2 * linux/fs/ext4/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
14 #include <linux/time.h>
15 #include <linux/capability.h>
17 #include <linux/jbd2.h>
18 #include <linux/ext4_fs.h>
19 #include <linux/ext4_jbd2.h>
20 #include <linux/quotaops.h>
21 #include <linux/buffer_head.h>
25 * balloc.c contains the blocks allocation and deallocation routines
29 * Calculate the block group number and offset, given a block number
31 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
32 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
34 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
37 blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
38 offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
46 /* Initializes an uninitialized block bitmap if given, and returns the
47 * number of blocks free in the group. */
48 unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
49 ext4_group_t block_group, struct ext4_group_desc *gdp)
53 unsigned free_blocks, group_blocks;
54 struct ext4_sb_info *sbi = EXT4_SB(sb);
57 J_ASSERT_BH(bh, buffer_locked(bh));
59 /* If checksum is bad mark all blocks used to prevent allocation
60 * essentially implementing a per-group read-only flag. */
61 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
62 ext4_error(sb, __FUNCTION__,
63 "Checksum bad for group %lu\n", block_group);
64 gdp->bg_free_blocks_count = 0;
65 gdp->bg_free_inodes_count = 0;
66 gdp->bg_itable_unused = 0;
67 memset(bh->b_data, 0xff, sb->s_blocksize);
70 memset(bh->b_data, 0, sb->s_blocksize);
73 /* Check for superblock and gdt backups in this group */
74 bit_max = ext4_bg_has_super(sb, block_group);
76 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
77 block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
78 sbi->s_desc_per_block) {
80 bit_max += ext4_bg_num_gdb(sb, block_group);
82 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
84 } else { /* For META_BG_BLOCK_GROUPS */
85 int group_rel = (block_group -
86 le32_to_cpu(sbi->s_es->s_first_meta_bg)) %
87 EXT4_DESC_PER_BLOCK(sb);
88 if (group_rel == 0 || group_rel == 1 ||
89 (group_rel == EXT4_DESC_PER_BLOCK(sb) - 1))
93 if (block_group == sbi->s_groups_count - 1) {
95 * Even though mke2fs always initialize first and last group
96 * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
97 * to make sure we calculate the right free blocks
99 group_blocks = ext4_blocks_count(sbi->s_es) -
100 le32_to_cpu(sbi->s_es->s_first_data_block) -
101 (EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count -1));
103 group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
106 free_blocks = group_blocks - bit_max;
109 for (bit = 0; bit < bit_max; bit++)
110 ext4_set_bit(bit, bh->b_data);
112 start = block_group * EXT4_BLOCKS_PER_GROUP(sb) +
113 le32_to_cpu(sbi->s_es->s_first_data_block);
115 /* Set bits for block and inode bitmaps, and inode table */
116 ext4_set_bit(ext4_block_bitmap(sb, gdp) - start, bh->b_data);
117 ext4_set_bit(ext4_inode_bitmap(sb, gdp) - start, bh->b_data);
118 for (bit = (ext4_inode_table(sb, gdp) - start),
119 bit_max = bit + sbi->s_itb_per_group; bit < bit_max; bit++)
120 ext4_set_bit(bit, bh->b_data);
123 * Also if the number of blocks within the group is
124 * less than the blocksize * 8 ( which is the size
125 * of bitmap ), set rest of the block bitmap to 1
127 mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
130 return free_blocks - sbi->s_itb_per_group - 2;
135 * The free blocks are managed by bitmaps. A file system contains several
136 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
137 * block for inodes, N blocks for the inode table and data blocks.
139 * The file system contains group descriptors which are located after the
140 * super block. Each descriptor contains the number of the bitmap block and
141 * the free blocks count in the block. The descriptors are loaded in memory
142 * when a file system is mounted (see ext4_fill_super).
146 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
149 * ext4_get_group_desc() -- load group descriptor from disk
151 * @block_group: given block group
152 * @bh: pointer to the buffer head to store the block
155 struct ext4_group_desc * ext4_get_group_desc(struct super_block * sb,
156 ext4_group_t block_group,
157 struct buffer_head ** bh)
159 unsigned long group_desc;
160 unsigned long offset;
161 struct ext4_group_desc * desc;
162 struct ext4_sb_info *sbi = EXT4_SB(sb);
164 if (block_group >= sbi->s_groups_count) {
165 ext4_error (sb, "ext4_get_group_desc",
166 "block_group >= groups_count - "
167 "block_group = %lu, groups_count = %lu",
168 block_group, sbi->s_groups_count);
174 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
175 offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
176 if (!sbi->s_group_desc[group_desc]) {
177 ext4_error (sb, "ext4_get_group_desc",
178 "Group descriptor not loaded - "
179 "block_group = %lu, group_desc = %lu, desc = %lu",
180 block_group, group_desc, offset);
184 desc = (struct ext4_group_desc *)(
185 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
186 offset * EXT4_DESC_SIZE(sb));
188 *bh = sbi->s_group_desc[group_desc];
193 * read_block_bitmap()
195 * @block_group: given block group
197 * Read the bitmap for a given block_group, reading into the specified
198 * slot in the superblock's bitmap cache.
200 * Return buffer_head on success or NULL in case of failure.
203 read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
205 struct ext4_group_desc * desc;
206 struct buffer_head * bh = NULL;
207 ext4_fsblk_t bitmap_blk;
209 desc = ext4_get_group_desc(sb, block_group, NULL);
212 bitmap_blk = ext4_block_bitmap(sb, desc);
213 if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
214 bh = sb_getblk(sb, bitmap_blk);
215 if (!buffer_uptodate(bh)) {
217 if (!buffer_uptodate(bh)) {
218 ext4_init_block_bitmap(sb, bh, block_group,
220 set_buffer_uptodate(bh);
225 bh = sb_bread(sb, bitmap_blk);
228 ext4_error (sb, __FUNCTION__,
229 "Cannot read block bitmap - "
230 "block_group = %lu, block_bitmap = %llu",
231 block_group, bitmap_blk);
235 * The reservation window structure operations
236 * --------------------------------------------
237 * Operations include:
238 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
240 * We use a red-black tree to represent per-filesystem reservation
246 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
247 * @rb_root: root of per-filesystem reservation rb tree
248 * @verbose: verbose mode
249 * @fn: function which wishes to dump the reservation map
251 * If verbose is turned on, it will print the whole block reservation
252 * windows(start, end). Otherwise, it will only print out the "bad" windows,
253 * those windows that overlap with their immediate neighbors.
256 static void __rsv_window_dump(struct rb_root *root, int verbose,
260 struct ext4_reserve_window_node *rsv, *prev;
268 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
270 rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node);
272 printk("reservation window 0x%p "
273 "start: %llu, end: %llu\n",
274 rsv, rsv->rsv_start, rsv->rsv_end);
275 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
276 printk("Bad reservation %p (start >= end)\n",
280 if (prev && prev->rsv_end >= rsv->rsv_start) {
281 printk("Bad reservation %p (prev->end >= start)\n",
287 printk("Restarting reservation walk in verbose mode\n");
295 printk("Window map complete.\n");
299 #define rsv_window_dump(root, verbose) \
300 __rsv_window_dump((root), (verbose), __FUNCTION__)
302 #define rsv_window_dump(root, verbose) do {} while (0)
306 * goal_in_my_reservation()
307 * @rsv: inode's reservation window
308 * @grp_goal: given goal block relative to the allocation block group
309 * @group: the current allocation block group
310 * @sb: filesystem super block
312 * Test if the given goal block (group relative) is within the file's
313 * own block reservation window range.
315 * If the reservation window is outside the goal allocation group, return 0;
316 * grp_goal (given goal block) could be -1, which means no specific
317 * goal block. In this case, always return 1.
318 * If the goal block is within the reservation window, return 1;
319 * otherwise, return 0;
322 goal_in_my_reservation(struct ext4_reserve_window *rsv, ext4_grpblk_t grp_goal,
323 ext4_group_t group, struct super_block *sb)
325 ext4_fsblk_t group_first_block, group_last_block;
327 group_first_block = ext4_group_first_block_no(sb, group);
328 group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1);
330 if ((rsv->_rsv_start > group_last_block) ||
331 (rsv->_rsv_end < group_first_block))
333 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
334 || (grp_goal + group_first_block > rsv->_rsv_end)))
340 * search_reserve_window()
341 * @rb_root: root of reservation tree
342 * @goal: target allocation block
344 * Find the reserved window which includes the goal, or the previous one
345 * if the goal is not in any window.
346 * Returns NULL if there are no windows or if all windows start after the goal.
348 static struct ext4_reserve_window_node *
349 search_reserve_window(struct rb_root *root, ext4_fsblk_t goal)
351 struct rb_node *n = root->rb_node;
352 struct ext4_reserve_window_node *rsv;
358 rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node);
360 if (goal < rsv->rsv_start)
362 else if (goal > rsv->rsv_end)
368 * We've fallen off the end of the tree: the goal wasn't inside
369 * any particular node. OK, the previous node must be to one
370 * side of the interval containing the goal. If it's the RHS,
371 * we need to back up one.
373 if (rsv->rsv_start > goal) {
374 n = rb_prev(&rsv->rsv_node);
375 rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node);
381 * ext4_rsv_window_add() -- Insert a window to the block reservation rb tree.
383 * @rsv: reservation window to add
385 * Must be called with rsv_lock hold.
387 void ext4_rsv_window_add(struct super_block *sb,
388 struct ext4_reserve_window_node *rsv)
390 struct rb_root *root = &EXT4_SB(sb)->s_rsv_window_root;
391 struct rb_node *node = &rsv->rsv_node;
392 ext4_fsblk_t start = rsv->rsv_start;
394 struct rb_node ** p = &root->rb_node;
395 struct rb_node * parent = NULL;
396 struct ext4_reserve_window_node *this;
401 this = rb_entry(parent, struct ext4_reserve_window_node, rsv_node);
403 if (start < this->rsv_start)
405 else if (start > this->rsv_end)
408 rsv_window_dump(root, 1);
413 rb_link_node(node, parent, p);
414 rb_insert_color(node, root);
418 * ext4_rsv_window_remove() -- unlink a window from the reservation rb tree
420 * @rsv: reservation window to remove
422 * Mark the block reservation window as not allocated, and unlink it
423 * from the filesystem reservation window rb tree. Must be called with
426 static void rsv_window_remove(struct super_block *sb,
427 struct ext4_reserve_window_node *rsv)
429 rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
430 rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
431 rsv->rsv_alloc_hit = 0;
432 rb_erase(&rsv->rsv_node, &EXT4_SB(sb)->s_rsv_window_root);
436 * rsv_is_empty() -- Check if the reservation window is allocated.
437 * @rsv: given reservation window to check
439 * returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED.
441 static inline int rsv_is_empty(struct ext4_reserve_window *rsv)
443 /* a valid reservation end block could not be 0 */
444 return rsv->_rsv_end == EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
448 * ext4_init_block_alloc_info()
449 * @inode: file inode structure
451 * Allocate and initialize the reservation window structure, and
452 * link the window to the ext4 inode structure at last
454 * The reservation window structure is only dynamically allocated
455 * and linked to ext4 inode the first time the open file
456 * needs a new block. So, before every ext4_new_block(s) call, for
457 * regular files, we should check whether the reservation window
458 * structure exists or not. In the latter case, this function is called.
459 * Fail to do so will result in block reservation being turned off for that
462 * This function is called from ext4_get_blocks_handle(), also called
463 * when setting the reservation window size through ioctl before the file
464 * is open for write (needs block allocation).
466 * Needs truncate_mutex protection prior to call this function.
468 void ext4_init_block_alloc_info(struct inode *inode)
470 struct ext4_inode_info *ei = EXT4_I(inode);
471 struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info;
472 struct super_block *sb = inode->i_sb;
474 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
476 struct ext4_reserve_window_node *rsv = &block_i->rsv_window_node;
478 rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
479 rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
482 * if filesystem is mounted with NORESERVATION, the goal
483 * reservation window size is set to zero to indicate
484 * block reservation is off
486 if (!test_opt(sb, RESERVATION))
487 rsv->rsv_goal_size = 0;
489 rsv->rsv_goal_size = EXT4_DEFAULT_RESERVE_BLOCKS;
490 rsv->rsv_alloc_hit = 0;
491 block_i->last_alloc_logical_block = 0;
492 block_i->last_alloc_physical_block = 0;
494 ei->i_block_alloc_info = block_i;
498 * ext4_discard_reservation()
501 * Discard(free) block reservation window on last file close, or truncate
504 * It is being called in three cases:
505 * ext4_release_file(): last writer close the file
506 * ext4_clear_inode(): last iput(), when nobody link to this file.
507 * ext4_truncate(): when the block indirect map is about to change.
510 void ext4_discard_reservation(struct inode *inode)
512 struct ext4_inode_info *ei = EXT4_I(inode);
513 struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info;
514 struct ext4_reserve_window_node *rsv;
515 spinlock_t *rsv_lock = &EXT4_SB(inode->i_sb)->s_rsv_window_lock;
520 rsv = &block_i->rsv_window_node;
521 if (!rsv_is_empty(&rsv->rsv_window)) {
523 if (!rsv_is_empty(&rsv->rsv_window))
524 rsv_window_remove(inode->i_sb, rsv);
525 spin_unlock(rsv_lock);
530 * ext4_free_blocks_sb() -- Free given blocks and update quota
531 * @handle: handle to this transaction
533 * @block: start physcial block to free
534 * @count: number of blocks to free
535 * @pdquot_freed_blocks: pointer to quota
537 void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb,
538 ext4_fsblk_t block, unsigned long count,
539 unsigned long *pdquot_freed_blocks)
541 struct buffer_head *bitmap_bh = NULL;
542 struct buffer_head *gd_bh;
543 ext4_group_t block_group;
546 unsigned long overflow;
547 struct ext4_group_desc * desc;
548 struct ext4_super_block * es;
549 struct ext4_sb_info *sbi;
551 ext4_grpblk_t group_freed;
553 *pdquot_freed_blocks = 0;
556 if (block < le32_to_cpu(es->s_first_data_block) ||
557 block + count < block ||
558 block + count > ext4_blocks_count(es)) {
559 ext4_error (sb, "ext4_free_blocks",
560 "Freeing blocks not in datazone - "
561 "block = %llu, count = %lu", block, count);
565 ext4_debug ("freeing block(s) %llu-%llu\n", block, block + count - 1);
569 ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
571 * Check to see if we are freeing blocks across a group
574 if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
575 overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb);
579 bitmap_bh = read_block_bitmap(sb, block_group);
582 desc = ext4_get_group_desc (sb, block_group, &gd_bh);
586 if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
587 in_range(ext4_inode_bitmap(sb, desc), block, count) ||
588 in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
589 in_range(block + count - 1, ext4_inode_table(sb, desc),
590 sbi->s_itb_per_group))
591 ext4_error (sb, "ext4_free_blocks",
592 "Freeing blocks in system zones - "
593 "Block = %llu, count = %lu",
597 * We are about to start releasing blocks in the bitmap,
598 * so we need undo access.
600 /* @@@ check errors */
601 BUFFER_TRACE(bitmap_bh, "getting undo access");
602 err = ext4_journal_get_undo_access(handle, bitmap_bh);
607 * We are about to modify some metadata. Call the journal APIs
608 * to unshare ->b_data if a currently-committing transaction is
611 BUFFER_TRACE(gd_bh, "get_write_access");
612 err = ext4_journal_get_write_access(handle, gd_bh);
616 jbd_lock_bh_state(bitmap_bh);
618 for (i = 0, group_freed = 0; i < count; i++) {
620 * An HJ special. This is expensive...
622 #ifdef CONFIG_JBD2_DEBUG
623 jbd_unlock_bh_state(bitmap_bh);
625 struct buffer_head *debug_bh;
626 debug_bh = sb_find_get_block(sb, block + i);
628 BUFFER_TRACE(debug_bh, "Deleted!");
629 if (!bh2jh(bitmap_bh)->b_committed_data)
630 BUFFER_TRACE(debug_bh,
631 "No commited data in bitmap");
632 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
636 jbd_lock_bh_state(bitmap_bh);
638 if (need_resched()) {
639 jbd_unlock_bh_state(bitmap_bh);
641 jbd_lock_bh_state(bitmap_bh);
643 /* @@@ This prevents newly-allocated data from being
644 * freed and then reallocated within the same
647 * Ideally we would want to allow that to happen, but to
648 * do so requires making jbd2_journal_forget() capable of
649 * revoking the queued write of a data block, which
650 * implies blocking on the journal lock. *forget()
651 * cannot block due to truncate races.
653 * Eventually we can fix this by making jbd2_journal_forget()
654 * return a status indicating whether or not it was able
655 * to revoke the buffer. On successful revoke, it is
656 * safe not to set the allocation bit in the committed
657 * bitmap, because we know that there is no outstanding
658 * activity on the buffer any more and so it is safe to
661 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
662 J_ASSERT_BH(bitmap_bh,
663 bh2jh(bitmap_bh)->b_committed_data != NULL);
664 ext4_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
665 bh2jh(bitmap_bh)->b_committed_data);
668 * We clear the bit in the bitmap after setting the committed
669 * data bit, because this is the reverse order to that which
670 * the allocator uses.
672 BUFFER_TRACE(bitmap_bh, "clear bit");
673 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
674 bit + i, bitmap_bh->b_data)) {
675 jbd_unlock_bh_state(bitmap_bh);
676 ext4_error(sb, __FUNCTION__,
677 "bit already cleared for block %llu",
678 (ext4_fsblk_t)(block + i));
679 jbd_lock_bh_state(bitmap_bh);
680 BUFFER_TRACE(bitmap_bh, "bit already cleared");
685 jbd_unlock_bh_state(bitmap_bh);
687 spin_lock(sb_bgl_lock(sbi, block_group));
688 desc->bg_free_blocks_count =
689 cpu_to_le16(le16_to_cpu(desc->bg_free_blocks_count) +
691 desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
692 spin_unlock(sb_bgl_lock(sbi, block_group));
693 percpu_counter_add(&sbi->s_freeblocks_counter, count);
695 /* We dirtied the bitmap block */
696 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
697 err = ext4_journal_dirty_metadata(handle, bitmap_bh);
699 /* And the group descriptor block */
700 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
701 ret = ext4_journal_dirty_metadata(handle, gd_bh);
703 *pdquot_freed_blocks += group_freed;
705 if (overflow && !err) {
713 ext4_std_error(sb, err);
718 * ext4_free_blocks() -- Free given blocks and update quota
719 * @handle: handle for this transaction
721 * @block: start physical block to free
722 * @count: number of blocks to count
724 void ext4_free_blocks(handle_t *handle, struct inode *inode,
725 ext4_fsblk_t block, unsigned long count)
727 struct super_block * sb;
728 unsigned long dquot_freed_blocks;
732 printk ("ext4_free_blocks: nonexistent device");
735 ext4_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
736 if (dquot_freed_blocks)
737 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
742 * ext4_test_allocatable()
743 * @nr: given allocation block group
744 * @bh: bufferhead contains the bitmap of the given block group
746 * For ext4 allocations, we must not reuse any blocks which are
747 * allocated in the bitmap buffer's "last committed data" copy. This
748 * prevents deletes from freeing up the page for reuse until we have
749 * committed the delete transaction.
751 * If we didn't do this, then deleting something and reallocating it as
752 * data would allow the old block to be overwritten before the
753 * transaction committed (because we force data to disk before commit).
754 * This would lead to corruption if we crashed between overwriting the
755 * data and committing the delete.
757 * @@@ We may want to make this allocation behaviour conditional on
758 * data-writes at some point, and disable it for metadata allocations or
761 static int ext4_test_allocatable(ext4_grpblk_t nr, struct buffer_head *bh)
764 struct journal_head *jh = bh2jh(bh);
766 if (ext4_test_bit(nr, bh->b_data))
769 jbd_lock_bh_state(bh);
770 if (!jh->b_committed_data)
773 ret = !ext4_test_bit(nr, jh->b_committed_data);
774 jbd_unlock_bh_state(bh);
779 * bitmap_search_next_usable_block()
780 * @start: the starting block (group relative) of the search
781 * @bh: bufferhead contains the block group bitmap
782 * @maxblocks: the ending block (group relative) of the reservation
784 * The bitmap search --- search forward alternately through the actual
785 * bitmap on disk and the last-committed copy in journal, until we find a
786 * bit free in both bitmaps.
789 bitmap_search_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh,
790 ext4_grpblk_t maxblocks)
793 struct journal_head *jh = bh2jh(bh);
795 while (start < maxblocks) {
796 next = ext4_find_next_zero_bit(bh->b_data, maxblocks, start);
797 if (next >= maxblocks)
799 if (ext4_test_allocatable(next, bh))
801 jbd_lock_bh_state(bh);
802 if (jh->b_committed_data)
803 start = ext4_find_next_zero_bit(jh->b_committed_data,
805 jbd_unlock_bh_state(bh);
811 * find_next_usable_block()
812 * @start: the starting block (group relative) to find next
813 * allocatable block in bitmap.
814 * @bh: bufferhead contains the block group bitmap
815 * @maxblocks: the ending block (group relative) for the search
817 * Find an allocatable block in a bitmap. We honor both the bitmap and
818 * its last-committed copy (if that exists), and perform the "most
819 * appropriate allocation" algorithm of looking for a free block near
820 * the initial goal; then for a free byte somewhere in the bitmap; then
821 * for any free bit in the bitmap.
824 find_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh,
825 ext4_grpblk_t maxblocks)
827 ext4_grpblk_t here, next;
832 * The goal was occupied; search forward for a free
833 * block within the next XX blocks.
835 * end_goal is more or less random, but it has to be
836 * less than EXT4_BLOCKS_PER_GROUP. Aligning up to the
837 * next 64-bit boundary is simple..
839 ext4_grpblk_t end_goal = (start + 63) & ~63;
840 if (end_goal > maxblocks)
841 end_goal = maxblocks;
842 here = ext4_find_next_zero_bit(bh->b_data, end_goal, start);
843 if (here < end_goal && ext4_test_allocatable(here, bh))
845 ext4_debug("Bit not found near goal\n");
852 p = ((char *)bh->b_data) + (here >> 3);
853 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
854 next = (r - ((char *)bh->b_data)) << 3;
856 if (next < maxblocks && next >= start && ext4_test_allocatable(next, bh))
860 * The bitmap search --- search forward alternately through the actual
861 * bitmap and the last-committed copy until we find a bit free in
864 here = bitmap_search_next_usable_block(here, bh, maxblocks);
870 * @block: the free block (group relative) to allocate
871 * @bh: the bufferhead containts the block group bitmap
873 * We think we can allocate this block in this bitmap. Try to set the bit.
874 * If that succeeds then check that nobody has allocated and then freed the
875 * block since we saw that is was not marked in b_committed_data. If it _was_
876 * allocated and freed then clear the bit in the bitmap again and return
880 claim_block(spinlock_t *lock, ext4_grpblk_t block, struct buffer_head *bh)
882 struct journal_head *jh = bh2jh(bh);
885 if (ext4_set_bit_atomic(lock, block, bh->b_data))
887 jbd_lock_bh_state(bh);
888 if (jh->b_committed_data && ext4_test_bit(block,jh->b_committed_data)) {
889 ext4_clear_bit_atomic(lock, block, bh->b_data);
894 jbd_unlock_bh_state(bh);
899 * ext4_try_to_allocate()
901 * @handle: handle to this transaction
902 * @group: given allocation block group
903 * @bitmap_bh: bufferhead holds the block bitmap
904 * @grp_goal: given target block within the group
905 * @count: target number of blocks to allocate
906 * @my_rsv: reservation window
908 * Attempt to allocate blocks within a give range. Set the range of allocation
909 * first, then find the first free bit(s) from the bitmap (within the range),
910 * and at last, allocate the blocks by claiming the found free bit as allocated.
912 * To set the range of this allocation:
913 * if there is a reservation window, only try to allocate block(s) from the
914 * file's own reservation window;
915 * Otherwise, the allocation range starts from the give goal block, ends at
916 * the block group's last block.
918 * If we failed to allocate the desired block then we may end up crossing to a
919 * new bitmap. In that case we must release write access to the old one via
920 * ext4_journal_release_buffer(), else we'll run out of credits.
923 ext4_try_to_allocate(struct super_block *sb, handle_t *handle,
924 ext4_group_t group, struct buffer_head *bitmap_bh,
925 ext4_grpblk_t grp_goal, unsigned long *count,
926 struct ext4_reserve_window *my_rsv)
928 ext4_fsblk_t group_first_block;
929 ext4_grpblk_t start, end;
930 unsigned long num = 0;
932 /* we do allocation within the reservation window if we have a window */
934 group_first_block = ext4_group_first_block_no(sb, group);
935 if (my_rsv->_rsv_start >= group_first_block)
936 start = my_rsv->_rsv_start - group_first_block;
938 /* reservation window cross group boundary */
940 end = my_rsv->_rsv_end - group_first_block + 1;
941 if (end > EXT4_BLOCKS_PER_GROUP(sb))
942 /* reservation window crosses group boundary */
943 end = EXT4_BLOCKS_PER_GROUP(sb);
944 if ((start <= grp_goal) && (grp_goal < end))
953 end = EXT4_BLOCKS_PER_GROUP(sb);
956 BUG_ON(start > EXT4_BLOCKS_PER_GROUP(sb));
959 if (grp_goal < 0 || !ext4_test_allocatable(grp_goal, bitmap_bh)) {
960 grp_goal = find_next_usable_block(start, bitmap_bh, end);
966 for (i = 0; i < 7 && grp_goal > start &&
967 ext4_test_allocatable(grp_goal - 1,
975 if (!claim_block(sb_bgl_lock(EXT4_SB(sb), group),
976 grp_goal, bitmap_bh)) {
978 * The block was allocated by another thread, or it was
979 * allocated and then freed by another thread
989 while (num < *count && grp_goal < end
990 && ext4_test_allocatable(grp_goal, bitmap_bh)
991 && claim_block(sb_bgl_lock(EXT4_SB(sb), group),
992 grp_goal, bitmap_bh)) {
997 return grp_goal - num;
1004 * find_next_reservable_window():
1005 * find a reservable space within the given range.
1006 * It does not allocate the reservation window for now:
1007 * alloc_new_reservation() will do the work later.
1009 * @search_head: the head of the searching list;
1010 * This is not necessarily the list head of the whole filesystem
1012 * We have both head and start_block to assist the search
1013 * for the reservable space. The list starts from head,
1014 * but we will shift to the place where start_block is,
1015 * then start from there, when looking for a reservable space.
1017 * @size: the target new reservation window size
1019 * @group_first_block: the first block we consider to start
1020 * the real search from
1023 * the maximum block number that our goal reservable space
1024 * could start from. This is normally the last block in this
1025 * group. The search will end when we found the start of next
1026 * possible reservable space is out of this boundary.
1027 * This could handle the cross boundary reservation window
1030 * basically we search from the given range, rather than the whole
1031 * reservation double linked list, (start_block, last_block)
1032 * to find a free region that is of my size and has not
1036 static int find_next_reservable_window(
1037 struct ext4_reserve_window_node *search_head,
1038 struct ext4_reserve_window_node *my_rsv,
1039 struct super_block * sb,
1040 ext4_fsblk_t start_block,
1041 ext4_fsblk_t last_block)
1043 struct rb_node *next;
1044 struct ext4_reserve_window_node *rsv, *prev;
1046 int size = my_rsv->rsv_goal_size;
1048 /* TODO: make the start of the reservation window byte-aligned */
1049 /* cur = *start_block & ~7;*/
1056 if (cur <= rsv->rsv_end)
1057 cur = rsv->rsv_end + 1;
1060 * in the case we could not find a reservable space
1061 * that is what is expected, during the re-search, we could
1062 * remember what's the largest reservable space we could have
1063 * and return that one.
1065 * For now it will fail if we could not find the reservable
1066 * space with expected-size (or more)...
1068 if (cur > last_block)
1069 return -1; /* fail */
1072 next = rb_next(&rsv->rsv_node);
1073 rsv = rb_entry(next,struct ext4_reserve_window_node,rsv_node);
1076 * Reached the last reservation, we can just append to the
1082 if (cur + size <= rsv->rsv_start) {
1084 * Found a reserveable space big enough. We could
1085 * have a reservation across the group boundary here
1091 * we come here either :
1092 * when we reach the end of the whole list,
1093 * and there is empty reservable space after last entry in the list.
1094 * append it to the end of the list.
1096 * or we found one reservable space in the middle of the list,
1097 * return the reservation window that we could append to.
1101 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
1102 rsv_window_remove(sb, my_rsv);
1105 * Let's book the whole avaliable window for now. We will check the
1106 * disk bitmap later and then, if there are free blocks then we adjust
1107 * the window size if it's larger than requested.
1108 * Otherwise, we will remove this node from the tree next time
1109 * call find_next_reservable_window.
1111 my_rsv->rsv_start = cur;
1112 my_rsv->rsv_end = cur + size - 1;
1113 my_rsv->rsv_alloc_hit = 0;
1116 ext4_rsv_window_add(sb, my_rsv);
1122 * alloc_new_reservation()--allocate a new reservation window
1124 * To make a new reservation, we search part of the filesystem
1125 * reservation list (the list that inside the group). We try to
1126 * allocate a new reservation window near the allocation goal,
1127 * or the beginning of the group, if there is no goal.
1129 * We first find a reservable space after the goal, then from
1130 * there, we check the bitmap for the first free block after
1131 * it. If there is no free block until the end of group, then the
1132 * whole group is full, we failed. Otherwise, check if the free
1133 * block is inside the expected reservable space, if so, we
1135 * If the first free block is outside the reservable space, then
1136 * start from the first free block, we search for next available
1139 * on succeed, a new reservation will be found and inserted into the list
1140 * It contains at least one free block, and it does not overlap with other
1141 * reservation windows.
1143 * failed: we failed to find a reservation window in this group
1145 * @rsv: the reservation
1147 * @grp_goal: The goal (group-relative). It is where the search for a
1148 * free reservable space should start from.
1149 * if we have a grp_goal(grp_goal >0 ), then start from there,
1150 * no grp_goal(grp_goal = -1), we start from the first block
1153 * @sb: the super block
1154 * @group: the group we are trying to allocate in
1155 * @bitmap_bh: the block group block bitmap
1158 static int alloc_new_reservation(struct ext4_reserve_window_node *my_rsv,
1159 ext4_grpblk_t grp_goal, struct super_block *sb,
1160 ext4_group_t group, struct buffer_head *bitmap_bh)
1162 struct ext4_reserve_window_node *search_head;
1163 ext4_fsblk_t group_first_block, group_end_block, start_block;
1164 ext4_grpblk_t first_free_block;
1165 struct rb_root *fs_rsv_root = &EXT4_SB(sb)->s_rsv_window_root;
1168 spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock;
1170 group_first_block = ext4_group_first_block_no(sb, group);
1171 group_end_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1174 start_block = group_first_block;
1176 start_block = grp_goal + group_first_block;
1178 size = my_rsv->rsv_goal_size;
1180 if (!rsv_is_empty(&my_rsv->rsv_window)) {
1182 * if the old reservation is cross group boundary
1183 * and if the goal is inside the old reservation window,
1184 * we will come here when we just failed to allocate from
1185 * the first part of the window. We still have another part
1186 * that belongs to the next group. In this case, there is no
1187 * point to discard our window and try to allocate a new one
1188 * in this group(which will fail). we should
1189 * keep the reservation window, just simply move on.
1191 * Maybe we could shift the start block of the reservation
1192 * window to the first block of next group.
1195 if ((my_rsv->rsv_start <= group_end_block) &&
1196 (my_rsv->rsv_end > group_end_block) &&
1197 (start_block >= my_rsv->rsv_start))
1200 if ((my_rsv->rsv_alloc_hit >
1201 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
1203 * if the previously allocation hit ratio is
1204 * greater than 1/2, then we double the size of
1205 * the reservation window the next time,
1206 * otherwise we keep the same size window
1209 if (size > EXT4_MAX_RESERVE_BLOCKS)
1210 size = EXT4_MAX_RESERVE_BLOCKS;
1211 my_rsv->rsv_goal_size= size;
1215 spin_lock(rsv_lock);
1217 * shift the search start to the window near the goal block
1219 search_head = search_reserve_window(fs_rsv_root, start_block);
1222 * find_next_reservable_window() simply finds a reservable window
1223 * inside the given range(start_block, group_end_block).
1225 * To make sure the reservation window has a free bit inside it, we
1226 * need to check the bitmap after we found a reservable window.
1229 ret = find_next_reservable_window(search_head, my_rsv, sb,
1230 start_block, group_end_block);
1233 if (!rsv_is_empty(&my_rsv->rsv_window))
1234 rsv_window_remove(sb, my_rsv);
1235 spin_unlock(rsv_lock);
1240 * On success, find_next_reservable_window() returns the
1241 * reservation window where there is a reservable space after it.
1242 * Before we reserve this reservable space, we need
1243 * to make sure there is at least a free block inside this region.
1245 * searching the first free bit on the block bitmap and copy of
1246 * last committed bitmap alternatively, until we found a allocatable
1247 * block. Search start from the start block of the reservable space
1250 spin_unlock(rsv_lock);
1251 first_free_block = bitmap_search_next_usable_block(
1252 my_rsv->rsv_start - group_first_block,
1253 bitmap_bh, group_end_block - group_first_block + 1);
1255 if (first_free_block < 0) {
1257 * no free block left on the bitmap, no point
1258 * to reserve the space. return failed.
1260 spin_lock(rsv_lock);
1261 if (!rsv_is_empty(&my_rsv->rsv_window))
1262 rsv_window_remove(sb, my_rsv);
1263 spin_unlock(rsv_lock);
1264 return -1; /* failed */
1267 start_block = first_free_block + group_first_block;
1269 * check if the first free block is within the
1270 * free space we just reserved
1272 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1273 return 0; /* success */
1275 * if the first free bit we found is out of the reservable space
1276 * continue search for next reservable space,
1277 * start from where the free block is,
1278 * we also shift the list head to where we stopped last time
1280 search_head = my_rsv;
1281 spin_lock(rsv_lock);
1286 * try_to_extend_reservation()
1287 * @my_rsv: given reservation window
1289 * @size: the delta to extend
1291 * Attempt to expand the reservation window large enough to have
1292 * required number of free blocks
1294 * Since ext4_try_to_allocate() will always allocate blocks within
1295 * the reservation window range, if the window size is too small,
1296 * multiple blocks allocation has to stop at the end of the reservation
1297 * window. To make this more efficient, given the total number of
1298 * blocks needed and the current size of the window, we try to
1299 * expand the reservation window size if necessary on a best-effort
1300 * basis before ext4_new_blocks() tries to allocate blocks,
1302 static void try_to_extend_reservation(struct ext4_reserve_window_node *my_rsv,
1303 struct super_block *sb, int size)
1305 struct ext4_reserve_window_node *next_rsv;
1306 struct rb_node *next;
1307 spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock;
1309 if (!spin_trylock(rsv_lock))
1312 next = rb_next(&my_rsv->rsv_node);
1315 my_rsv->rsv_end += size;
1317 next_rsv = rb_entry(next, struct ext4_reserve_window_node, rsv_node);
1319 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1320 my_rsv->rsv_end += size;
1322 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1324 spin_unlock(rsv_lock);
1328 * ext4_try_to_allocate_with_rsv()
1330 * @handle: handle to this transaction
1331 * @group: given allocation block group
1332 * @bitmap_bh: bufferhead holds the block bitmap
1333 * @grp_goal: given target block within the group
1334 * @count: target number of blocks to allocate
1335 * @my_rsv: reservation window
1336 * @errp: pointer to store the error code
1338 * This is the main function used to allocate a new block and its reservation
1341 * Each time when a new block allocation is need, first try to allocate from
1342 * its own reservation. If it does not have a reservation window, instead of
1343 * looking for a free bit on bitmap first, then look up the reservation list to
1344 * see if it is inside somebody else's reservation window, we try to allocate a
1345 * reservation window for it starting from the goal first. Then do the block
1346 * allocation within the reservation window.
1348 * This will avoid keeping on searching the reservation list again and
1349 * again when somebody is looking for a free block (without
1350 * reservation), and there are lots of free blocks, but they are all
1353 * We use a red-black tree for the per-filesystem reservation list.
1356 static ext4_grpblk_t
1357 ext4_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1358 ext4_group_t group, struct buffer_head *bitmap_bh,
1359 ext4_grpblk_t grp_goal,
1360 struct ext4_reserve_window_node * my_rsv,
1361 unsigned long *count, int *errp)
1363 ext4_fsblk_t group_first_block, group_last_block;
1364 ext4_grpblk_t ret = 0;
1366 unsigned long num = *count;
1371 * Make sure we use undo access for the bitmap, because it is critical
1372 * that we do the frozen_data COW on bitmap buffers in all cases even
1373 * if the buffer is in BJ_Forget state in the committing transaction.
1375 BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1376 fatal = ext4_journal_get_undo_access(handle, bitmap_bh);
1383 * we don't deal with reservation when
1384 * filesystem is mounted without reservation
1385 * or the file is not a regular file
1386 * or last attempt to allocate a block with reservation turned on failed
1388 if (my_rsv == NULL ) {
1389 ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh,
1390 grp_goal, count, NULL);
1394 * grp_goal is a group relative block number (if there is a goal)
1395 * 0 <= grp_goal < EXT4_BLOCKS_PER_GROUP(sb)
1396 * first block is a filesystem wide block number
1397 * first block is the block number of the first block in this group
1399 group_first_block = ext4_group_first_block_no(sb, group);
1400 group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1403 * Basically we will allocate a new block from inode's reservation
1406 * We need to allocate a new reservation window, if:
1407 * a) inode does not have a reservation window; or
1408 * b) last attempt to allocate a block from existing reservation
1410 * c) we come here with a goal and with a reservation window
1412 * We do not need to allocate a new reservation window if we come here
1413 * at the beginning with a goal and the goal is inside the window, or
1414 * we don't have a goal but already have a reservation window.
1415 * then we could go to allocate from the reservation window directly.
1418 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1419 !goal_in_my_reservation(&my_rsv->rsv_window,
1420 grp_goal, group, sb)) {
1421 if (my_rsv->rsv_goal_size < *count)
1422 my_rsv->rsv_goal_size = *count;
1423 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1428 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1429 grp_goal, group, sb))
1431 } else if (grp_goal >= 0) {
1432 int curr = my_rsv->rsv_end -
1433 (grp_goal + group_first_block) + 1;
1436 try_to_extend_reservation(my_rsv, sb,
1440 if ((my_rsv->rsv_start > group_last_block) ||
1441 (my_rsv->rsv_end < group_first_block)) {
1442 rsv_window_dump(&EXT4_SB(sb)->s_rsv_window_root, 1);
1445 ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh,
1446 grp_goal, &num, &my_rsv->rsv_window);
1448 my_rsv->rsv_alloc_hit += num;
1450 break; /* succeed */
1456 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1458 fatal = ext4_journal_dirty_metadata(handle, bitmap_bh);
1466 BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1467 ext4_journal_release_buffer(handle, bitmap_bh);
1472 * ext4_has_free_blocks()
1473 * @sbi: in-core super block structure.
1475 * Check if filesystem has at least 1 free block available for allocation.
1477 static int ext4_has_free_blocks(struct ext4_sb_info *sbi)
1479 ext4_fsblk_t free_blocks, root_blocks;
1481 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1482 root_blocks = ext4_r_blocks_count(sbi->s_es);
1483 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1484 sbi->s_resuid != current->fsuid &&
1485 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1492 * ext4_should_retry_alloc()
1494 * @retries number of attemps has been made
1496 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
1497 * it is profitable to retry the operation, this function will wait
1498 * for the current or commiting transaction to complete, and then
1501 * if the total number of retries exceed three times, return FALSE.
1503 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
1505 if (!ext4_has_free_blocks(EXT4_SB(sb)) || (*retries)++ > 3)
1508 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1510 return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
1514 * ext4_new_blocks() -- core block(s) allocation function
1515 * @handle: handle to this transaction
1516 * @inode: file inode
1517 * @goal: given target block(filesystem wide)
1518 * @count: target number of blocks to allocate
1521 * ext4_new_blocks uses a goal block to assist allocation. It tries to
1522 * allocate block(s) from the block group contains the goal block first. If that
1523 * fails, it will try to allocate block(s) from other block groups without
1524 * any specific goal block.
1527 ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode,
1528 ext4_fsblk_t goal, unsigned long *count, int *errp)
1530 struct buffer_head *bitmap_bh = NULL;
1531 struct buffer_head *gdp_bh;
1532 ext4_group_t group_no;
1533 ext4_group_t goal_group;
1534 ext4_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1535 ext4_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1536 ext4_fsblk_t ret_block; /* filesyetem-wide allocated block */
1537 ext4_group_t bgi; /* blockgroup iteration index */
1539 int performed_allocation = 0;
1540 ext4_grpblk_t free_blocks; /* number of free blocks in a group */
1541 struct super_block *sb;
1542 struct ext4_group_desc *gdp;
1543 struct ext4_super_block *es;
1544 struct ext4_sb_info *sbi;
1545 struct ext4_reserve_window_node *my_rsv = NULL;
1546 struct ext4_block_alloc_info *block_i;
1547 unsigned short windowsz = 0;
1548 ext4_group_t ngroups;
1549 unsigned long num = *count;
1554 printk("ext4_new_block: nonexistent device");
1559 * Check quota for allocation of this block.
1561 if (DQUOT_ALLOC_BLOCK(inode, num)) {
1567 es = EXT4_SB(sb)->s_es;
1568 ext4_debug("goal=%lu.\n", goal);
1570 * Allocate a block from reservation only when
1571 * filesystem is mounted with reservation(default,-o reservation), and
1572 * it's a regular file, and
1573 * the desired window size is greater than 0 (One could use ioctl
1574 * command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off
1575 * reservation on that particular file)
1577 block_i = EXT4_I(inode)->i_block_alloc_info;
1578 if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1579 my_rsv = &block_i->rsv_window_node;
1581 if (!ext4_has_free_blocks(sbi)) {
1587 * First, test whether the goal block is free.
1589 if (goal < le32_to_cpu(es->s_first_data_block) ||
1590 goal >= ext4_blocks_count(es))
1591 goal = le32_to_cpu(es->s_first_data_block);
1592 ext4_get_group_no_and_offset(sb, goal, &group_no, &grp_target_blk);
1593 goal_group = group_no;
1595 gdp = ext4_get_group_desc(sb, group_no, &gdp_bh);
1599 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1601 * if there is not enough free blocks to make a new resevation
1602 * turn off reservation for this allocation
1604 if (my_rsv && (free_blocks < windowsz)
1605 && (rsv_is_empty(&my_rsv->rsv_window)))
1608 if (free_blocks > 0) {
1609 bitmap_bh = read_block_bitmap(sb, group_no);
1612 grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle,
1613 group_no, bitmap_bh, grp_target_blk,
1614 my_rsv, &num, &fatal);
1617 if (grp_alloc_blk >= 0)
1621 ngroups = EXT4_SB(sb)->s_groups_count;
1625 * Now search the rest of the groups. We assume that
1626 * i and gdp correctly point to the last group visited.
1628 for (bgi = 0; bgi < ngroups; bgi++) {
1630 if (group_no >= ngroups)
1632 gdp = ext4_get_group_desc(sb, group_no, &gdp_bh);
1635 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1637 * skip this group if the number of
1638 * free blocks is less than half of the reservation
1641 if (free_blocks <= (windowsz/2))
1645 bitmap_bh = read_block_bitmap(sb, group_no);
1649 * try to allocate block(s) from this group, without a goal(-1).
1651 grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle,
1652 group_no, bitmap_bh, -1, my_rsv,
1656 if (grp_alloc_blk >= 0)
1660 * We may end up a bogus ealier ENOSPC error due to
1661 * filesystem is "full" of reservations, but
1662 * there maybe indeed free blocks avaliable on disk
1663 * In this case, we just forget about the reservations
1664 * just do block allocation as without reservations.
1669 group_no = goal_group;
1672 /* No space left on the device */
1678 ext4_debug("using block group %d(%d)\n",
1679 group_no, gdp->bg_free_blocks_count);
1681 BUFFER_TRACE(gdp_bh, "get_write_access");
1682 fatal = ext4_journal_get_write_access(handle, gdp_bh);
1686 ret_block = grp_alloc_blk + ext4_group_first_block_no(sb, group_no);
1688 if (in_range(ext4_block_bitmap(sb, gdp), ret_block, num) ||
1689 in_range(ext4_inode_bitmap(sb, gdp), ret_block, num) ||
1690 in_range(ret_block, ext4_inode_table(sb, gdp),
1691 EXT4_SB(sb)->s_itb_per_group) ||
1692 in_range(ret_block + num - 1, ext4_inode_table(sb, gdp),
1693 EXT4_SB(sb)->s_itb_per_group))
1694 ext4_error(sb, "ext4_new_block",
1695 "Allocating block in system zone - "
1696 "blocks from %llu, length %lu",
1699 performed_allocation = 1;
1701 #ifdef CONFIG_JBD2_DEBUG
1703 struct buffer_head *debug_bh;
1705 /* Record bitmap buffer state in the newly allocated block */
1706 debug_bh = sb_find_get_block(sb, ret_block);
1708 BUFFER_TRACE(debug_bh, "state when allocated");
1709 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1713 jbd_lock_bh_state(bitmap_bh);
1714 spin_lock(sb_bgl_lock(sbi, group_no));
1715 if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1718 for (i = 0; i < num; i++) {
1719 if (ext4_test_bit(grp_alloc_blk+i,
1720 bh2jh(bitmap_bh)->b_committed_data)) {
1721 printk("%s: block was unexpectedly set in "
1722 "b_committed_data\n", __FUNCTION__);
1726 ext4_debug("found bit %d\n", grp_alloc_blk);
1727 spin_unlock(sb_bgl_lock(sbi, group_no));
1728 jbd_unlock_bh_state(bitmap_bh);
1731 if (ret_block + num - 1 >= ext4_blocks_count(es)) {
1732 ext4_error(sb, "ext4_new_block",
1733 "block(%llu) >= blocks count(%llu) - "
1734 "block_group = %lu, es == %p ", ret_block,
1735 ext4_blocks_count(es), group_no, es);
1740 * It is up to the caller to add the new buffer to a journal
1741 * list of some description. We don't know in advance whether
1742 * the caller wants to use it as metadata or data.
1744 spin_lock(sb_bgl_lock(sbi, group_no));
1745 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))
1746 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1747 gdp->bg_free_blocks_count =
1748 cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count)-num);
1749 gdp->bg_checksum = ext4_group_desc_csum(sbi, group_no, gdp);
1750 spin_unlock(sb_bgl_lock(sbi, group_no));
1751 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1753 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1754 err = ext4_journal_dirty_metadata(handle, gdp_bh);
1764 DQUOT_FREE_BLOCK(inode, *count-num);
1773 ext4_std_error(sb, fatal);
1776 * Undo the block allocation
1778 if (!performed_allocation)
1779 DQUOT_FREE_BLOCK(inode, *count);
1784 ext4_fsblk_t ext4_new_block(handle_t *handle, struct inode *inode,
1785 ext4_fsblk_t goal, int *errp)
1787 unsigned long count = 1;
1789 return ext4_new_blocks(handle, inode, goal, &count, errp);
1793 * ext4_count_free_blocks() -- count filesystem free blocks
1796 * Adds up the number of free blocks from each block group.
1798 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
1800 ext4_fsblk_t desc_count;
1801 struct ext4_group_desc *gdp;
1803 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
1805 struct ext4_super_block *es;
1806 ext4_fsblk_t bitmap_count;
1808 struct buffer_head *bitmap_bh = NULL;
1810 es = EXT4_SB(sb)->s_es;
1816 for (i = 0; i < ngroups; i++) {
1817 gdp = ext4_get_group_desc(sb, i, NULL);
1820 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1822 bitmap_bh = read_block_bitmap(sb, i);
1823 if (bitmap_bh == NULL)
1826 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
1827 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1828 i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1832 printk("ext4_count_free_blocks: stored = %llu"
1833 ", computed = %llu, %llu\n",
1834 EXT4_FREE_BLOCKS_COUNT(es),
1835 desc_count, bitmap_count);
1836 return bitmap_count;
1840 for (i = 0; i < ngroups; i++) {
1841 gdp = ext4_get_group_desc(sb, i, NULL);
1844 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1851 static inline int test_root(ext4_group_t a, int b)
1860 static int ext4_group_sparse(ext4_group_t group)
1866 return (test_root(group, 7) || test_root(group, 5) ||
1867 test_root(group, 3));
1871 * ext4_bg_has_super - number of blocks used by the superblock in group
1872 * @sb: superblock for filesystem
1873 * @group: group number to check
1875 * Return the number of blocks used by the superblock (primary or backup)
1876 * in this group. Currently this will be only 0 or 1.
1878 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
1880 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1881 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1882 !ext4_group_sparse(group))
1887 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
1890 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
1891 ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
1892 ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
1894 if (group == first || group == first + 1 || group == last)
1899 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
1902 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1903 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1904 !ext4_group_sparse(group))
1906 return EXT4_SB(sb)->s_gdb_count;
1910 * ext4_bg_num_gdb - number of blocks used by the group table in group
1911 * @sb: superblock for filesystem
1912 * @group: group number to check
1914 * Return the number of blocks used by the group descriptor table
1915 * (primary or backup) in this group. In the future there may be a
1916 * different number of descriptor blocks in each group.
1918 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
1920 unsigned long first_meta_bg =
1921 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
1922 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
1924 if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
1925 metagroup < first_meta_bg)
1926 return ext4_bg_num_gdb_nometa(sb,group);
1928 return ext4_bg_num_gdb_meta(sb,group);