2 * linux/fs/ext2/balloc.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
9 * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
10 * Big-endian to little-endian byte-swapping/bitmaps by
11 * David S. Miller (davem@caip.rutgers.edu), 1995
15 #include <linux/quotaops.h>
16 #include <linux/sched.h>
17 #include <linux/buffer_head.h>
18 #include <linux/capability.h>
21 * balloc.c contains the blocks allocation and deallocation routines
25 * The free blocks are managed by bitmaps. A file system contains several
26 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
27 * block for inodes, N blocks for the inode table and data blocks.
29 * The file system contains group descriptors which are located after the
30 * super block. Each descriptor contains the number of the bitmap block and
31 * the free blocks count in the block. The descriptors are loaded in memory
32 * when a file system is mounted (see ext2_fill_super).
36 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
38 struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
39 unsigned int block_group,
40 struct buffer_head ** bh)
42 unsigned long group_desc;
44 struct ext2_group_desc * desc;
45 struct ext2_sb_info *sbi = EXT2_SB(sb);
47 if (block_group >= sbi->s_groups_count) {
48 ext2_error (sb, "ext2_get_group_desc",
49 "block_group >= groups_count - "
50 "block_group = %d, groups_count = %lu",
51 block_group, sbi->s_groups_count);
56 group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
57 offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
58 if (!sbi->s_group_desc[group_desc]) {
59 ext2_error (sb, "ext2_get_group_desc",
60 "Group descriptor not loaded - "
61 "block_group = %d, group_desc = %lu, desc = %lu",
62 block_group, group_desc, offset);
66 desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
68 *bh = sbi->s_group_desc[group_desc];
73 block_in_use(unsigned long block, struct super_block *sb, unsigned char *map)
75 return ext2_test_bit ((block -
76 le32_to_cpu(EXT2_SB(sb)->s_es->s_first_data_block)) %
77 EXT2_BLOCKS_PER_GROUP(sb), map);
81 * Read the bitmap for a given block_group, reading into the specified
82 * slot in the superblock's bitmap cache.
84 * Return buffer_head on success or NULL in case of failure.
86 static struct buffer_head *
87 read_block_bitmap(struct super_block *sb, unsigned int block_group)
90 struct ext2_group_desc * desc;
91 struct buffer_head * bh = NULL;
92 unsigned int bitmap_blk;
94 desc = ext2_get_group_desc (sb, block_group, NULL);
97 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
98 bh = sb_bread(sb, bitmap_blk);
100 ext2_error (sb, __FUNCTION__,
101 "Cannot read block bitmap - "
102 "block_group = %d, block_bitmap = %u",
103 block_group, le32_to_cpu(desc->bg_block_bitmap));
105 /* check whether block bitmap block number is set */
106 if (!block_in_use(bitmap_blk, sb, bh->b_data)) {
107 /* bad block bitmap */
110 /* check whether the inode bitmap block number is set */
111 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
112 if (!block_in_use(bitmap_blk, sb, bh->b_data)) {
113 /* bad block bitmap */
116 /* check whether the inode table block number is set */
117 bitmap_blk = le32_to_cpu(desc->bg_inode_table);
118 for (i = 0; i < EXT2_SB(sb)->s_itb_per_group; i++, bitmap_blk++) {
119 if (!block_in_use(bitmap_blk, sb, bh->b_data)) {
120 /* bad block bitmap */
129 ext2_error(sb, __FUNCTION__,
130 "Invalid block bitmap - "
131 "block_group = %d, block = %u",
132 block_group, bitmap_blk);
136 static void release_blocks(struct super_block *sb, int count)
139 struct ext2_sb_info *sbi = EXT2_SB(sb);
141 percpu_counter_add(&sbi->s_freeblocks_counter, count);
146 static void group_adjust_blocks(struct super_block *sb, int group_no,
147 struct ext2_group_desc *desc, struct buffer_head *bh, int count)
150 struct ext2_sb_info *sbi = EXT2_SB(sb);
151 unsigned free_blocks;
153 spin_lock(sb_bgl_lock(sbi, group_no));
154 free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
155 desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
156 spin_unlock(sb_bgl_lock(sbi, group_no));
158 mark_buffer_dirty(bh);
163 * The reservation window structure operations
164 * --------------------------------------------
165 * Operations include:
166 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
168 * We use a red-black tree to represent per-filesystem reservation
174 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
175 * @rb_root: root of per-filesystem reservation rb tree
176 * @verbose: verbose mode
177 * @fn: function which wishes to dump the reservation map
179 * If verbose is turned on, it will print the whole block reservation
180 * windows(start, end). Otherwise, it will only print out the "bad" windows,
181 * those windows that overlap with their immediate neighbors.
184 static void __rsv_window_dump(struct rb_root *root, int verbose,
188 struct ext2_reserve_window_node *rsv, *prev;
196 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
198 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
200 printk("reservation window 0x%p "
201 "start: %lu, end: %lu\n",
202 rsv, rsv->rsv_start, rsv->rsv_end);
203 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
204 printk("Bad reservation %p (start >= end)\n",
208 if (prev && prev->rsv_end >= rsv->rsv_start) {
209 printk("Bad reservation %p (prev->end >= start)\n",
215 printk("Restarting reservation walk in verbose mode\n");
223 printk("Window map complete.\n");
227 #define rsv_window_dump(root, verbose) \
228 __rsv_window_dump((root), (verbose), __FUNCTION__)
230 #define rsv_window_dump(root, verbose) do {} while (0)
234 * goal_in_my_reservation()
235 * @rsv: inode's reservation window
236 * @grp_goal: given goal block relative to the allocation block group
237 * @group: the current allocation block group
238 * @sb: filesystem super block
240 * Test if the given goal block (group relative) is within the file's
241 * own block reservation window range.
243 * If the reservation window is outside the goal allocation group, return 0;
244 * grp_goal (given goal block) could be -1, which means no specific
245 * goal block. In this case, always return 1.
246 * If the goal block is within the reservation window, return 1;
247 * otherwise, return 0;
250 goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
251 unsigned int group, struct super_block * sb)
253 ext2_fsblk_t group_first_block, group_last_block;
255 group_first_block = ext2_group_first_block_no(sb, group);
256 group_last_block = group_first_block + EXT2_BLOCKS_PER_GROUP(sb) - 1;
258 if ((rsv->_rsv_start > group_last_block) ||
259 (rsv->_rsv_end < group_first_block))
261 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
262 || (grp_goal + group_first_block > rsv->_rsv_end)))
268 * search_reserve_window()
269 * @rb_root: root of reservation tree
270 * @goal: target allocation block
272 * Find the reserved window which includes the goal, or the previous one
273 * if the goal is not in any window.
274 * Returns NULL if there are no windows or if all windows start after the goal.
276 static struct ext2_reserve_window_node *
277 search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
279 struct rb_node *n = root->rb_node;
280 struct ext2_reserve_window_node *rsv;
286 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
288 if (goal < rsv->rsv_start)
290 else if (goal > rsv->rsv_end)
296 * We've fallen off the end of the tree: the goal wasn't inside
297 * any particular node. OK, the previous node must be to one
298 * side of the interval containing the goal. If it's the RHS,
299 * we need to back up one.
301 if (rsv->rsv_start > goal) {
302 n = rb_prev(&rsv->rsv_node);
303 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
309 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
311 * @rsv: reservation window to add
313 * Must be called with rsv_lock held.
315 void ext2_rsv_window_add(struct super_block *sb,
316 struct ext2_reserve_window_node *rsv)
318 struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
319 struct rb_node *node = &rsv->rsv_node;
320 ext2_fsblk_t start = rsv->rsv_start;
322 struct rb_node ** p = &root->rb_node;
323 struct rb_node * parent = NULL;
324 struct ext2_reserve_window_node *this;
329 this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
331 if (start < this->rsv_start)
333 else if (start > this->rsv_end)
336 rsv_window_dump(root, 1);
341 rb_link_node(node, parent, p);
342 rb_insert_color(node, root);
346 * rsv_window_remove() -- unlink a window from the reservation rb tree
348 * @rsv: reservation window to remove
350 * Mark the block reservation window as not allocated, and unlink it
351 * from the filesystem reservation window rb tree. Must be called with
354 static void rsv_window_remove(struct super_block *sb,
355 struct ext2_reserve_window_node *rsv)
357 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
358 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
359 rsv->rsv_alloc_hit = 0;
360 rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
364 * rsv_is_empty() -- Check if the reservation window is allocated.
365 * @rsv: given reservation window to check
367 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
369 static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
371 /* a valid reservation end block could not be 0 */
372 return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
376 * ext2_init_block_alloc_info()
377 * @inode: file inode structure
379 * Allocate and initialize the reservation window structure, and
380 * link the window to the ext2 inode structure at last
382 * The reservation window structure is only dynamically allocated
383 * and linked to ext2 inode the first time the open file
384 * needs a new block. So, before every ext2_new_block(s) call, for
385 * regular files, we should check whether the reservation window
386 * structure exists or not. In the latter case, this function is called.
387 * Fail to do so will result in block reservation being turned off for that
390 * This function is called from ext2_get_blocks_handle(), also called
391 * when setting the reservation window size through ioctl before the file
392 * is open for write (needs block allocation).
394 * Needs truncate_mutex protection prior to calling this function.
396 void ext2_init_block_alloc_info(struct inode *inode)
398 struct ext2_inode_info *ei = EXT2_I(inode);
399 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
400 struct super_block *sb = inode->i_sb;
402 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
404 struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
406 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
407 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
410 * if filesystem is mounted with NORESERVATION, the goal
411 * reservation window size is set to zero to indicate
412 * block reservation is off
414 if (!test_opt(sb, RESERVATION))
415 rsv->rsv_goal_size = 0;
417 rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
418 rsv->rsv_alloc_hit = 0;
419 block_i->last_alloc_logical_block = 0;
420 block_i->last_alloc_physical_block = 0;
422 ei->i_block_alloc_info = block_i;
426 * ext2_discard_reservation()
429 * Discard(free) block reservation window on last file close, or truncate
432 * It is being called in three cases:
433 * ext2_release_file(): last writer closes the file
434 * ext2_clear_inode(): last iput(), when nobody links to this file.
435 * ext2_truncate(): when the block indirect map is about to change.
437 void ext2_discard_reservation(struct inode *inode)
439 struct ext2_inode_info *ei = EXT2_I(inode);
440 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
441 struct ext2_reserve_window_node *rsv;
442 spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
447 rsv = &block_i->rsv_window_node;
448 if (!rsv_is_empty(&rsv->rsv_window)) {
450 if (!rsv_is_empty(&rsv->rsv_window))
451 rsv_window_remove(inode->i_sb, rsv);
452 spin_unlock(rsv_lock);
457 * ext2_free_blocks_sb() -- Free given blocks and update quota and i_blocks
459 * @block: start physcial block to free
460 * @count: number of blocks to free
462 void ext2_free_blocks (struct inode * inode, unsigned long block,
465 struct buffer_head *bitmap_bh = NULL;
466 struct buffer_head * bh2;
467 unsigned long block_group;
470 unsigned long overflow;
471 struct super_block * sb = inode->i_sb;
472 struct ext2_sb_info * sbi = EXT2_SB(sb);
473 struct ext2_group_desc * desc;
474 struct ext2_super_block * es = sbi->s_es;
475 unsigned freed = 0, group_freed;
477 if (block < le32_to_cpu(es->s_first_data_block) ||
478 block + count < block ||
479 block + count > le32_to_cpu(es->s_blocks_count)) {
480 ext2_error (sb, "ext2_free_blocks",
481 "Freeing blocks not in datazone - "
482 "block = %lu, count = %lu", block, count);
486 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
490 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
491 EXT2_BLOCKS_PER_GROUP(sb);
492 bit = (block - le32_to_cpu(es->s_first_data_block)) %
493 EXT2_BLOCKS_PER_GROUP(sb);
495 * Check to see if we are freeing blocks across a group
498 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
499 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
503 bitmap_bh = read_block_bitmap(sb, block_group);
507 desc = ext2_get_group_desc (sb, block_group, &bh2);
511 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
512 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
513 in_range (block, le32_to_cpu(desc->bg_inode_table),
514 sbi->s_itb_per_group) ||
515 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
516 sbi->s_itb_per_group))
517 ext2_error (sb, "ext2_free_blocks",
518 "Freeing blocks in system zones - "
519 "Block = %lu, count = %lu",
522 for (i = 0, group_freed = 0; i < count; i++) {
523 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
524 bit + i, bitmap_bh->b_data)) {
525 ext2_error(sb, __FUNCTION__,
526 "bit already cleared for block %lu", block + i);
532 mark_buffer_dirty(bitmap_bh);
533 if (sb->s_flags & MS_SYNCHRONOUS)
534 sync_dirty_buffer(bitmap_bh);
536 group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
537 freed += group_freed;
546 release_blocks(sb, freed);
547 DQUOT_FREE_BLOCK(inode, freed);
551 * bitmap_search_next_usable_block()
552 * @start: the starting block (group relative) of the search
553 * @bh: bufferhead contains the block group bitmap
554 * @maxblocks: the ending block (group relative) of the reservation
556 * The bitmap search --- search forward through the actual bitmap on disk until
557 * we find a bit free.
560 bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
561 ext2_grpblk_t maxblocks)
565 next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
566 if (next >= maxblocks)
572 * find_next_usable_block()
573 * @start: the starting block (group relative) to find next
574 * allocatable block in bitmap.
575 * @bh: bufferhead contains the block group bitmap
576 * @maxblocks: the ending block (group relative) for the search
578 * Find an allocatable block in a bitmap. We perform the "most
579 * appropriate allocation" algorithm of looking for a free block near
580 * the initial goal; then for a free byte somewhere in the bitmap;
581 * then for any free bit in the bitmap.
584 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
586 ext2_grpblk_t here, next;
591 * The goal was occupied; search forward for a free
592 * block within the next XX blocks.
594 * end_goal is more or less random, but it has to be
595 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
596 * next 64-bit boundary is simple..
598 ext2_grpblk_t end_goal = (start + 63) & ~63;
599 if (end_goal > maxblocks)
600 end_goal = maxblocks;
601 here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
604 ext2_debug("Bit not found near goal\n");
611 p = ((char *)bh->b_data) + (here >> 3);
612 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
613 next = (r - ((char *)bh->b_data)) << 3;
615 if (next < maxblocks && next >= here)
618 here = bitmap_search_next_usable_block(here, bh, maxblocks);
623 * ext2_try_to_allocate()
625 * @handle: handle to this transaction
626 * @group: given allocation block group
627 * @bitmap_bh: bufferhead holds the block bitmap
628 * @grp_goal: given target block within the group
629 * @count: target number of blocks to allocate
630 * @my_rsv: reservation window
632 * Attempt to allocate blocks within a give range. Set the range of allocation
633 * first, then find the first free bit(s) from the bitmap (within the range),
634 * and at last, allocate the blocks by claiming the found free bit as allocated.
636 * To set the range of this allocation:
637 * if there is a reservation window, only try to allocate block(s)
638 * from the file's own reservation window;
639 * Otherwise, the allocation range starts from the give goal block,
640 * ends at the block group's last block.
642 * If we failed to allocate the desired block then we may end up crossing to a
646 ext2_try_to_allocate(struct super_block *sb, int group,
647 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
648 unsigned long *count,
649 struct ext2_reserve_window *my_rsv)
651 ext2_fsblk_t group_first_block;
652 ext2_grpblk_t start, end;
653 unsigned long num = 0;
655 /* we do allocation within the reservation window if we have a window */
657 group_first_block = ext2_group_first_block_no(sb, group);
658 if (my_rsv->_rsv_start >= group_first_block)
659 start = my_rsv->_rsv_start - group_first_block;
661 /* reservation window cross group boundary */
663 end = my_rsv->_rsv_end - group_first_block + 1;
664 if (end > EXT2_BLOCKS_PER_GROUP(sb))
665 /* reservation window crosses group boundary */
666 end = EXT2_BLOCKS_PER_GROUP(sb);
667 if ((start <= grp_goal) && (grp_goal < end))
676 end = EXT2_BLOCKS_PER_GROUP(sb);
679 BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
683 grp_goal = find_next_usable_block(start, bitmap_bh, end);
689 for (i = 0; i < 7 && grp_goal > start &&
690 !ext2_test_bit(grp_goal - 1,
698 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), grp_goal,
699 bitmap_bh->b_data)) {
701 * The block was allocated by another thread, or it was
702 * allocated and then freed by another thread
712 while (num < *count && grp_goal < end
713 && !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
714 grp_goal, bitmap_bh->b_data)) {
719 return grp_goal - num;
726 * find_next_reservable_window():
727 * find a reservable space within the given range.
728 * It does not allocate the reservation window for now:
729 * alloc_new_reservation() will do the work later.
731 * @search_head: the head of the searching list;
732 * This is not necessarily the list head of the whole filesystem
734 * We have both head and start_block to assist the search
735 * for the reservable space. The list starts from head,
736 * but we will shift to the place where start_block is,
737 * then start from there, when looking for a reservable space.
739 * @size: the target new reservation window size
741 * @group_first_block: the first block we consider to start
742 * the real search from
745 * the maximum block number that our goal reservable space
746 * could start from. This is normally the last block in this
747 * group. The search will end when we found the start of next
748 * possible reservable space is out of this boundary.
749 * This could handle the cross boundary reservation window
752 * basically we search from the given range, rather than the whole
753 * reservation double linked list, (start_block, last_block)
754 * to find a free region that is of my size and has not
758 static int find_next_reservable_window(
759 struct ext2_reserve_window_node *search_head,
760 struct ext2_reserve_window_node *my_rsv,
761 struct super_block * sb,
762 ext2_fsblk_t start_block,
763 ext2_fsblk_t last_block)
765 struct rb_node *next;
766 struct ext2_reserve_window_node *rsv, *prev;
768 int size = my_rsv->rsv_goal_size;
770 /* TODO: make the start of the reservation window byte-aligned */
771 /* cur = *start_block & ~7;*/
778 if (cur <= rsv->rsv_end)
779 cur = rsv->rsv_end + 1;
782 * in the case we could not find a reservable space
783 * that is what is expected, during the re-search, we could
784 * remember what's the largest reservable space we could have
785 * and return that one.
787 * For now it will fail if we could not find the reservable
788 * space with expected-size (or more)...
790 if (cur > last_block)
791 return -1; /* fail */
794 next = rb_next(&rsv->rsv_node);
795 rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
798 * Reached the last reservation, we can just append to the
804 if (cur + size <= rsv->rsv_start) {
806 * Found a reserveable space big enough. We could
807 * have a reservation across the group boundary here
813 * we come here either :
814 * when we reach the end of the whole list,
815 * and there is empty reservable space after last entry in the list.
816 * append it to the end of the list.
818 * or we found one reservable space in the middle of the list,
819 * return the reservation window that we could append to.
823 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
824 rsv_window_remove(sb, my_rsv);
827 * Let's book the whole avaliable window for now. We will check the
828 * disk bitmap later and then, if there are free blocks then we adjust
829 * the window size if it's larger than requested.
830 * Otherwise, we will remove this node from the tree next time
831 * call find_next_reservable_window.
833 my_rsv->rsv_start = cur;
834 my_rsv->rsv_end = cur + size - 1;
835 my_rsv->rsv_alloc_hit = 0;
838 ext2_rsv_window_add(sb, my_rsv);
844 * alloc_new_reservation()--allocate a new reservation window
846 * To make a new reservation, we search part of the filesystem
847 * reservation list (the list that inside the group). We try to
848 * allocate a new reservation window near the allocation goal,
849 * or the beginning of the group, if there is no goal.
851 * We first find a reservable space after the goal, then from
852 * there, we check the bitmap for the first free block after
853 * it. If there is no free block until the end of group, then the
854 * whole group is full, we failed. Otherwise, check if the free
855 * block is inside the expected reservable space, if so, we
857 * If the first free block is outside the reservable space, then
858 * start from the first free block, we search for next available
861 * on succeed, a new reservation will be found and inserted into the list
862 * It contains at least one free block, and it does not overlap with other
863 * reservation windows.
865 * failed: we failed to find a reservation window in this group
867 * @rsv: the reservation
869 * @grp_goal: The goal (group-relative). It is where the search for a
870 * free reservable space should start from.
871 * if we have a goal(goal >0 ), then start from there,
872 * no goal(goal = -1), we start from the first block
875 * @sb: the super block
876 * @group: the group we are trying to allocate in
877 * @bitmap_bh: the block group block bitmap
880 static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
881 ext2_grpblk_t grp_goal, struct super_block *sb,
882 unsigned int group, struct buffer_head *bitmap_bh)
884 struct ext2_reserve_window_node *search_head;
885 ext2_fsblk_t group_first_block, group_end_block, start_block;
886 ext2_grpblk_t first_free_block;
887 struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
890 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
892 group_first_block = ext2_group_first_block_no(sb, group);
893 group_end_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
896 start_block = group_first_block;
898 start_block = grp_goal + group_first_block;
900 size = my_rsv->rsv_goal_size;
902 if (!rsv_is_empty(&my_rsv->rsv_window)) {
904 * if the old reservation is cross group boundary
905 * and if the goal is inside the old reservation window,
906 * we will come here when we just failed to allocate from
907 * the first part of the window. We still have another part
908 * that belongs to the next group. In this case, there is no
909 * point to discard our window and try to allocate a new one
910 * in this group(which will fail). we should
911 * keep the reservation window, just simply move on.
913 * Maybe we could shift the start block of the reservation
914 * window to the first block of next group.
917 if ((my_rsv->rsv_start <= group_end_block) &&
918 (my_rsv->rsv_end > group_end_block) &&
919 (start_block >= my_rsv->rsv_start))
922 if ((my_rsv->rsv_alloc_hit >
923 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
925 * if the previously allocation hit ratio is
926 * greater than 1/2, then we double the size of
927 * the reservation window the next time,
928 * otherwise we keep the same size window
931 if (size > EXT2_MAX_RESERVE_BLOCKS)
932 size = EXT2_MAX_RESERVE_BLOCKS;
933 my_rsv->rsv_goal_size= size;
939 * shift the search start to the window near the goal block
941 search_head = search_reserve_window(fs_rsv_root, start_block);
944 * find_next_reservable_window() simply finds a reservable window
945 * inside the given range(start_block, group_end_block).
947 * To make sure the reservation window has a free bit inside it, we
948 * need to check the bitmap after we found a reservable window.
951 ret = find_next_reservable_window(search_head, my_rsv, sb,
952 start_block, group_end_block);
955 if (!rsv_is_empty(&my_rsv->rsv_window))
956 rsv_window_remove(sb, my_rsv);
957 spin_unlock(rsv_lock);
962 * On success, find_next_reservable_window() returns the
963 * reservation window where there is a reservable space after it.
964 * Before we reserve this reservable space, we need
965 * to make sure there is at least a free block inside this region.
967 * Search the first free bit on the block bitmap. Search starts from
968 * the start block of the reservable space we just found.
970 spin_unlock(rsv_lock);
971 first_free_block = bitmap_search_next_usable_block(
972 my_rsv->rsv_start - group_first_block,
973 bitmap_bh, group_end_block - group_first_block + 1);
975 if (first_free_block < 0) {
977 * no free block left on the bitmap, no point
978 * to reserve the space. return failed.
981 if (!rsv_is_empty(&my_rsv->rsv_window))
982 rsv_window_remove(sb, my_rsv);
983 spin_unlock(rsv_lock);
984 return -1; /* failed */
987 start_block = first_free_block + group_first_block;
989 * check if the first free block is within the
990 * free space we just reserved
992 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
993 return 0; /* success */
995 * if the first free bit we found is out of the reservable space
996 * continue search for next reservable space,
997 * start from where the free block is,
998 * we also shift the list head to where we stopped last time
1000 search_head = my_rsv;
1001 spin_lock(rsv_lock);
1006 * try_to_extend_reservation()
1007 * @my_rsv: given reservation window
1009 * @size: the delta to extend
1011 * Attempt to expand the reservation window large enough to have
1012 * required number of free blocks
1014 * Since ext2_try_to_allocate() will always allocate blocks within
1015 * the reservation window range, if the window size is too small,
1016 * multiple blocks allocation has to stop at the end of the reservation
1017 * window. To make this more efficient, given the total number of
1018 * blocks needed and the current size of the window, we try to
1019 * expand the reservation window size if necessary on a best-effort
1020 * basis before ext2_new_blocks() tries to allocate blocks.
1022 static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1023 struct super_block *sb, int size)
1025 struct ext2_reserve_window_node *next_rsv;
1026 struct rb_node *next;
1027 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1029 if (!spin_trylock(rsv_lock))
1032 next = rb_next(&my_rsv->rsv_node);
1035 my_rsv->rsv_end += size;
1037 next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1039 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1040 my_rsv->rsv_end += size;
1042 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1044 spin_unlock(rsv_lock);
1048 * ext2_try_to_allocate_with_rsv()
1050 * @group: given allocation block group
1051 * @bitmap_bh: bufferhead holds the block bitmap
1052 * @grp_goal: given target block within the group
1053 * @count: target number of blocks to allocate
1054 * @my_rsv: reservation window
1056 * This is the main function used to allocate a new block and its reservation
1059 * Each time when a new block allocation is need, first try to allocate from
1060 * its own reservation. If it does not have a reservation window, instead of
1061 * looking for a free bit on bitmap first, then look up the reservation list to
1062 * see if it is inside somebody else's reservation window, we try to allocate a
1063 * reservation window for it starting from the goal first. Then do the block
1064 * allocation within the reservation window.
1066 * This will avoid keeping on searching the reservation list again and
1067 * again when somebody is looking for a free block (without
1068 * reservation), and there are lots of free blocks, but they are all
1071 * We use a red-black tree for the per-filesystem reservation list.
1073 static ext2_grpblk_t
1074 ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1075 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1076 struct ext2_reserve_window_node * my_rsv,
1077 unsigned long *count)
1079 ext2_fsblk_t group_first_block, group_last_block;
1080 ext2_grpblk_t ret = 0;
1081 unsigned long num = *count;
1084 * we don't deal with reservation when
1085 * filesystem is mounted without reservation
1086 * or the file is not a regular file
1087 * or last attempt to allocate a block with reservation turned on failed
1089 if (my_rsv == NULL) {
1090 return ext2_try_to_allocate(sb, group, bitmap_bh,
1091 grp_goal, count, NULL);
1094 * grp_goal is a group relative block number (if there is a goal)
1095 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1096 * first block is a filesystem wide block number
1097 * first block is the block number of the first block in this group
1099 group_first_block = ext2_group_first_block_no(sb, group);
1100 group_last_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
1103 * Basically we will allocate a new block from inode's reservation
1106 * We need to allocate a new reservation window, if:
1107 * a) inode does not have a reservation window; or
1108 * b) last attempt to allocate a block from existing reservation
1110 * c) we come here with a goal and with a reservation window
1112 * We do not need to allocate a new reservation window if we come here
1113 * at the beginning with a goal and the goal is inside the window, or
1114 * we don't have a goal but already have a reservation window.
1115 * then we could go to allocate from the reservation window directly.
1118 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1119 !goal_in_my_reservation(&my_rsv->rsv_window,
1120 grp_goal, group, sb)) {
1121 if (my_rsv->rsv_goal_size < *count)
1122 my_rsv->rsv_goal_size = *count;
1123 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1128 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1129 grp_goal, group, sb))
1131 } else if (grp_goal >= 0) {
1132 int curr = my_rsv->rsv_end -
1133 (grp_goal + group_first_block) + 1;
1136 try_to_extend_reservation(my_rsv, sb,
1140 if ((my_rsv->rsv_start > group_last_block) ||
1141 (my_rsv->rsv_end < group_first_block)) {
1142 rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1145 ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1146 &num, &my_rsv->rsv_window);
1148 my_rsv->rsv_alloc_hit += num;
1150 break; /* succeed */
1158 * ext2_has_free_blocks()
1159 * @sbi: in-core super block structure.
1161 * Check if filesystem has at least 1 free block available for allocation.
1163 static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1165 ext2_fsblk_t free_blocks, root_blocks;
1167 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1168 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1169 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1170 sbi->s_resuid != current->fsuid &&
1171 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1178 * ext2_new_blocks() -- core block(s) allocation function
1179 * @inode: file inode
1180 * @goal: given target block(filesystem wide)
1181 * @count: target number of blocks to allocate
1184 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
1185 * free, or there is a free block within 32 blocks of the goal, that block
1186 * is allocated. Otherwise a forward search is made for a free block; within
1187 * each block group the search first looks for an entire free byte in the block
1188 * bitmap, and then for any free bit if that fails.
1189 * This function also updates quota and i_blocks field.
1191 ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1192 unsigned long *count, int *errp)
1194 struct buffer_head *bitmap_bh = NULL;
1195 struct buffer_head *gdp_bh;
1198 ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1199 ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1200 ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
1201 int bgi; /* blockgroup iteration index */
1202 int performed_allocation = 0;
1203 ext2_grpblk_t free_blocks; /* number of free blocks in a group */
1204 struct super_block *sb;
1205 struct ext2_group_desc *gdp;
1206 struct ext2_super_block *es;
1207 struct ext2_sb_info *sbi;
1208 struct ext2_reserve_window_node *my_rsv = NULL;
1209 struct ext2_block_alloc_info *block_i;
1210 unsigned short windowsz = 0;
1211 unsigned long ngroups;
1212 unsigned long num = *count;
1217 printk("ext2_new_blocks: nonexistent device");
1222 * Check quota for allocation of this block.
1224 if (DQUOT_ALLOC_BLOCK(inode, num)) {
1230 es = EXT2_SB(sb)->s_es;
1231 ext2_debug("goal=%lu.\n", goal);
1233 * Allocate a block from reservation only when
1234 * filesystem is mounted with reservation(default,-o reservation), and
1235 * it's a regular file, and
1236 * the desired window size is greater than 0 (One could use ioctl
1237 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1238 * reservation on that particular file)
1240 block_i = EXT2_I(inode)->i_block_alloc_info;
1242 windowsz = block_i->rsv_window_node.rsv_goal_size;
1244 my_rsv = &block_i->rsv_window_node;
1247 if (!ext2_has_free_blocks(sbi)) {
1253 * First, test whether the goal block is free.
1255 if (goal < le32_to_cpu(es->s_first_data_block) ||
1256 goal >= le32_to_cpu(es->s_blocks_count))
1257 goal = le32_to_cpu(es->s_first_data_block);
1258 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1259 EXT2_BLOCKS_PER_GROUP(sb);
1260 goal_group = group_no;
1262 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1266 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1268 * if there is not enough free blocks to make a new resevation
1269 * turn off reservation for this allocation
1271 if (my_rsv && (free_blocks < windowsz)
1272 && (rsv_is_empty(&my_rsv->rsv_window)))
1275 if (free_blocks > 0) {
1276 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1277 EXT2_BLOCKS_PER_GROUP(sb));
1278 bitmap_bh = read_block_bitmap(sb, group_no);
1281 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1282 bitmap_bh, grp_target_blk,
1284 if (grp_alloc_blk >= 0)
1288 ngroups = EXT2_SB(sb)->s_groups_count;
1292 * Now search the rest of the groups. We assume that
1293 * i and gdp correctly point to the last group visited.
1295 for (bgi = 0; bgi < ngroups; bgi++) {
1297 if (group_no >= ngroups)
1299 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1303 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1305 * skip this group if the number of
1306 * free blocks is less than half of the reservation
1309 if (free_blocks <= (windowsz/2))
1313 bitmap_bh = read_block_bitmap(sb, group_no);
1317 * try to allocate block(s) from this group, without a goal(-1).
1319 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1320 bitmap_bh, -1, my_rsv, &num);
1321 if (grp_alloc_blk >= 0)
1325 * We may end up a bogus ealier ENOSPC error due to
1326 * filesystem is "full" of reservations, but
1327 * there maybe indeed free blocks avaliable on disk
1328 * In this case, we just forget about the reservations
1329 * just do block allocation as without reservations.
1334 group_no = goal_group;
1337 /* No space left on the device */
1343 ext2_debug("using block group %d(%d)\n",
1344 group_no, gdp->bg_free_blocks_count);
1346 ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1348 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1349 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1350 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1351 EXT2_SB(sb)->s_itb_per_group) ||
1352 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1353 EXT2_SB(sb)->s_itb_per_group))
1354 ext2_error(sb, "ext2_new_blocks",
1355 "Allocating block in system zone - "
1356 "blocks from "E2FSBLK", length %lu",
1359 performed_allocation = 1;
1361 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1362 ext2_error(sb, "ext2_new_blocks",
1363 "block("E2FSBLK") >= blocks count(%d) - "
1364 "block_group = %d, es == %p ", ret_block,
1365 le32_to_cpu(es->s_blocks_count), group_no, es);
1369 group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1370 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1372 mark_buffer_dirty(bitmap_bh);
1373 if (sb->s_flags & MS_SYNCHRONOUS)
1374 sync_dirty_buffer(bitmap_bh);
1378 DQUOT_FREE_BLOCK(inode, *count-num);
1386 * Undo the block allocation
1388 if (!performed_allocation)
1389 DQUOT_FREE_BLOCK(inode, *count);
1394 ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1396 unsigned long count = 1;
1398 return ext2_new_blocks(inode, goal, &count, errp);
1403 static const int nibblemap[] = {4, 3, 3, 2, 3, 2, 2, 1, 3, 2, 2, 1, 2, 1, 1, 0};
1405 unsigned long ext2_count_free (struct buffer_head * map, unsigned int numchars)
1408 unsigned long sum = 0;
1412 for (i = 0; i < numchars; i++)
1413 sum += nibblemap[map->b_data[i] & 0xf] +
1414 nibblemap[(map->b_data[i] >> 4) & 0xf];
1418 #endif /* EXT2FS_DEBUG */
1420 unsigned long ext2_count_free_blocks (struct super_block * sb)
1422 struct ext2_group_desc * desc;
1423 unsigned long desc_count = 0;
1426 unsigned long bitmap_count, x;
1427 struct ext2_super_block *es;
1429 es = EXT2_SB(sb)->s_es;
1433 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1434 struct buffer_head *bitmap_bh;
1435 desc = ext2_get_group_desc (sb, i, NULL);
1438 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1439 bitmap_bh = read_block_bitmap(sb, i);
1443 x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1444 printk ("group %d: stored = %d, counted = %lu\n",
1445 i, le16_to_cpu(desc->bg_free_blocks_count), x);
1449 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1450 (long)le32_to_cpu(es->s_free_blocks_count),
1451 desc_count, bitmap_count);
1452 return bitmap_count;
1454 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1455 desc = ext2_get_group_desc (sb, i, NULL);
1458 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1465 static inline int test_root(int a, int b)
1474 static int ext2_group_sparse(int group)
1478 return (test_root(group, 3) || test_root(group, 5) ||
1479 test_root(group, 7));
1483 * ext2_bg_has_super - number of blocks used by the superblock in group
1484 * @sb: superblock for filesystem
1485 * @group: group number to check
1487 * Return the number of blocks used by the superblock (primary or backup)
1488 * in this group. Currently this will be only 0 or 1.
1490 int ext2_bg_has_super(struct super_block *sb, int group)
1492 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1493 !ext2_group_sparse(group))
1499 * ext2_bg_num_gdb - number of blocks used by the group table in group
1500 * @sb: superblock for filesystem
1501 * @group: group number to check
1503 * Return the number of blocks used by the group descriptor table
1504 * (primary or backup) in this group. In the future there may be a
1505 * different number of descriptor blocks in each group.
1507 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1509 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1510 !ext2_group_sparse(group))
1512 return EXT2_SB(sb)->s_gdb_count;