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/slab.h>
17 #include <linux/sched.h>
18 #include <linux/buffer_head.h>
19 #include <linux/capability.h>
22 * balloc.c contains the blocks allocation and deallocation routines
26 * The free blocks are managed by bitmaps. A file system contains several
27 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
28 * block for inodes, N blocks for the inode table and data blocks.
30 * The file system contains group descriptors which are located after the
31 * super block. Each descriptor contains the number of the bitmap block and
32 * the free blocks count in the block. The descriptors are loaded in memory
33 * when a file system is mounted (see ext2_fill_super).
37 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
39 struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
40 unsigned int block_group,
41 struct buffer_head ** bh)
43 unsigned long group_desc;
45 struct ext2_group_desc * desc;
46 struct ext2_sb_info *sbi = EXT2_SB(sb);
48 if (block_group >= sbi->s_groups_count) {
49 WARN(1, "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 WARN(1, "Group descriptor not loaded - "
60 "block_group = %d, group_desc = %lu, desc = %lu",
61 block_group, group_desc, offset);
65 desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
67 *bh = sbi->s_group_desc[group_desc];
71 static int ext2_valid_block_bitmap(struct super_block *sb,
72 struct ext2_group_desc *desc,
73 unsigned int block_group,
74 struct buffer_head *bh)
77 ext2_grpblk_t next_zero_bit;
78 ext2_fsblk_t bitmap_blk;
79 ext2_fsblk_t group_first_block;
81 group_first_block = ext2_group_first_block_no(sb, block_group);
83 /* check whether block bitmap block number is set */
84 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
85 offset = bitmap_blk - group_first_block;
86 if (!ext2_test_bit(offset, bh->b_data))
87 /* bad block bitmap */
90 /* check whether the inode bitmap block number is set */
91 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
92 offset = bitmap_blk - group_first_block;
93 if (!ext2_test_bit(offset, bh->b_data))
94 /* bad block bitmap */
97 /* check whether the inode table block number is set */
98 bitmap_blk = le32_to_cpu(desc->bg_inode_table);
99 offset = bitmap_blk - group_first_block;
100 next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
101 offset + EXT2_SB(sb)->s_itb_per_group,
103 if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
104 /* good bitmap for inode tables */
108 ext2_error(sb, __func__,
109 "Invalid block bitmap - "
110 "block_group = %d, block = %lu",
111 block_group, bitmap_blk);
116 * Read the bitmap for a given block_group,and validate the
117 * bits for block/inode/inode tables are set in the bitmaps
119 * Return buffer_head on success or NULL in case of failure.
121 static struct buffer_head *
122 read_block_bitmap(struct super_block *sb, unsigned int block_group)
124 struct ext2_group_desc * desc;
125 struct buffer_head * bh = NULL;
126 ext2_fsblk_t bitmap_blk;
128 desc = ext2_get_group_desc(sb, block_group, NULL);
131 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
132 bh = sb_getblk(sb, bitmap_blk);
134 ext2_error(sb, __func__,
135 "Cannot read block bitmap - "
136 "block_group = %d, block_bitmap = %u",
137 block_group, le32_to_cpu(desc->bg_block_bitmap));
140 if (likely(bh_uptodate_or_lock(bh)))
143 if (bh_submit_read(bh) < 0) {
145 ext2_error(sb, __func__,
146 "Cannot read block bitmap - "
147 "block_group = %d, block_bitmap = %u",
148 block_group, le32_to_cpu(desc->bg_block_bitmap));
152 ext2_valid_block_bitmap(sb, desc, block_group, bh);
154 * file system mounted not to panic on error, continue with corrupt
160 static void group_adjust_blocks(struct super_block *sb, int group_no,
161 struct ext2_group_desc *desc, struct buffer_head *bh, int count)
164 struct ext2_sb_info *sbi = EXT2_SB(sb);
165 unsigned free_blocks;
167 spin_lock(sb_bgl_lock(sbi, group_no));
168 free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
169 desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
170 spin_unlock(sb_bgl_lock(sbi, group_no));
171 mark_buffer_dirty(bh);
176 * The reservation window structure operations
177 * --------------------------------------------
178 * Operations include:
179 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
181 * We use a red-black tree to represent per-filesystem reservation
187 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
188 * @rb_root: root of per-filesystem reservation rb tree
189 * @verbose: verbose mode
190 * @fn: function which wishes to dump the reservation map
192 * If verbose is turned on, it will print the whole block reservation
193 * windows(start, end). Otherwise, it will only print out the "bad" windows,
194 * those windows that overlap with their immediate neighbors.
197 static void __rsv_window_dump(struct rb_root *root, int verbose,
201 struct ext2_reserve_window_node *rsv, *prev;
209 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
211 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
213 printk("reservation window 0x%p "
214 "start: %lu, end: %lu\n",
215 rsv, rsv->rsv_start, rsv->rsv_end);
216 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
217 printk("Bad reservation %p (start >= end)\n",
221 if (prev && prev->rsv_end >= rsv->rsv_start) {
222 printk("Bad reservation %p (prev->end >= start)\n",
228 printk("Restarting reservation walk in verbose mode\n");
236 printk("Window map complete.\n");
239 #define rsv_window_dump(root, verbose) \
240 __rsv_window_dump((root), (verbose), __func__)
242 #define rsv_window_dump(root, verbose) do {} while (0)
246 * goal_in_my_reservation()
247 * @rsv: inode's reservation window
248 * @grp_goal: given goal block relative to the allocation block group
249 * @group: the current allocation block group
250 * @sb: filesystem super block
252 * Test if the given goal block (group relative) is within the file's
253 * own block reservation window range.
255 * If the reservation window is outside the goal allocation group, return 0;
256 * grp_goal (given goal block) could be -1, which means no specific
257 * goal block. In this case, always return 1.
258 * If the goal block is within the reservation window, return 1;
259 * otherwise, return 0;
262 goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
263 unsigned int group, struct super_block * sb)
265 ext2_fsblk_t group_first_block, group_last_block;
267 group_first_block = ext2_group_first_block_no(sb, group);
268 group_last_block = group_first_block + EXT2_BLOCKS_PER_GROUP(sb) - 1;
270 if ((rsv->_rsv_start > group_last_block) ||
271 (rsv->_rsv_end < group_first_block))
273 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
274 || (grp_goal + group_first_block > rsv->_rsv_end)))
280 * search_reserve_window()
281 * @rb_root: root of reservation tree
282 * @goal: target allocation block
284 * Find the reserved window which includes the goal, or the previous one
285 * if the goal is not in any window.
286 * Returns NULL if there are no windows or if all windows start after the goal.
288 static struct ext2_reserve_window_node *
289 search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
291 struct rb_node *n = root->rb_node;
292 struct ext2_reserve_window_node *rsv;
298 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
300 if (goal < rsv->rsv_start)
302 else if (goal > rsv->rsv_end)
308 * We've fallen off the end of the tree: the goal wasn't inside
309 * any particular node. OK, the previous node must be to one
310 * side of the interval containing the goal. If it's the RHS,
311 * we need to back up one.
313 if (rsv->rsv_start > goal) {
314 n = rb_prev(&rsv->rsv_node);
315 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
321 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
323 * @rsv: reservation window to add
325 * Must be called with rsv_lock held.
327 void ext2_rsv_window_add(struct super_block *sb,
328 struct ext2_reserve_window_node *rsv)
330 struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
331 struct rb_node *node = &rsv->rsv_node;
332 ext2_fsblk_t start = rsv->rsv_start;
334 struct rb_node ** p = &root->rb_node;
335 struct rb_node * parent = NULL;
336 struct ext2_reserve_window_node *this;
341 this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
343 if (start < this->rsv_start)
345 else if (start > this->rsv_end)
348 rsv_window_dump(root, 1);
353 rb_link_node(node, parent, p);
354 rb_insert_color(node, root);
358 * rsv_window_remove() -- unlink a window from the reservation rb tree
360 * @rsv: reservation window to remove
362 * Mark the block reservation window as not allocated, and unlink it
363 * from the filesystem reservation window rb tree. Must be called with
366 static void rsv_window_remove(struct super_block *sb,
367 struct ext2_reserve_window_node *rsv)
369 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
370 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
371 rsv->rsv_alloc_hit = 0;
372 rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
376 * rsv_is_empty() -- Check if the reservation window is allocated.
377 * @rsv: given reservation window to check
379 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
381 static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
383 /* a valid reservation end block could not be 0 */
384 return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
388 * ext2_init_block_alloc_info()
389 * @inode: file inode structure
391 * Allocate and initialize the reservation window structure, and
392 * link the window to the ext2 inode structure at last
394 * The reservation window structure is only dynamically allocated
395 * and linked to ext2 inode the first time the open file
396 * needs a new block. So, before every ext2_new_block(s) call, for
397 * regular files, we should check whether the reservation window
398 * structure exists or not. In the latter case, this function is called.
399 * Fail to do so will result in block reservation being turned off for that
402 * This function is called from ext2_get_blocks_handle(), also called
403 * when setting the reservation window size through ioctl before the file
404 * is open for write (needs block allocation).
406 * Needs truncate_mutex protection prior to calling this function.
408 void ext2_init_block_alloc_info(struct inode *inode)
410 struct ext2_inode_info *ei = EXT2_I(inode);
411 struct ext2_block_alloc_info *block_i;
412 struct super_block *sb = inode->i_sb;
414 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
416 struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
418 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
419 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
422 * if filesystem is mounted with NORESERVATION, the goal
423 * reservation window size is set to zero to indicate
424 * block reservation is off
426 if (!test_opt(sb, RESERVATION))
427 rsv->rsv_goal_size = 0;
429 rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
430 rsv->rsv_alloc_hit = 0;
431 block_i->last_alloc_logical_block = 0;
432 block_i->last_alloc_physical_block = 0;
434 ei->i_block_alloc_info = block_i;
438 * ext2_discard_reservation()
441 * Discard(free) block reservation window on last file close, or truncate
444 * It is being called in three cases:
445 * ext2_release_file(): last writer closes the file
446 * ext2_clear_inode(): last iput(), when nobody links to this file.
447 * ext2_truncate(): when the block indirect map is about to change.
449 void ext2_discard_reservation(struct inode *inode)
451 struct ext2_inode_info *ei = EXT2_I(inode);
452 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
453 struct ext2_reserve_window_node *rsv;
454 spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
459 rsv = &block_i->rsv_window_node;
460 if (!rsv_is_empty(&rsv->rsv_window)) {
462 if (!rsv_is_empty(&rsv->rsv_window))
463 rsv_window_remove(inode->i_sb, rsv);
464 spin_unlock(rsv_lock);
469 * ext2_free_blocks() -- Free given blocks and update quota and i_blocks
471 * @block: start physical block to free
472 * @count: number of blocks to free
474 void ext2_free_blocks (struct inode * inode, unsigned long block,
477 struct buffer_head *bitmap_bh = NULL;
478 struct buffer_head * bh2;
479 unsigned long block_group;
482 unsigned long overflow;
483 struct super_block * sb = inode->i_sb;
484 struct ext2_sb_info * sbi = EXT2_SB(sb);
485 struct ext2_group_desc * desc;
486 struct ext2_super_block * es = sbi->s_es;
487 unsigned freed = 0, group_freed;
489 if (block < le32_to_cpu(es->s_first_data_block) ||
490 block + count < block ||
491 block + count > le32_to_cpu(es->s_blocks_count)) {
492 ext2_error (sb, "ext2_free_blocks",
493 "Freeing blocks not in datazone - "
494 "block = %lu, count = %lu", block, count);
498 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
502 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
503 EXT2_BLOCKS_PER_GROUP(sb);
504 bit = (block - le32_to_cpu(es->s_first_data_block)) %
505 EXT2_BLOCKS_PER_GROUP(sb);
507 * Check to see if we are freeing blocks across a group
510 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
511 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
515 bitmap_bh = read_block_bitmap(sb, block_group);
519 desc = ext2_get_group_desc (sb, block_group, &bh2);
523 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
524 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
525 in_range (block, le32_to_cpu(desc->bg_inode_table),
526 sbi->s_itb_per_group) ||
527 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
528 sbi->s_itb_per_group)) {
529 ext2_error (sb, "ext2_free_blocks",
530 "Freeing blocks in system zones - "
531 "Block = %lu, count = %lu",
536 for (i = 0, group_freed = 0; i < count; i++) {
537 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
538 bit + i, bitmap_bh->b_data)) {
539 ext2_error(sb, __func__,
540 "bit already cleared for block %lu", block + i);
546 mark_buffer_dirty(bitmap_bh);
547 if (sb->s_flags & MS_SYNCHRONOUS)
548 sync_dirty_buffer(bitmap_bh);
550 group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
551 freed += group_freed;
561 percpu_counter_add(&sbi->s_freeblocks_counter, freed);
562 dquot_free_block_nodirty(inode, freed);
563 mark_inode_dirty(inode);
568 * bitmap_search_next_usable_block()
569 * @start: the starting block (group relative) of the search
570 * @bh: bufferhead contains the block group bitmap
571 * @maxblocks: the ending block (group relative) of the reservation
573 * The bitmap search --- search forward through the actual bitmap on disk until
574 * we find a bit free.
577 bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
578 ext2_grpblk_t maxblocks)
582 next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
583 if (next >= maxblocks)
589 * find_next_usable_block()
590 * @start: the starting block (group relative) to find next
591 * allocatable block in bitmap.
592 * @bh: bufferhead contains the block group bitmap
593 * @maxblocks: the ending block (group relative) for the search
595 * Find an allocatable block in a bitmap. We perform the "most
596 * appropriate allocation" algorithm of looking for a free block near
597 * the initial goal; then for a free byte somewhere in the bitmap;
598 * then for any free bit in the bitmap.
601 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
603 ext2_grpblk_t here, next;
608 * The goal was occupied; search forward for a free
609 * block within the next XX blocks.
611 * end_goal is more or less random, but it has to be
612 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
613 * next 64-bit boundary is simple..
615 ext2_grpblk_t end_goal = (start + 63) & ~63;
616 if (end_goal > maxblocks)
617 end_goal = maxblocks;
618 here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
621 ext2_debug("Bit not found near goal\n");
628 p = ((char *)bh->b_data) + (here >> 3);
629 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
630 next = (r - ((char *)bh->b_data)) << 3;
632 if (next < maxblocks && next >= here)
635 here = bitmap_search_next_usable_block(here, bh, maxblocks);
640 * ext2_try_to_allocate()
642 * @group: given allocation block group
643 * @bitmap_bh: bufferhead holds the block bitmap
644 * @grp_goal: given target block within the group
645 * @count: target number of blocks to allocate
646 * @my_rsv: reservation window
648 * Attempt to allocate blocks within a give range. Set the range of allocation
649 * first, then find the first free bit(s) from the bitmap (within the range),
650 * and at last, allocate the blocks by claiming the found free bit as allocated.
652 * To set the range of this allocation:
653 * if there is a reservation window, only try to allocate block(s)
654 * from the file's own reservation window;
655 * Otherwise, the allocation range starts from the give goal block,
656 * ends at the block group's last block.
658 * If we failed to allocate the desired block then we may end up crossing to a
662 ext2_try_to_allocate(struct super_block *sb, int group,
663 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
664 unsigned long *count,
665 struct ext2_reserve_window *my_rsv)
667 ext2_fsblk_t group_first_block;
668 ext2_grpblk_t start, end;
669 unsigned long num = 0;
671 /* we do allocation within the reservation window if we have a window */
673 group_first_block = ext2_group_first_block_no(sb, group);
674 if (my_rsv->_rsv_start >= group_first_block)
675 start = my_rsv->_rsv_start - group_first_block;
677 /* reservation window cross group boundary */
679 end = my_rsv->_rsv_end - group_first_block + 1;
680 if (end > EXT2_BLOCKS_PER_GROUP(sb))
681 /* reservation window crosses group boundary */
682 end = EXT2_BLOCKS_PER_GROUP(sb);
683 if ((start <= grp_goal) && (grp_goal < end))
692 end = EXT2_BLOCKS_PER_GROUP(sb);
695 BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
699 grp_goal = find_next_usable_block(start, bitmap_bh, end);
705 for (i = 0; i < 7 && grp_goal > start &&
706 !ext2_test_bit(grp_goal - 1,
714 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), grp_goal,
715 bitmap_bh->b_data)) {
717 * The block was allocated by another thread, or it was
718 * allocated and then freed by another thread
728 while (num < *count && grp_goal < end
729 && !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
730 grp_goal, bitmap_bh->b_data)) {
735 return grp_goal - num;
742 * find_next_reservable_window():
743 * find a reservable space within the given range.
744 * It does not allocate the reservation window for now:
745 * alloc_new_reservation() will do the work later.
747 * @search_head: the head of the searching list;
748 * This is not necessarily the list head of the whole filesystem
750 * We have both head and start_block to assist the search
751 * for the reservable space. The list starts from head,
752 * but we will shift to the place where start_block is,
753 * then start from there, when looking for a reservable space.
755 * @size: the target new reservation window size
757 * @group_first_block: the first block we consider to start
758 * the real search from
761 * the maximum block number that our goal reservable space
762 * could start from. This is normally the last block in this
763 * group. The search will end when we found the start of next
764 * possible reservable space is out of this boundary.
765 * This could handle the cross boundary reservation window
768 * basically we search from the given range, rather than the whole
769 * reservation double linked list, (start_block, last_block)
770 * to find a free region that is of my size and has not
774 static int find_next_reservable_window(
775 struct ext2_reserve_window_node *search_head,
776 struct ext2_reserve_window_node *my_rsv,
777 struct super_block * sb,
778 ext2_fsblk_t start_block,
779 ext2_fsblk_t last_block)
781 struct rb_node *next;
782 struct ext2_reserve_window_node *rsv, *prev;
784 int size = my_rsv->rsv_goal_size;
786 /* TODO: make the start of the reservation window byte-aligned */
787 /* cur = *start_block & ~7;*/
794 if (cur <= rsv->rsv_end)
795 cur = rsv->rsv_end + 1;
798 * in the case we could not find a reservable space
799 * that is what is expected, during the re-search, we could
800 * remember what's the largest reservable space we could have
801 * and return that one.
803 * For now it will fail if we could not find the reservable
804 * space with expected-size (or more)...
806 if (cur > last_block)
807 return -1; /* fail */
810 next = rb_next(&rsv->rsv_node);
811 rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
814 * Reached the last reservation, we can just append to the
820 if (cur + size <= rsv->rsv_start) {
822 * Found a reserveable space big enough. We could
823 * have a reservation across the group boundary here
829 * we come here either :
830 * when we reach the end of the whole list,
831 * and there is empty reservable space after last entry in the list.
832 * append it to the end of the list.
834 * or we found one reservable space in the middle of the list,
835 * return the reservation window that we could append to.
839 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
840 rsv_window_remove(sb, my_rsv);
843 * Let's book the whole available window for now. We will check the
844 * disk bitmap later and then, if there are free blocks then we adjust
845 * the window size if it's larger than requested.
846 * Otherwise, we will remove this node from the tree next time
847 * call find_next_reservable_window.
849 my_rsv->rsv_start = cur;
850 my_rsv->rsv_end = cur + size - 1;
851 my_rsv->rsv_alloc_hit = 0;
854 ext2_rsv_window_add(sb, my_rsv);
860 * alloc_new_reservation()--allocate a new reservation window
862 * To make a new reservation, we search part of the filesystem
863 * reservation list (the list that inside the group). We try to
864 * allocate a new reservation window near the allocation goal,
865 * or the beginning of the group, if there is no goal.
867 * We first find a reservable space after the goal, then from
868 * there, we check the bitmap for the first free block after
869 * it. If there is no free block until the end of group, then the
870 * whole group is full, we failed. Otherwise, check if the free
871 * block is inside the expected reservable space, if so, we
873 * If the first free block is outside the reservable space, then
874 * start from the first free block, we search for next available
877 * on succeed, a new reservation will be found and inserted into the list
878 * It contains at least one free block, and it does not overlap with other
879 * reservation windows.
881 * failed: we failed to find a reservation window in this group
883 * @rsv: the reservation
885 * @grp_goal: The goal (group-relative). It is where the search for a
886 * free reservable space should start from.
887 * if we have a goal(goal >0 ), then start from there,
888 * no goal(goal = -1), we start from the first block
891 * @sb: the super block
892 * @group: the group we are trying to allocate in
893 * @bitmap_bh: the block group block bitmap
896 static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
897 ext2_grpblk_t grp_goal, struct super_block *sb,
898 unsigned int group, struct buffer_head *bitmap_bh)
900 struct ext2_reserve_window_node *search_head;
901 ext2_fsblk_t group_first_block, group_end_block, start_block;
902 ext2_grpblk_t first_free_block;
903 struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
906 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
908 group_first_block = ext2_group_first_block_no(sb, group);
909 group_end_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
912 start_block = group_first_block;
914 start_block = grp_goal + group_first_block;
916 size = my_rsv->rsv_goal_size;
918 if (!rsv_is_empty(&my_rsv->rsv_window)) {
920 * if the old reservation is cross group boundary
921 * and if the goal is inside the old reservation window,
922 * we will come here when we just failed to allocate from
923 * the first part of the window. We still have another part
924 * that belongs to the next group. In this case, there is no
925 * point to discard our window and try to allocate a new one
926 * in this group(which will fail). we should
927 * keep the reservation window, just simply move on.
929 * Maybe we could shift the start block of the reservation
930 * window to the first block of next group.
933 if ((my_rsv->rsv_start <= group_end_block) &&
934 (my_rsv->rsv_end > group_end_block) &&
935 (start_block >= my_rsv->rsv_start))
938 if ((my_rsv->rsv_alloc_hit >
939 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
941 * if the previously allocation hit ratio is
942 * greater than 1/2, then we double the size of
943 * the reservation window the next time,
944 * otherwise we keep the same size window
947 if (size > EXT2_MAX_RESERVE_BLOCKS)
948 size = EXT2_MAX_RESERVE_BLOCKS;
949 my_rsv->rsv_goal_size= size;
955 * shift the search start to the window near the goal block
957 search_head = search_reserve_window(fs_rsv_root, start_block);
960 * find_next_reservable_window() simply finds a reservable window
961 * inside the given range(start_block, group_end_block).
963 * To make sure the reservation window has a free bit inside it, we
964 * need to check the bitmap after we found a reservable window.
967 ret = find_next_reservable_window(search_head, my_rsv, sb,
968 start_block, group_end_block);
971 if (!rsv_is_empty(&my_rsv->rsv_window))
972 rsv_window_remove(sb, my_rsv);
973 spin_unlock(rsv_lock);
978 * On success, find_next_reservable_window() returns the
979 * reservation window where there is a reservable space after it.
980 * Before we reserve this reservable space, we need
981 * to make sure there is at least a free block inside this region.
983 * Search the first free bit on the block bitmap. Search starts from
984 * the start block of the reservable space we just found.
986 spin_unlock(rsv_lock);
987 first_free_block = bitmap_search_next_usable_block(
988 my_rsv->rsv_start - group_first_block,
989 bitmap_bh, group_end_block - group_first_block + 1);
991 if (first_free_block < 0) {
993 * no free block left on the bitmap, no point
994 * to reserve the space. return failed.
997 if (!rsv_is_empty(&my_rsv->rsv_window))
998 rsv_window_remove(sb, my_rsv);
999 spin_unlock(rsv_lock);
1000 return -1; /* failed */
1003 start_block = first_free_block + group_first_block;
1005 * check if the first free block is within the
1006 * free space we just reserved
1008 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1009 return 0; /* success */
1011 * if the first free bit we found is out of the reservable space
1012 * continue search for next reservable space,
1013 * start from where the free block is,
1014 * we also shift the list head to where we stopped last time
1016 search_head = my_rsv;
1017 spin_lock(rsv_lock);
1022 * try_to_extend_reservation()
1023 * @my_rsv: given reservation window
1025 * @size: the delta to extend
1027 * Attempt to expand the reservation window large enough to have
1028 * required number of free blocks
1030 * Since ext2_try_to_allocate() will always allocate blocks within
1031 * the reservation window range, if the window size is too small,
1032 * multiple blocks allocation has to stop at the end of the reservation
1033 * window. To make this more efficient, given the total number of
1034 * blocks needed and the current size of the window, we try to
1035 * expand the reservation window size if necessary on a best-effort
1036 * basis before ext2_new_blocks() tries to allocate blocks.
1038 static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1039 struct super_block *sb, int size)
1041 struct ext2_reserve_window_node *next_rsv;
1042 struct rb_node *next;
1043 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1045 if (!spin_trylock(rsv_lock))
1048 next = rb_next(&my_rsv->rsv_node);
1051 my_rsv->rsv_end += size;
1053 next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1055 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1056 my_rsv->rsv_end += size;
1058 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1060 spin_unlock(rsv_lock);
1064 * ext2_try_to_allocate_with_rsv()
1066 * @group: given allocation block group
1067 * @bitmap_bh: bufferhead holds the block bitmap
1068 * @grp_goal: given target block within the group
1069 * @count: target number of blocks to allocate
1070 * @my_rsv: reservation window
1072 * This is the main function used to allocate a new block and its reservation
1075 * Each time when a new block allocation is need, first try to allocate from
1076 * its own reservation. If it does not have a reservation window, instead of
1077 * looking for a free bit on bitmap first, then look up the reservation list to
1078 * see if it is inside somebody else's reservation window, we try to allocate a
1079 * reservation window for it starting from the goal first. Then do the block
1080 * allocation within the reservation window.
1082 * This will avoid keeping on searching the reservation list again and
1083 * again when somebody is looking for a free block (without
1084 * reservation), and there are lots of free blocks, but they are all
1087 * We use a red-black tree for the per-filesystem reservation list.
1089 static ext2_grpblk_t
1090 ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1091 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1092 struct ext2_reserve_window_node * my_rsv,
1093 unsigned long *count)
1095 ext2_fsblk_t group_first_block, group_last_block;
1096 ext2_grpblk_t ret = 0;
1097 unsigned long num = *count;
1100 * we don't deal with reservation when
1101 * filesystem is mounted without reservation
1102 * or the file is not a regular file
1103 * or last attempt to allocate a block with reservation turned on failed
1105 if (my_rsv == NULL) {
1106 return ext2_try_to_allocate(sb, group, bitmap_bh,
1107 grp_goal, count, NULL);
1110 * grp_goal is a group relative block number (if there is a goal)
1111 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1112 * first block is a filesystem wide block number
1113 * first block is the block number of the first block in this group
1115 group_first_block = ext2_group_first_block_no(sb, group);
1116 group_last_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
1119 * Basically we will allocate a new block from inode's reservation
1122 * We need to allocate a new reservation window, if:
1123 * a) inode does not have a reservation window; or
1124 * b) last attempt to allocate a block from existing reservation
1126 * c) we come here with a goal and with a reservation window
1128 * We do not need to allocate a new reservation window if we come here
1129 * at the beginning with a goal and the goal is inside the window, or
1130 * we don't have a goal but already have a reservation window.
1131 * then we could go to allocate from the reservation window directly.
1134 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1135 !goal_in_my_reservation(&my_rsv->rsv_window,
1136 grp_goal, group, sb)) {
1137 if (my_rsv->rsv_goal_size < *count)
1138 my_rsv->rsv_goal_size = *count;
1139 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1144 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1145 grp_goal, group, sb))
1147 } else if (grp_goal >= 0) {
1148 int curr = my_rsv->rsv_end -
1149 (grp_goal + group_first_block) + 1;
1152 try_to_extend_reservation(my_rsv, sb,
1156 if ((my_rsv->rsv_start > group_last_block) ||
1157 (my_rsv->rsv_end < group_first_block)) {
1158 rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1161 ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1162 &num, &my_rsv->rsv_window);
1164 my_rsv->rsv_alloc_hit += num;
1166 break; /* succeed */
1174 * ext2_has_free_blocks()
1175 * @sbi: in-core super block structure.
1177 * Check if filesystem has at least 1 free block available for allocation.
1179 static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1181 ext2_fsblk_t free_blocks, root_blocks;
1183 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1184 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1185 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1186 !uid_eq(sbi->s_resuid, current_fsuid()) &&
1187 (gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1188 !in_group_p (sbi->s_resgid))) {
1195 * Returns 1 if the passed-in block region is valid; 0 if some part overlaps
1196 * with filesystem metadata blocksi.
1198 int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk,
1201 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
1202 (start_blk + count < start_blk) ||
1203 (start_blk > le32_to_cpu(sbi->s_es->s_blocks_count)))
1206 /* Ensure we do not step over superblock */
1207 if ((start_blk <= sbi->s_sb_block) &&
1208 (start_blk + count >= sbi->s_sb_block))
1216 * ext2_new_blocks() -- core block(s) allocation function
1217 * @inode: file inode
1218 * @goal: given target block(filesystem wide)
1219 * @count: target number of blocks to allocate
1222 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
1223 * free, or there is a free block within 32 blocks of the goal, that block
1224 * is allocated. Otherwise a forward search is made for a free block; within
1225 * each block group the search first looks for an entire free byte in the block
1226 * bitmap, and then for any free bit if that fails.
1227 * This function also updates quota and i_blocks field.
1229 ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1230 unsigned long *count, int *errp)
1232 struct buffer_head *bitmap_bh = NULL;
1233 struct buffer_head *gdp_bh;
1236 ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1237 ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1238 ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
1239 int bgi; /* blockgroup iteration index */
1240 int performed_allocation = 0;
1241 ext2_grpblk_t free_blocks; /* number of free blocks in a group */
1242 struct super_block *sb;
1243 struct ext2_group_desc *gdp;
1244 struct ext2_super_block *es;
1245 struct ext2_sb_info *sbi;
1246 struct ext2_reserve_window_node *my_rsv = NULL;
1247 struct ext2_block_alloc_info *block_i;
1248 unsigned short windowsz = 0;
1249 unsigned long ngroups;
1250 unsigned long num = *count;
1257 * Check quota for allocation of this block.
1259 ret = dquot_alloc_block(inode, num);
1266 es = EXT2_SB(sb)->s_es;
1267 ext2_debug("goal=%lu.\n", goal);
1269 * Allocate a block from reservation only when
1270 * filesystem is mounted with reservation(default,-o reservation), and
1271 * it's a regular file, and
1272 * the desired window size is greater than 0 (One could use ioctl
1273 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1274 * reservation on that particular file)
1276 block_i = EXT2_I(inode)->i_block_alloc_info;
1278 windowsz = block_i->rsv_window_node.rsv_goal_size;
1280 my_rsv = &block_i->rsv_window_node;
1283 if (!ext2_has_free_blocks(sbi)) {
1289 * First, test whether the goal block is free.
1291 if (goal < le32_to_cpu(es->s_first_data_block) ||
1292 goal >= le32_to_cpu(es->s_blocks_count))
1293 goal = le32_to_cpu(es->s_first_data_block);
1294 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1295 EXT2_BLOCKS_PER_GROUP(sb);
1296 goal_group = group_no;
1298 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1302 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1304 * if there is not enough free blocks to make a new resevation
1305 * turn off reservation for this allocation
1307 if (my_rsv && (free_blocks < windowsz)
1308 && (free_blocks > 0)
1309 && (rsv_is_empty(&my_rsv->rsv_window)))
1312 if (free_blocks > 0) {
1313 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1314 EXT2_BLOCKS_PER_GROUP(sb));
1315 bitmap_bh = read_block_bitmap(sb, group_no);
1318 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1319 bitmap_bh, grp_target_blk,
1321 if (grp_alloc_blk >= 0)
1325 ngroups = EXT2_SB(sb)->s_groups_count;
1329 * Now search the rest of the groups. We assume that
1330 * group_no and gdp correctly point to the last group visited.
1332 for (bgi = 0; bgi < ngroups; bgi++) {
1334 if (group_no >= ngroups)
1336 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1340 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1342 * skip this group (and avoid loading bitmap) if there
1343 * are no free blocks
1348 * skip this group if the number of
1349 * free blocks is less than half of the reservation
1352 if (my_rsv && (free_blocks <= (windowsz/2)))
1356 bitmap_bh = read_block_bitmap(sb, group_no);
1360 * try to allocate block(s) from this group, without a goal(-1).
1362 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1363 bitmap_bh, -1, my_rsv, &num);
1364 if (grp_alloc_blk >= 0)
1368 * We may end up a bogus earlier ENOSPC error due to
1369 * filesystem is "full" of reservations, but
1370 * there maybe indeed free blocks available on disk
1371 * In this case, we just forget about the reservations
1372 * just do block allocation as without reservations.
1377 group_no = goal_group;
1380 /* No space left on the device */
1386 ext2_debug("using block group %d(%d)\n",
1387 group_no, gdp->bg_free_blocks_count);
1389 ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1391 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1392 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1393 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1394 EXT2_SB(sb)->s_itb_per_group) ||
1395 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1396 EXT2_SB(sb)->s_itb_per_group)) {
1397 ext2_error(sb, "ext2_new_blocks",
1398 "Allocating block in system zone - "
1399 "blocks from "E2FSBLK", length %lu",
1402 * ext2_try_to_allocate marked the blocks we allocated as in
1403 * use. So we may want to selectively mark some of the blocks
1409 performed_allocation = 1;
1411 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1412 ext2_error(sb, "ext2_new_blocks",
1413 "block("E2FSBLK") >= blocks count(%d) - "
1414 "block_group = %d, es == %p ", ret_block,
1415 le32_to_cpu(es->s_blocks_count), group_no, es);
1419 group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1420 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1422 mark_buffer_dirty(bitmap_bh);
1423 if (sb->s_flags & MS_SYNCHRONOUS)
1424 sync_dirty_buffer(bitmap_bh);
1429 dquot_free_block_nodirty(inode, *count-num);
1430 mark_inode_dirty(inode);
1439 * Undo the block allocation
1441 if (!performed_allocation) {
1442 dquot_free_block_nodirty(inode, *count);
1443 mark_inode_dirty(inode);
1449 ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1451 unsigned long count = 1;
1453 return ext2_new_blocks(inode, goal, &count, errp);
1458 unsigned long ext2_count_free(struct buffer_head *map, unsigned int numchars)
1460 return numchars * BITS_PER_BYTE - memweight(map->b_data, numchars);
1463 #endif /* EXT2FS_DEBUG */
1465 unsigned long ext2_count_free_blocks (struct super_block * sb)
1467 struct ext2_group_desc * desc;
1468 unsigned long desc_count = 0;
1471 unsigned long bitmap_count, x;
1472 struct ext2_super_block *es;
1474 es = EXT2_SB(sb)->s_es;
1478 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1479 struct buffer_head *bitmap_bh;
1480 desc = ext2_get_group_desc (sb, i, NULL);
1483 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1484 bitmap_bh = read_block_bitmap(sb, i);
1488 x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1489 printk ("group %d: stored = %d, counted = %lu\n",
1490 i, le16_to_cpu(desc->bg_free_blocks_count), x);
1494 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1495 (long)le32_to_cpu(es->s_free_blocks_count),
1496 desc_count, bitmap_count);
1497 return bitmap_count;
1499 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1500 desc = ext2_get_group_desc (sb, i, NULL);
1503 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1509 static inline int test_root(int a, int b)
1518 static int ext2_group_sparse(int group)
1522 return (test_root(group, 3) || test_root(group, 5) ||
1523 test_root(group, 7));
1527 * ext2_bg_has_super - number of blocks used by the superblock in group
1528 * @sb: superblock for filesystem
1529 * @group: group number to check
1531 * Return the number of blocks used by the superblock (primary or backup)
1532 * in this group. Currently this will be only 0 or 1.
1534 int ext2_bg_has_super(struct super_block *sb, int group)
1536 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1537 !ext2_group_sparse(group))
1543 * ext2_bg_num_gdb - number of blocks used by the group table in group
1544 * @sb: superblock for filesystem
1545 * @group: group number to check
1547 * Return the number of blocks used by the group descriptor table
1548 * (primary or backup) in this group. In the future there may be a
1549 * different number of descriptor blocks in each group.
1551 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1553 return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;