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 int is_reusable(struct super_block *s, b_blocknr_t block, int bit_value)
63 if (block == 0 || block >= SB_BLOCK_COUNT(s)) {
65 "vs-4010: is_reusable: block number is out of range %lu (%u)",
66 block, SB_BLOCK_COUNT(s));
70 get_bit_address(s, block, &bmap, &offset);
72 /* Old format filesystem? Unlikely, but the bitmaps are all up front so
73 * we need to account for it. */
74 if (unlikely(test_bit(REISERFS_OLD_FORMAT,
75 &(REISERFS_SB(s)->s_properties)))) {
76 b_blocknr_t bmap1 = REISERFS_SB(s)->s_sbh->b_blocknr + 1;
77 if (block >= bmap1 && block <= bmap1 + SB_BMAP_NR(s)) {
78 reiserfs_warning(s, "vs: 4019: is_reusable: "
79 "bitmap block %lu(%u) can't be freed or reused",
80 block, SB_BMAP_NR(s));
85 reiserfs_warning(s, "vs: 4020: is_reusable: "
86 "bitmap block %lu(%u) can't be freed or reused",
87 block, SB_BMAP_NR(s));
92 if (bmap >= SB_BMAP_NR(s)) {
94 "vs-4030: is_reusable: there is no so many bitmap blocks: "
95 "block=%lu, bitmap_nr=%d", block, bmap);
99 if (bit_value == 0 && block == SB_ROOT_BLOCK(s)) {
101 "vs-4050: is_reusable: this is root block (%u), "
102 "it must be busy", SB_ROOT_BLOCK(s));
109 /* searches in journal structures for a given block number (bmap, off). If block
110 is found in reiserfs journal it suggests next free block candidate to test. */
111 static inline int is_block_in_journal(struct super_block *s, int bmap, int
116 if (reiserfs_in_journal(s, bmap, off, 1, &tmp)) {
117 if (tmp) { /* hint supplied */
119 PROC_INFO_INC(s, scan_bitmap.in_journal_hint);
121 (*next) = off + 1; /* inc offset to avoid looping. */
122 PROC_INFO_INC(s, scan_bitmap.in_journal_nohint);
124 PROC_INFO_INC(s, scan_bitmap.retry);
130 /* it searches for a window of zero bits with given minimum and maximum lengths in one bitmap
132 static int scan_bitmap_block(struct reiserfs_transaction_handle *th,
133 int bmap_n, int *beg, int boundary, int min,
136 struct super_block *s = th->t_super;
137 struct reiserfs_bitmap_info *bi = &SB_AP_BITMAP(s)[bmap_n];
138 struct buffer_head *bh;
142 BUG_ON(!th->t_trans_id);
144 RFALSE(bmap_n >= SB_BMAP_NR(s), "Bitmap %d is out of range (0..%d)",
145 bmap_n, SB_BMAP_NR(s) - 1);
146 PROC_INFO_INC(s, scan_bitmap.bmap);
147 /* this is unclear and lacks comments, explain how journal bitmaps
148 work here for the reader. Convey a sense of the design here. What
150 /* - I mean `a window of zero bits' as in description of this function - Zam. */
153 reiserfs_warning(s, "NULL bitmap info pointer for bitmap %d",
158 bh = reiserfs_read_bitmap_block(s, bmap_n);
164 if (bi->free_count < min) {
166 return 0; // No free blocks in this bitmap
169 /* search for a first zero bit -- beggining of a window */
170 *beg = reiserfs_find_next_zero_le_bit
171 ((unsigned long *)(bh->b_data), boundary, *beg);
173 if (*beg + min > boundary) { /* search for a zero bit fails or the rest of bitmap block
174 * cannot contain a zero window of minimum size */
179 if (unfm && is_block_in_journal(s, bmap_n, *beg, beg))
181 /* first zero bit found; we check next bits */
182 for (end = *beg + 1;; end++) {
183 if (end >= *beg + max || end >= boundary
184 || reiserfs_test_le_bit(end, bh->b_data)) {
188 /* finding the other end of zero bit window requires looking into journal structures (in
189 * case of searching for free blocks for unformatted nodes) */
190 if (unfm && is_block_in_journal(s, bmap_n, end, &next))
194 /* now (*beg) points to beginning of zero bits window,
195 * (end) points to one bit after the window end */
196 if (end - *beg >= min) { /* it seems we have found window of proper size */
198 reiserfs_prepare_for_journal(s, bh, 1);
199 /* try to set all blocks used checking are they still free */
200 for (i = *beg; i < end; i++) {
201 /* It seems that we should not check in journal again. */
202 if (reiserfs_test_and_set_le_bit
204 /* bit was set by another process
205 * while we slept in prepare_for_journal() */
206 PROC_INFO_INC(s, scan_bitmap.stolen);
207 if (i >= *beg + min) { /* we can continue with smaller set of allocated blocks,
208 * if length of this set is more or equal to `min' */
212 /* otherwise we clear all bit were set ... */
214 reiserfs_test_and_clear_le_bit
216 reiserfs_restore_prepared_buffer(s, bh);
218 /* ... and search again in current block from beginning */
222 bi->free_count -= (end - *beg);
223 journal_mark_dirty(th, s, bh);
226 /* free block count calculation */
227 reiserfs_prepare_for_journal(s, SB_BUFFER_WITH_SB(s),
229 PUT_SB_FREE_BLOCKS(s, SB_FREE_BLOCKS(s) - (end - *beg));
230 journal_mark_dirty(th, s, SB_BUFFER_WITH_SB(s));
239 static int bmap_hash_id(struct super_block *s, u32 id)
241 char *hash_in = NULL;
248 hash_in = (char *)(&id);
249 hash = keyed_hash(hash_in, 4);
250 bm = hash % SB_BMAP_NR(s);
254 /* this can only be true when SB_BMAP_NR = 1 */
255 if (bm >= SB_BMAP_NR(s))
261 * hashes the id and then returns > 0 if the block group for the
262 * corresponding hash is full
264 static inline int block_group_used(struct super_block *s, u32 id)
266 int bm = bmap_hash_id(s, id);
267 struct reiserfs_bitmap_info *info = &SB_AP_BITMAP(s)[bm];
269 /* If we don't have cached information on this bitmap block, we're
270 * going to have to load it later anyway. Loading it here allows us
271 * to make a better decision. This favors long-term performace gain
272 * with a better on-disk layout vs. a short term gain of skipping the
273 * read and potentially having a bad placement. */
274 if (info->first_zero_hint == 0) {
275 struct buffer_head *bh = reiserfs_read_bitmap_block(s, bm);
279 if (info->free_count > ((s->s_blocksize << 3) * 60 / 100)) {
286 * the packing is returned in disk byte order
288 __le32 reiserfs_choose_packing(struct inode * dir)
291 if (TEST_OPTION(packing_groups, dir->i_sb)) {
292 u32 parent_dir = le32_to_cpu(INODE_PKEY(dir)->k_dir_id);
294 * some versions of reiserfsck expect packing locality 1 to be
297 if (parent_dir == 1 || block_group_used(dir->i_sb, parent_dir))
298 packing = INODE_PKEY(dir)->k_objectid;
300 packing = INODE_PKEY(dir)->k_dir_id;
302 packing = INODE_PKEY(dir)->k_objectid;
306 /* Tries to find contiguous zero bit window (given size) in given region of
307 * bitmap and place new blocks there. Returns number of allocated blocks. */
308 static int scan_bitmap(struct reiserfs_transaction_handle *th,
309 b_blocknr_t * start, b_blocknr_t finish,
310 int min, int max, int unfm, unsigned long file_block)
312 int nr_allocated = 0;
313 struct super_block *s = th->t_super;
314 /* find every bm and bmap and bmap_nr in this file, and change them all to bitmap_blocknr
315 * - Hans, it is not a block number - Zam. */
319 int off_max = s->s_blocksize << 3;
321 BUG_ON(!th->t_trans_id);
323 PROC_INFO_INC(s, scan_bitmap.call);
324 if (SB_FREE_BLOCKS(s) <= 0)
325 return 0; // No point in looking for more free blocks
327 get_bit_address(s, *start, &bm, &off);
328 get_bit_address(s, finish, &end_bm, &end_off);
329 if (bm > SB_BMAP_NR(s))
331 if (end_bm > SB_BMAP_NR(s))
332 end_bm = SB_BMAP_NR(s);
334 /* When the bitmap is more than 10% free, anyone can allocate.
335 * When it's less than 10% free, only files that already use the
336 * bitmap are allowed. Once we pass 80% full, this restriction
339 * We do this so that files that grow later still have space close to
340 * their original allocation. This improves locality, and presumably
341 * performance as a result.
343 * This is only an allocation policy and does not make up for getting a
344 * bad hint. Decent hinting must be implemented for this to work well.
346 if (TEST_OPTION(skip_busy, s)
347 && SB_FREE_BLOCKS(s) > SB_BLOCK_COUNT(s) / 20) {
348 for (; bm < end_bm; bm++, off = 0) {
349 if ((off && (!unfm || (file_block != 0)))
350 || SB_AP_BITMAP(s)[bm].free_count >
351 (s->s_blocksize << 3) / 10)
353 scan_bitmap_block(th, bm, &off, off_max,
358 /* we know from above that start is a reasonable number */
359 get_bit_address(s, *start, &bm, &off);
362 for (; bm < end_bm; bm++, off = 0) {
364 scan_bitmap_block(th, bm, &off, off_max, min, max, unfm);
370 scan_bitmap_block(th, bm, &off, end_off + 1, min, max, unfm);
373 *start = bm * off_max + off;
378 static void _reiserfs_free_block(struct reiserfs_transaction_handle *th,
379 struct inode *inode, b_blocknr_t block,
382 struct super_block *s = th->t_super;
383 struct reiserfs_super_block *rs;
384 struct buffer_head *sbh, *bmbh;
385 struct reiserfs_bitmap_info *apbi;
388 BUG_ON(!th->t_trans_id);
390 PROC_INFO_INC(s, free_block);
392 rs = SB_DISK_SUPER_BLOCK(s);
393 sbh = SB_BUFFER_WITH_SB(s);
394 apbi = SB_AP_BITMAP(s);
396 get_bit_address(s, block, &nr, &offset);
398 if (nr >= sb_bmap_nr(rs)) {
399 reiserfs_warning(s, "vs-4075: reiserfs_free_block: "
400 "block %lu is out of range on %s",
401 block, reiserfs_bdevname(s));
405 bmbh = reiserfs_read_bitmap_block(s, nr);
409 reiserfs_prepare_for_journal(s, bmbh, 1);
411 /* clear bit for the given block in bit map */
412 if (!reiserfs_test_and_clear_le_bit(offset, bmbh->b_data)) {
413 reiserfs_warning(s, "vs-4080: reiserfs_free_block: "
414 "free_block (%s:%lu)[dev:blocknr]: bit already cleared",
415 reiserfs_bdevname(s), block);
417 apbi[nr].free_count++;
418 journal_mark_dirty(th, s, bmbh);
421 reiserfs_prepare_for_journal(s, sbh, 1);
422 /* update super block */
423 set_sb_free_blocks(rs, sb_free_blocks(rs) + 1);
425 journal_mark_dirty(th, s, sbh);
427 DQUOT_FREE_BLOCK_NODIRTY(inode, 1);
430 void reiserfs_free_block(struct reiserfs_transaction_handle *th,
431 struct inode *inode, b_blocknr_t block,
434 struct super_block *s = th->t_super;
435 BUG_ON(!th->t_trans_id);
437 RFALSE(!s, "vs-4061: trying to free block on nonexistent device");
438 if (!is_reusable(s, block, 1))
441 if (block > sb_block_count(REISERFS_SB(s)->s_rs)) {
442 reiserfs_panic(th->t_super, "bitmap-4072",
443 "Trying to free block outside file system "
444 "boundaries (%lu > %lu)",
445 block, sb_block_count(REISERFS_SB(s)->s_rs));
448 /* mark it before we clear it, just in case */
449 journal_mark_freed(th, s, block);
450 _reiserfs_free_block(th, inode, block, for_unformatted);
453 /* preallocated blocks don't need to be run through journal_mark_freed */
454 static void reiserfs_free_prealloc_block(struct reiserfs_transaction_handle *th,
455 struct inode *inode, b_blocknr_t block)
457 BUG_ON(!th->t_trans_id);
459 "vs-4060: trying to free block on nonexistent device");
460 if (!is_reusable(th->t_super, block, 1))
462 _reiserfs_free_block(th, inode, block, 1);
465 static void __discard_prealloc(struct reiserfs_transaction_handle *th,
466 struct reiserfs_inode_info *ei)
468 unsigned long save = ei->i_prealloc_block;
470 struct inode *inode = &ei->vfs_inode;
471 BUG_ON(!th->t_trans_id);
472 #ifdef CONFIG_REISERFS_CHECK
473 if (ei->i_prealloc_count < 0)
474 reiserfs_warning(th->t_super,
475 "zam-4001:%s: inode has negative prealloc blocks count.",
478 while (ei->i_prealloc_count > 0) {
479 reiserfs_free_prealloc_block(th, inode, ei->i_prealloc_block);
480 ei->i_prealloc_block++;
481 ei->i_prealloc_count--;
485 reiserfs_update_sd(th, inode);
486 ei->i_prealloc_block = save;
487 list_del_init(&(ei->i_prealloc_list));
490 /* FIXME: It should be inline function */
491 void reiserfs_discard_prealloc(struct reiserfs_transaction_handle *th,
494 struct reiserfs_inode_info *ei = REISERFS_I(inode);
495 BUG_ON(!th->t_trans_id);
496 if (ei->i_prealloc_count)
497 __discard_prealloc(th, ei);
500 void reiserfs_discard_all_prealloc(struct reiserfs_transaction_handle *th)
502 struct list_head *plist = &SB_JOURNAL(th->t_super)->j_prealloc_list;
504 BUG_ON(!th->t_trans_id);
506 while (!list_empty(plist)) {
507 struct reiserfs_inode_info *ei;
508 ei = list_entry(plist->next, struct reiserfs_inode_info,
510 #ifdef CONFIG_REISERFS_CHECK
511 if (!ei->i_prealloc_count) {
512 reiserfs_warning(th->t_super,
513 "zam-4001:%s: inode is in prealloc list but has no preallocated blocks.",
517 __discard_prealloc(th, ei);
521 void reiserfs_init_alloc_options(struct super_block *s)
523 set_bit(_ALLOC_skip_busy, &SB_ALLOC_OPTS(s));
524 set_bit(_ALLOC_dirid_groups, &SB_ALLOC_OPTS(s));
525 set_bit(_ALLOC_packing_groups, &SB_ALLOC_OPTS(s));
528 /* block allocator related options are parsed here */
529 int reiserfs_parse_alloc_options(struct super_block *s, char *options)
531 char *this_char, *value;
533 REISERFS_SB(s)->s_alloc_options.bits = 0; /* clear default settings */
535 while ((this_char = strsep(&options, ":")) != NULL) {
536 if ((value = strchr(this_char, '=')) != NULL)
539 if (!strcmp(this_char, "concentrating_formatted_nodes")) {
541 SET_OPTION(concentrating_formatted_nodes);
543 && *value) ? simple_strtoul(value, &value,
545 if (temp <= 0 || temp > 100) {
546 REISERFS_SB(s)->s_alloc_options.border = 10;
548 REISERFS_SB(s)->s_alloc_options.border =
553 if (!strcmp(this_char, "displacing_large_files")) {
554 SET_OPTION(displacing_large_files);
555 REISERFS_SB(s)->s_alloc_options.large_file_size =
557 && *value) ? simple_strtoul(value, &value, 0) : 16;
560 if (!strcmp(this_char, "displacing_new_packing_localities")) {
561 SET_OPTION(displacing_new_packing_localities);
565 if (!strcmp(this_char, "old_hashed_relocation")) {
566 SET_OPTION(old_hashed_relocation);
570 if (!strcmp(this_char, "new_hashed_relocation")) {
571 SET_OPTION(new_hashed_relocation);
575 if (!strcmp(this_char, "dirid_groups")) {
576 SET_OPTION(dirid_groups);
579 if (!strcmp(this_char, "oid_groups")) {
580 SET_OPTION(oid_groups);
583 if (!strcmp(this_char, "packing_groups")) {
584 SET_OPTION(packing_groups);
587 if (!strcmp(this_char, "hashed_formatted_nodes")) {
588 SET_OPTION(hashed_formatted_nodes);
592 if (!strcmp(this_char, "skip_busy")) {
593 SET_OPTION(skip_busy);
597 if (!strcmp(this_char, "hundredth_slices")) {
598 SET_OPTION(hundredth_slices);
602 if (!strcmp(this_char, "old_way")) {
607 if (!strcmp(this_char, "displace_based_on_dirid")) {
608 SET_OPTION(displace_based_on_dirid);
612 if (!strcmp(this_char, "preallocmin")) {
613 REISERFS_SB(s)->s_alloc_options.preallocmin =
615 && *value) ? simple_strtoul(value, &value, 0) : 4;
619 if (!strcmp(this_char, "preallocsize")) {
620 REISERFS_SB(s)->s_alloc_options.preallocsize =
622 && *value) ? simple_strtoul(value, &value,
628 reiserfs_warning(s, "zam-4001: %s : unknown option - %s",
629 __FUNCTION__, this_char);
633 reiserfs_warning(s, "allocator options = [%08x]\n", SB_ALLOC_OPTS(s));
637 static inline void new_hashed_relocation(reiserfs_blocknr_hint_t * hint)
640 if (hint->formatted_node) {
641 hash_in = (char *)&hint->key.k_dir_id;
644 //hint->search_start = hint->beg;
645 hash_in = (char *)&hint->key.k_dir_id;
647 if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
648 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_dir_id);
651 (char *)(&INODE_PKEY(hint->inode)->k_objectid);
655 hint->beg + keyed_hash(hash_in, 4) % (hint->end - hint->beg);
659 * Relocation based on dirid, hashing them into a given bitmap block
660 * files. Formatted nodes are unaffected, a seperate policy covers them
662 static void dirid_groups(reiserfs_blocknr_hint_t * hint)
667 struct super_block *sb = hint->th->t_super;
669 dirid = le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id);
670 else if (hint->formatted_node)
671 dirid = hint->key.k_dir_id;
674 bm = bmap_hash_id(sb, dirid);
675 hash = bm * (sb->s_blocksize << 3);
676 /* give a portion of the block group to metadata */
678 hash += sb->s_blocksize / 2;
679 hint->search_start = hash;
684 * Relocation based on oid, hashing them into a given bitmap block
685 * files. Formatted nodes are unaffected, a seperate policy covers them
687 static void oid_groups(reiserfs_blocknr_hint_t * hint)
695 dirid = le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id);
697 /* keep the root dir and it's first set of subdirs close to
698 * the start of the disk
701 hash = (hint->inode->i_sb->s_blocksize << 3);
703 oid = le32_to_cpu(INODE_PKEY(hint->inode)->k_objectid);
704 bm = bmap_hash_id(hint->inode->i_sb, oid);
705 hash = bm * (hint->inode->i_sb->s_blocksize << 3);
707 hint->search_start = hash;
711 /* returns 1 if it finds an indirect item and gets valid hint info
712 * from it, otherwise 0
714 static int get_left_neighbor(reiserfs_blocknr_hint_t * hint)
716 struct treepath *path;
717 struct buffer_head *bh;
718 struct item_head *ih;
723 if (!hint->path) /* reiserfs code can call this function w/o pointer to path
724 * structure supplied; then we rely on supplied search_start */
728 bh = get_last_bh(path);
729 RFALSE(!bh, "green-4002: Illegal path specified to get_left_neighbor");
731 pos_in_item = path->pos_in_item;
732 item = get_item(path);
734 hint->search_start = bh->b_blocknr;
736 if (!hint->formatted_node && is_indirect_le_ih(ih)) {
737 /* for indirect item: go to left and look for the first non-hole entry
738 in the indirect item */
739 if (pos_in_item == I_UNFM_NUM(ih))
741 // pos_in_item = I_UNFM_NUM (ih) - 1;
742 while (pos_in_item >= 0) {
743 int t = get_block_num(item, pos_in_item);
745 hint->search_start = t;
753 /* does result value fit into specified region? */
757 /* should be, if formatted node, then try to put on first part of the device
758 specified as number of percent with mount option device, else try to put
759 on last of device. This is not to say it is good code to do so,
760 but the effect should be measured. */
761 static inline void set_border_in_hint(struct super_block *s,
762 reiserfs_blocknr_hint_t * hint)
765 SB_BLOCK_COUNT(s) / REISERFS_SB(s)->s_alloc_options.border;
767 if (hint->formatted_node)
768 hint->end = border - 1;
773 static inline void displace_large_file(reiserfs_blocknr_hint_t * hint)
775 if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
778 keyed_hash((char *)(&INODE_PKEY(hint->inode)->k_dir_id),
779 4) % (hint->end - hint->beg);
783 keyed_hash((char *)(&INODE_PKEY(hint->inode)->k_objectid),
784 4) % (hint->end - hint->beg);
787 static inline void hash_formatted_node(reiserfs_blocknr_hint_t * hint)
792 hash_in = (char *)&hint->key.k_dir_id;
793 else if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super))
794 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_dir_id);
796 hash_in = (char *)(&INODE_PKEY(hint->inode)->k_objectid);
799 hint->beg + keyed_hash(hash_in, 4) % (hint->end - hint->beg);
803 this_blocknr_allocation_would_make_it_a_large_file(reiserfs_blocknr_hint_t *
806 return hint->block ==
807 REISERFS_SB(hint->th->t_super)->s_alloc_options.large_file_size;
810 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
811 static inline void displace_new_packing_locality(reiserfs_blocknr_hint_t * hint)
813 struct in_core_key *key = &hint->key;
815 hint->th->displace_new_blocks = 0;
817 hint->beg + keyed_hash((char *)(&key->k_objectid),
818 4) % (hint->end - hint->beg);
822 static inline int old_hashed_relocation(reiserfs_blocknr_hint_t * hint)
827 if (hint->formatted_node || hint->inode == NULL) {
831 hash_in = le32_to_cpu((INODE_PKEY(hint->inode))->k_dir_id);
833 hint->beg + (u32) keyed_hash(((char *)(&hash_in)),
834 4) % (hint->end - hint->beg - 1);
835 if (border > hint->search_start)
836 hint->search_start = border;
841 static inline int old_way(reiserfs_blocknr_hint_t * hint)
845 if (hint->formatted_node || hint->inode == NULL) {
851 le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id) % (hint->end -
853 if (border > hint->search_start)
854 hint->search_start = border;
859 static inline void hundredth_slices(reiserfs_blocknr_hint_t * hint)
861 struct in_core_key *key = &hint->key;
862 b_blocknr_t slice_start;
865 (keyed_hash((char *)(&key->k_dir_id), 4) % 100) * (hint->end / 100);
866 if (slice_start > hint->search_start
867 || slice_start + (hint->end / 100) <= hint->search_start) {
868 hint->search_start = slice_start;
872 static void determine_search_start(reiserfs_blocknr_hint_t * hint,
875 struct super_block *s = hint->th->t_super;
879 hint->end = SB_BLOCK_COUNT(s) - 1;
881 /* This is former border algorithm. Now with tunable border offset */
882 if (concentrating_formatted_nodes(s))
883 set_border_in_hint(s, hint);
885 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
886 /* whenever we create a new directory, we displace it. At first we will
887 hash for location, later we might look for a moderately empty place for
889 if (displacing_new_packing_localities(s)
890 && hint->th->displace_new_blocks) {
891 displace_new_packing_locality(hint);
893 /* we do not continue determine_search_start,
894 * if new packing locality is being displaced */
899 /* all persons should feel encouraged to add more special cases here and
902 if (displacing_large_files(s) && !hint->formatted_node
903 && this_blocknr_allocation_would_make_it_a_large_file(hint)) {
904 displace_large_file(hint);
908 /* if none of our special cases is relevant, use the left neighbor in the
909 tree order of the new node we are allocating for */
910 if (hint->formatted_node && TEST_OPTION(hashed_formatted_nodes, s)) {
911 hash_formatted_node(hint);
915 unfm_hint = get_left_neighbor(hint);
917 /* Mimic old block allocator behaviour, that is if VFS allowed for preallocation,
918 new blocks are displaced based on directory ID. Also, if suggested search_start
919 is less than last preallocated block, we start searching from it, assuming that
920 HDD dataflow is faster in forward direction */
921 if (TEST_OPTION(old_way, s)) {
922 if (!hint->formatted_node) {
923 if (!reiserfs_hashed_relocation(s))
925 else if (!reiserfs_no_unhashed_relocation(s))
926 old_hashed_relocation(hint);
929 && hint->search_start <
930 REISERFS_I(hint->inode)->i_prealloc_block)
932 REISERFS_I(hint->inode)->i_prealloc_block;
937 /* This is an approach proposed by Hans */
938 if (TEST_OPTION(hundredth_slices, s)
939 && !(displacing_large_files(s) && !hint->formatted_node)) {
940 hundredth_slices(hint);
944 /* old_hashed_relocation only works on unformatted */
945 if (!unfm_hint && !hint->formatted_node &&
946 TEST_OPTION(old_hashed_relocation, s)) {
947 old_hashed_relocation(hint);
949 /* new_hashed_relocation works with both formatted/unformatted nodes */
950 if ((!unfm_hint || hint->formatted_node) &&
951 TEST_OPTION(new_hashed_relocation, s)) {
952 new_hashed_relocation(hint);
954 /* dirid grouping works only on unformatted nodes */
955 if (!unfm_hint && !hint->formatted_node && TEST_OPTION(dirid_groups, s)) {
958 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
959 if (hint->formatted_node && TEST_OPTION(dirid_groups, s)) {
964 /* oid grouping works only on unformatted nodes */
965 if (!unfm_hint && !hint->formatted_node && TEST_OPTION(oid_groups, s)) {
971 static int determine_prealloc_size(reiserfs_blocknr_hint_t * hint)
973 /* make minimum size a mount option and benchmark both ways */
974 /* we preallocate blocks only for regular files, specific size */
975 /* benchmark preallocating always and see what happens */
977 hint->prealloc_size = 0;
979 if (!hint->formatted_node && hint->preallocate) {
980 if (S_ISREG(hint->inode->i_mode)
981 && hint->inode->i_size >=
982 REISERFS_SB(hint->th->t_super)->s_alloc_options.
983 preallocmin * hint->inode->i_sb->s_blocksize)
984 hint->prealloc_size =
985 REISERFS_SB(hint->th->t_super)->s_alloc_options.
991 /* XXX I know it could be merged with upper-level function;
992 but may be result function would be too complex. */
993 static inline int allocate_without_wrapping_disk(reiserfs_blocknr_hint_t * hint,
994 b_blocknr_t * new_blocknrs,
996 b_blocknr_t finish, int min,
1000 int rest = amount_needed;
1003 while (rest > 0 && start <= finish) {
1004 nr_allocated = scan_bitmap(hint->th, &start, finish, min,
1005 rest + prealloc_size,
1006 !hint->formatted_node, hint->block);
1008 if (nr_allocated == 0) /* no new blocks allocated, return */
1011 /* fill free_blocknrs array first */
1012 while (rest > 0 && nr_allocated > 0) {
1013 *new_blocknrs++ = start++;
1018 /* do we have something to fill prealloc. array also ? */
1019 if (nr_allocated > 0) {
1020 /* it means prealloc_size was greater that 0 and we do preallocation */
1021 list_add(&REISERFS_I(hint->inode)->i_prealloc_list,
1022 &SB_JOURNAL(hint->th->t_super)->
1024 REISERFS_I(hint->inode)->i_prealloc_block = start;
1025 REISERFS_I(hint->inode)->i_prealloc_count =
1031 return (amount_needed - rest);
1034 static inline int blocknrs_and_prealloc_arrays_from_search_start
1035 (reiserfs_blocknr_hint_t * hint, b_blocknr_t * new_blocknrs,
1036 int amount_needed) {
1037 struct super_block *s = hint->th->t_super;
1038 b_blocknr_t start = hint->search_start;
1039 b_blocknr_t finish = SB_BLOCK_COUNT(s) - 1;
1041 int nr_allocated = 0;
1043 determine_prealloc_size(hint);
1044 if (!hint->formatted_node) {
1046 #ifdef REISERQUOTA_DEBUG
1047 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1048 "reiserquota: allocating %d blocks id=%u",
1049 amount_needed, hint->inode->i_uid);
1052 DQUOT_ALLOC_BLOCK_NODIRTY(hint->inode, amount_needed);
1053 if (quota_ret) /* Quota exceeded? */
1054 return QUOTA_EXCEEDED;
1055 if (hint->preallocate && hint->prealloc_size) {
1056 #ifdef REISERQUOTA_DEBUG
1057 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1058 "reiserquota: allocating (prealloc) %d blocks id=%u",
1059 hint->prealloc_size, hint->inode->i_uid);
1062 DQUOT_PREALLOC_BLOCK_NODIRTY(hint->inode,
1063 hint->prealloc_size);
1065 hint->preallocate = hint->prealloc_size = 0;
1067 /* for unformatted nodes, force large allocations */
1072 case 0: /* Search from hint->search_start to end of disk */
1073 start = hint->search_start;
1074 finish = SB_BLOCK_COUNT(s) - 1;
1076 case 1: /* Search from hint->beg to hint->search_start */
1078 finish = hint->search_start;
1080 case 2: /* Last chance: Search from 0 to hint->beg */
1084 default: /* We've tried searching everywhere, not enough space */
1085 /* Free the blocks */
1086 if (!hint->formatted_node) {
1087 #ifdef REISERQUOTA_DEBUG
1088 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1089 "reiserquota: freeing (nospace) %d blocks id=%u",
1091 hint->prealloc_size -
1093 hint->inode->i_uid);
1095 DQUOT_FREE_BLOCK_NODIRTY(hint->inode, amount_needed + hint->prealloc_size - nr_allocated); /* Free not allocated blocks */
1097 while (nr_allocated--)
1098 reiserfs_free_block(hint->th, hint->inode,
1099 new_blocknrs[nr_allocated],
1100 !hint->formatted_node);
1102 return NO_DISK_SPACE;
1104 } while ((nr_allocated += allocate_without_wrapping_disk(hint,
1114 if (!hint->formatted_node &&
1115 amount_needed + hint->prealloc_size >
1116 nr_allocated + REISERFS_I(hint->inode)->i_prealloc_count) {
1117 /* Some of preallocation blocks were not allocated */
1118 #ifdef REISERQUOTA_DEBUG
1119 reiserfs_debug(s, REISERFS_DEBUG_CODE,
1120 "reiserquota: freeing (failed prealloc) %d blocks id=%u",
1121 amount_needed + hint->prealloc_size -
1123 REISERFS_I(hint->inode)->i_prealloc_count,
1124 hint->inode->i_uid);
1126 DQUOT_FREE_BLOCK_NODIRTY(hint->inode, amount_needed +
1127 hint->prealloc_size - nr_allocated -
1128 REISERFS_I(hint->inode)->
1135 /* grab new blocknrs from preallocated list */
1136 /* return amount still needed after using them */
1137 static int use_preallocated_list_if_available(reiserfs_blocknr_hint_t * hint,
1138 b_blocknr_t * new_blocknrs,
1141 struct inode *inode = hint->inode;
1143 if (REISERFS_I(inode)->i_prealloc_count > 0) {
1144 while (amount_needed) {
1146 *new_blocknrs++ = REISERFS_I(inode)->i_prealloc_block++;
1147 REISERFS_I(inode)->i_prealloc_count--;
1151 if (REISERFS_I(inode)->i_prealloc_count <= 0) {
1152 list_del(&REISERFS_I(inode)->i_prealloc_list);
1157 /* return amount still needed after using preallocated blocks */
1158 return amount_needed;
1161 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
1162 already reserved */ )
1164 int initial_amount_needed = amount_needed;
1166 struct super_block *s = hint->th->t_super;
1168 /* Check if there is enough space, taking into account reserved space */
1169 if (SB_FREE_BLOCKS(s) - REISERFS_SB(s)->reserved_blocks <
1170 amount_needed - reserved_by_us)
1171 return NO_DISK_SPACE;
1172 /* should this be if !hint->inode && hint->preallocate? */
1173 /* do you mean hint->formatted_node can be removed ? - Zam */
1174 /* hint->formatted_node cannot be removed because we try to access
1175 inode information here, and there is often no inode assotiated with
1176 metadata allocations - green */
1178 if (!hint->formatted_node && hint->preallocate) {
1179 amount_needed = use_preallocated_list_if_available
1180 (hint, new_blocknrs, amount_needed);
1181 if (amount_needed == 0) /* all blocknrs we need we got from
1184 new_blocknrs += (initial_amount_needed - amount_needed);
1187 /* find search start and save it in hint structure */
1188 determine_search_start(hint, amount_needed);
1189 if (hint->search_start >= SB_BLOCK_COUNT(s))
1190 hint->search_start = SB_BLOCK_COUNT(s) - 1;
1192 /* allocation itself; fill new_blocknrs and preallocation arrays */
1193 ret = blocknrs_and_prealloc_arrays_from_search_start
1194 (hint, new_blocknrs, amount_needed);
1196 /* we used prealloc. list to fill (partially) new_blocknrs array. If final allocation fails we
1197 * need to return blocks back to prealloc. list or just free them. -- Zam (I chose second
1200 if (ret != CARRY_ON) {
1201 while (amount_needed++ < initial_amount_needed) {
1202 reiserfs_free_block(hint->th, hint->inode,
1203 *(--new_blocknrs), 1);
1209 void reiserfs_cache_bitmap_metadata(struct super_block *sb,
1210 struct buffer_head *bh,
1211 struct reiserfs_bitmap_info *info)
1213 unsigned long *cur = (unsigned long *)(bh->b_data + bh->b_size);
1215 info->first_zero_hint = 1 << (sb->s_blocksize_bits + 3);
1217 while (--cur >= (unsigned long *)bh->b_data) {
1218 int base = ((char *)cur - bh->b_data) << 3;
1220 /* 0 and ~0 are special, we can optimize for them */
1222 info->first_zero_hint = base;
1223 info->free_count += BITS_PER_LONG;
1224 } else if (*cur != ~0L) { /* A mix, investigate */
1226 for (b = BITS_PER_LONG - 1; b >= 0; b--) {
1227 if (!reiserfs_test_le_bit(b, cur)) {
1228 info->first_zero_hint = base + b;
1234 /* The first bit must ALWAYS be 1 */
1235 BUG_ON(info->first_zero_hint == 0);
1238 struct buffer_head *reiserfs_read_bitmap_block(struct super_block *sb,
1239 unsigned int bitmap)
1241 b_blocknr_t block = (sb->s_blocksize << 3) * bitmap;
1242 struct reiserfs_bitmap_info *info = SB_AP_BITMAP(sb) + bitmap;
1243 struct buffer_head *bh;
1245 /* Way old format filesystems had the bitmaps packed up front.
1246 * I doubt there are any of these left, but just in case... */
1247 if (unlikely(test_bit(REISERFS_OLD_FORMAT,
1248 &(REISERFS_SB(sb)->s_properties))))
1249 block = REISERFS_SB(sb)->s_sbh->b_blocknr + 1 + bitmap;
1250 else if (bitmap == 0)
1251 block = (REISERFS_DISK_OFFSET_IN_BYTES >> sb->s_blocksize_bits) + 1;
1253 bh = sb_bread(sb, block);
1255 reiserfs_warning(sb, "sh-2029: %s: bitmap block (#%u) "
1256 "reading failed", __FUNCTION__, block);
1258 if (buffer_locked(bh)) {
1259 PROC_INFO_INC(sb, scan_bitmap.wait);
1260 __wait_on_buffer(bh);
1262 BUG_ON(!buffer_uptodate(bh));
1263 BUG_ON(atomic_read(&bh->b_count) == 0);
1265 if (info->first_zero_hint == 0)
1266 reiserfs_cache_bitmap_metadata(sb, bh, info);
1272 int reiserfs_init_bitmap_cache(struct super_block *sb)
1274 struct reiserfs_bitmap_info *bitmap;
1276 bitmap = vmalloc(sizeof (*bitmap) * SB_BMAP_NR(sb));
1280 memset(bitmap, 0, sizeof (*bitmap) * SB_BMAP_NR(sb));
1282 SB_AP_BITMAP(sb) = bitmap;
1287 void reiserfs_free_bitmap_cache(struct super_block *sb)
1289 if (SB_AP_BITMAP(sb)) {
1290 vfree(SB_AP_BITMAP(sb));
1291 SB_AP_BITMAP(sb) = NULL;