REISERFS_SB(sb)->bitmap_lock);
1265 return space > 0 ? space : 0;