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>
22 #include <linux/vs_dlimit.h>
23 #include <linux/vs_tag.h>
26 * balloc.c contains the blocks allocation and deallocation routines
30 * Calculate the block group number and offset, given a block number
32 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
33 unsigned long *blockgrpp, ext4_grpblk_t *offsetp)
35 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
38 blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
39 offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
48 * The free blocks are managed by bitmaps. A file system contains several
49 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
50 * block for inodes, N blocks for the inode table and data blocks.
52 * The file system contains group descriptors which are located after the
53 * super block. Each descriptor contains the number of the bitmap block and
54 * the free blocks count in the block. The descriptors are loaded in memory
55 * when a file system is mounted (see ext4_read_super).
59 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
62 * ext4_get_group_desc() -- load group descriptor from disk
64 * @block_group: given block group
65 * @bh: pointer to the buffer head to store the block
68 struct ext4_group_desc * ext4_get_group_desc(struct super_block * sb,
69 unsigned int block_group,
70 struct buffer_head ** bh)
72 unsigned long group_desc;
74 struct ext4_group_desc * desc;
75 struct ext4_sb_info *sbi = EXT4_SB(sb);
77 if (block_group >= sbi->s_groups_count) {
78 ext4_error (sb, "ext4_get_group_desc",
79 "block_group >= groups_count - "
80 "block_group = %d, groups_count = %lu",
81 block_group, sbi->s_groups_count);
87 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
88 offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
89 if (!sbi->s_group_desc[group_desc]) {
90 ext4_error (sb, "ext4_get_group_desc",
91 "Group descriptor not loaded - "
92 "block_group = %d, group_desc = %lu, desc = %lu",
93 block_group, group_desc, offset);
97 desc = (struct ext4_group_desc *)(
98 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
99 offset * EXT4_DESC_SIZE(sb));
101 *bh = sbi->s_group_desc[group_desc];
106 * read_block_bitmap()
108 * @block_group: given block group
110 * Read the bitmap for a given block_group, reading into the specified
111 * slot in the superblock's bitmap cache.
113 * Return buffer_head on success or NULL in case of failure.
115 static struct buffer_head *
116 read_block_bitmap(struct super_block *sb, unsigned int block_group)
118 struct ext4_group_desc * desc;
119 struct buffer_head * bh = NULL;
121 desc = ext4_get_group_desc (sb, block_group, NULL);
124 bh = sb_bread(sb, ext4_block_bitmap(sb, desc));
126 ext4_error (sb, "read_block_bitmap",
127 "Cannot read block bitmap - "
128 "block_group = %d, block_bitmap = %llu",
130 ext4_block_bitmap(sb, desc));
135 * The reservation window structure operations
136 * --------------------------------------------
137 * Operations include:
138 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
140 * We use a red-black tree to represent per-filesystem reservation
146 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
147 * @rb_root: root of per-filesystem reservation rb tree
148 * @verbose: verbose mode
149 * @fn: function which wishes to dump the reservation map
151 * If verbose is turned on, it will print the whole block reservation
152 * windows(start, end). Otherwise, it will only print out the "bad" windows,
153 * those windows that overlap with their immediate neighbors.
156 static void __rsv_window_dump(struct rb_root *root, int verbose,
160 struct ext4_reserve_window_node *rsv, *prev;
168 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
170 rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node);
172 printk("reservation window 0x%p "
173 "start: %llu, end: %llu\n",
174 rsv, rsv->rsv_start, rsv->rsv_end);
175 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
176 printk("Bad reservation %p (start >= end)\n",
180 if (prev && prev->rsv_end >= rsv->rsv_start) {
181 printk("Bad reservation %p (prev->end >= start)\n",
187 printk("Restarting reservation walk in verbose mode\n");
195 printk("Window map complete.\n");
199 #define rsv_window_dump(root, verbose) \
200 __rsv_window_dump((root), (verbose), __FUNCTION__)
202 #define rsv_window_dump(root, verbose) do {} while (0)
206 * goal_in_my_reservation()
207 * @rsv: inode's reservation window
208 * @grp_goal: given goal block relative to the allocation block group
209 * @group: the current allocation block group
210 * @sb: filesystem super block
212 * Test if the given goal block (group relative) is within the file's
213 * own block reservation window range.
215 * If the reservation window is outside the goal allocation group, return 0;
216 * grp_goal (given goal block) could be -1, which means no specific
217 * goal block. In this case, always return 1.
218 * If the goal block is within the reservation window, return 1;
219 * otherwise, return 0;
222 goal_in_my_reservation(struct ext4_reserve_window *rsv, ext4_grpblk_t grp_goal,
223 unsigned int group, struct super_block * sb)
225 ext4_fsblk_t group_first_block, group_last_block;
227 group_first_block = ext4_group_first_block_no(sb, group);
228 group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1);
230 if ((rsv->_rsv_start > group_last_block) ||
231 (rsv->_rsv_end < group_first_block))
233 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
234 || (grp_goal + group_first_block > rsv->_rsv_end)))
240 * search_reserve_window()
241 * @rb_root: root of reservation tree
242 * @goal: target allocation block
244 * Find the reserved window which includes the goal, or the previous one
245 * if the goal is not in any window.
246 * Returns NULL if there are no windows or if all windows start after the goal.
248 static struct ext4_reserve_window_node *
249 search_reserve_window(struct rb_root *root, ext4_fsblk_t goal)
251 struct rb_node *n = root->rb_node;
252 struct ext4_reserve_window_node *rsv;
258 rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node);
260 if (goal < rsv->rsv_start)
262 else if (goal > rsv->rsv_end)
268 * We've fallen off the end of the tree: the goal wasn't inside
269 * any particular node. OK, the previous node must be to one
270 * side of the interval containing the goal. If it's the RHS,
271 * we need to back up one.
273 if (rsv->rsv_start > goal) {
274 n = rb_prev(&rsv->rsv_node);
275 rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node);
281 * ext4_rsv_window_add() -- Insert a window to the block reservation rb tree.
283 * @rsv: reservation window to add
285 * Must be called with rsv_lock hold.
287 void ext4_rsv_window_add(struct super_block *sb,
288 struct ext4_reserve_window_node *rsv)
290 struct rb_root *root = &EXT4_SB(sb)->s_rsv_window_root;
291 struct rb_node *node = &rsv->rsv_node;
292 ext4_fsblk_t start = rsv->rsv_start;
294 struct rb_node ** p = &root->rb_node;
295 struct rb_node * parent = NULL;
296 struct ext4_reserve_window_node *this;
301 this = rb_entry(parent, struct ext4_reserve_window_node, rsv_node);
303 if (start < this->rsv_start)
305 else if (start > this->rsv_end)
308 rsv_window_dump(root, 1);
313 rb_link_node(node, parent, p);
314 rb_insert_color(node, root);
318 * ext4_rsv_window_remove() -- unlink a window from the reservation rb tree
320 * @rsv: reservation window to remove
322 * Mark the block reservation window as not allocated, and unlink it
323 * from the filesystem reservation window rb tree. Must be called with
326 static void rsv_window_remove(struct super_block *sb,
327 struct ext4_reserve_window_node *rsv)
329 rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
330 rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
331 rsv->rsv_alloc_hit = 0;
332 rb_erase(&rsv->rsv_node, &EXT4_SB(sb)->s_rsv_window_root);
336 * rsv_is_empty() -- Check if the reservation window is allocated.
337 * @rsv: given reservation window to check
339 * returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED.
341 static inline int rsv_is_empty(struct ext4_reserve_window *rsv)
343 /* a valid reservation end block could not be 0 */
344 return rsv->_rsv_end == EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
348 * ext4_init_block_alloc_info()
349 * @inode: file inode structure
351 * Allocate and initialize the reservation window structure, and
352 * link the window to the ext4 inode structure at last
354 * The reservation window structure is only dynamically allocated
355 * and linked to ext4 inode the first time the open file
356 * needs a new block. So, before every ext4_new_block(s) call, for
357 * regular files, we should check whether the reservation window
358 * structure exists or not. In the latter case, this function is called.
359 * Fail to do so will result in block reservation being turned off for that
362 * This function is called from ext4_get_blocks_handle(), also called
363 * when setting the reservation window size through ioctl before the file
364 * is open for write (needs block allocation).
366 * Needs truncate_mutex protection prior to call this function.
368 void ext4_init_block_alloc_info(struct inode *inode)
370 struct ext4_inode_info *ei = EXT4_I(inode);
371 struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info;
372 struct super_block *sb = inode->i_sb;
374 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
376 struct ext4_reserve_window_node *rsv = &block_i->rsv_window_node;
378 rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
379 rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
382 * if filesystem is mounted with NORESERVATION, the goal
383 * reservation window size is set to zero to indicate
384 * block reservation is off
386 if (!test_opt(sb, RESERVATION))
387 rsv->rsv_goal_size = 0;
389 rsv->rsv_goal_size = EXT4_DEFAULT_RESERVE_BLOCKS;
390 rsv->rsv_alloc_hit = 0;
391 block_i->last_alloc_logical_block = 0;
392 block_i->last_alloc_physical_block = 0;
394 ei->i_block_alloc_info = block_i;
398 * ext4_discard_reservation()
401 * Discard(free) block reservation window on last file close, or truncate
404 * It is being called in three cases:
405 * ext4_release_file(): last writer close the file
406 * ext4_clear_inode(): last iput(), when nobody link to this file.
407 * ext4_truncate(): when the block indirect map is about to change.
410 void ext4_discard_reservation(struct inode *inode)
412 struct ext4_inode_info *ei = EXT4_I(inode);
413 struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info;
414 struct ext4_reserve_window_node *rsv;
415 spinlock_t *rsv_lock = &EXT4_SB(inode->i_sb)->s_rsv_window_lock;
420 rsv = &block_i->rsv_window_node;
421 if (!rsv_is_empty(&rsv->rsv_window)) {
423 if (!rsv_is_empty(&rsv->rsv_window))
424 rsv_window_remove(inode->i_sb, rsv);
425 spin_unlock(rsv_lock);
430 * ext4_free_blocks_sb() -- Free given blocks and update quota
431 * @handle: handle to this transaction
433 * @block: start physcial block to free
434 * @count: number of blocks to free
435 * @pdquot_freed_blocks: pointer to quota
437 void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb,
438 ext4_fsblk_t block, unsigned long count,
439 unsigned long *pdquot_freed_blocks)
441 struct buffer_head *bitmap_bh = NULL;
442 struct buffer_head *gd_bh;
443 unsigned long block_group;
446 unsigned long overflow;
447 struct ext4_group_desc * desc;
448 struct ext4_super_block * es;
449 struct ext4_sb_info *sbi;
451 ext4_grpblk_t group_freed;
453 *pdquot_freed_blocks = 0;
456 if (block < le32_to_cpu(es->s_first_data_block) ||
457 block + count < block ||
458 block + count > ext4_blocks_count(es)) {
459 ext4_error (sb, "ext4_free_blocks",
460 "Freeing blocks not in datazone - "
461 "block = %llu, count = %lu", block, count);
465 ext4_debug ("freeing block(s) %llu-%llu\n", block, block + count - 1);
469 ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
471 * Check to see if we are freeing blocks across a group
474 if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
475 overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb);
479 bitmap_bh = read_block_bitmap(sb, block_group);
482 desc = ext4_get_group_desc (sb, block_group, &gd_bh);
486 if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
487 in_range(ext4_inode_bitmap(sb, desc), block, count) ||
488 in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
489 in_range(block + count - 1, ext4_inode_table(sb, desc),
490 sbi->s_itb_per_group))
491 ext4_error (sb, "ext4_free_blocks",
492 "Freeing blocks in system zones - "
493 "Block = %llu, count = %lu",
497 * We are about to start releasing blocks in the bitmap,
498 * so we need undo access.
500 /* @@@ check errors */
501 BUFFER_TRACE(bitmap_bh, "getting undo access");
502 err = ext4_journal_get_undo_access(handle, bitmap_bh);
507 * We are about to modify some metadata. Call the journal APIs
508 * to unshare ->b_data if a currently-committing transaction is
511 BUFFER_TRACE(gd_bh, "get_write_access");
512 err = ext4_journal_get_write_access(handle, gd_bh);
516 jbd_lock_bh_state(bitmap_bh);
518 for (i = 0, group_freed = 0; i < count; i++) {
520 * An HJ special. This is expensive...
522 #ifdef CONFIG_JBD_DEBUG
523 jbd_unlock_bh_state(bitmap_bh);
525 struct buffer_head *debug_bh;
526 debug_bh = sb_find_get_block(sb, block + i);
528 BUFFER_TRACE(debug_bh, "Deleted!");
529 if (!bh2jh(bitmap_bh)->b_committed_data)
530 BUFFER_TRACE(debug_bh,
531 "No commited data in bitmap");
532 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
536 jbd_lock_bh_state(bitmap_bh);
538 if (need_resched()) {
539 jbd_unlock_bh_state(bitmap_bh);
541 jbd_lock_bh_state(bitmap_bh);
543 /* @@@ This prevents newly-allocated data from being
544 * freed and then reallocated within the same
547 * Ideally we would want to allow that to happen, but to
548 * do so requires making jbd2_journal_forget() capable of
549 * revoking the queued write of a data block, which
550 * implies blocking on the journal lock. *forget()
551 * cannot block due to truncate races.
553 * Eventually we can fix this by making jbd2_journal_forget()
554 * return a status indicating whether or not it was able
555 * to revoke the buffer. On successful revoke, it is
556 * safe not to set the allocation bit in the committed
557 * bitmap, because we know that there is no outstanding
558 * activity on the buffer any more and so it is safe to
561 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
562 J_ASSERT_BH(bitmap_bh,
563 bh2jh(bitmap_bh)->b_committed_data != NULL);
564 ext4_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
565 bh2jh(bitmap_bh)->b_committed_data);
568 * We clear the bit in the bitmap after setting the committed
569 * data bit, because this is the reverse order to that which
570 * the allocator uses.
572 BUFFER_TRACE(bitmap_bh, "clear bit");
573 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
574 bit + i, bitmap_bh->b_data)) {
575 jbd_unlock_bh_state(bitmap_bh);
576 ext4_error(sb, __FUNCTION__,
577 "bit already cleared for block %llu",
578 (ext4_fsblk_t)(block + i));
579 jbd_lock_bh_state(bitmap_bh);
580 BUFFER_TRACE(bitmap_bh, "bit already cleared");
585 jbd_unlock_bh_state(bitmap_bh);
587 spin_lock(sb_bgl_lock(sbi, block_group));
588 desc->bg_free_blocks_count =
589 cpu_to_le16(le16_to_cpu(desc->bg_free_blocks_count) +
591 spin_unlock(sb_bgl_lock(sbi, block_group));
592 percpu_counter_mod(&sbi->s_freeblocks_counter, count);
594 /* We dirtied the bitmap block */
595 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
596 err = ext4_journal_dirty_metadata(handle, bitmap_bh);
598 /* And the group descriptor block */
599 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
600 ret = ext4_journal_dirty_metadata(handle, gd_bh);
602 *pdquot_freed_blocks += group_freed;
604 if (overflow && !err) {
612 ext4_std_error(sb, err);
617 * ext4_free_blocks() -- Free given blocks and update quota
618 * @handle: handle for this transaction
620 * @block: start physical block to free
621 * @count: number of blocks to count
623 void ext4_free_blocks(handle_t *handle, struct inode *inode,
624 ext4_fsblk_t block, unsigned long count)
626 struct super_block * sb;
627 unsigned long dquot_freed_blocks;
631 printk ("ext4_free_blocks: nonexistent device");
634 ext4_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
635 if (dquot_freed_blocks) {
636 DLIMIT_FREE_BLOCK(inode, dquot_freed_blocks);
637 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
643 * ext4_test_allocatable()
644 * @nr: given allocation block group
645 * @bh: bufferhead contains the bitmap of the given block group
647 * For ext4 allocations, we must not reuse any blocks which are
648 * allocated in the bitmap buffer's "last committed data" copy. This
649 * prevents deletes from freeing up the page for reuse until we have
650 * committed the delete transaction.
652 * If we didn't do this, then deleting something and reallocating it as
653 * data would allow the old block to be overwritten before the
654 * transaction committed (because we force data to disk before commit).
655 * This would lead to corruption if we crashed between overwriting the
656 * data and committing the delete.
658 * @@@ We may want to make this allocation behaviour conditional on
659 * data-writes at some point, and disable it for metadata allocations or
662 static int ext4_test_allocatable(ext4_grpblk_t nr, struct buffer_head *bh)
665 struct journal_head *jh = bh2jh(bh);
667 if (ext4_test_bit(nr, bh->b_data))
670 jbd_lock_bh_state(bh);
671 if (!jh->b_committed_data)
674 ret = !ext4_test_bit(nr, jh->b_committed_data);
675 jbd_unlock_bh_state(bh);
680 * bitmap_search_next_usable_block()
681 * @start: the starting block (group relative) of the search
682 * @bh: bufferhead contains the block group bitmap
683 * @maxblocks: the ending block (group relative) of the reservation
685 * The bitmap search --- search forward alternately through the actual
686 * bitmap on disk and the last-committed copy in journal, until we find a
687 * bit free in both bitmaps.
690 bitmap_search_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh,
691 ext4_grpblk_t maxblocks)
694 struct journal_head *jh = bh2jh(bh);
696 while (start < maxblocks) {
697 next = ext4_find_next_zero_bit(bh->b_data, maxblocks, start);
698 if (next >= maxblocks)
700 if (ext4_test_allocatable(next, bh))
702 jbd_lock_bh_state(bh);
703 if (jh->b_committed_data)
704 start = ext4_find_next_zero_bit(jh->b_committed_data,
706 jbd_unlock_bh_state(bh);
712 * find_next_usable_block()
713 * @start: the starting block (group relative) to find next
714 * allocatable block in bitmap.
715 * @bh: bufferhead contains the block group bitmap
716 * @maxblocks: the ending block (group relative) for the search
718 * Find an allocatable block in a bitmap. We honor both the bitmap and
719 * its last-committed copy (if that exists), and perform the "most
720 * appropriate allocation" algorithm of looking for a free block near
721 * the initial goal; then for a free byte somewhere in the bitmap; then
722 * for any free bit in the bitmap.
725 find_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh,
726 ext4_grpblk_t maxblocks)
728 ext4_grpblk_t here, next;
733 * The goal was occupied; search forward for a free
734 * block within the next XX blocks.
736 * end_goal is more or less random, but it has to be
737 * less than EXT4_BLOCKS_PER_GROUP. Aligning up to the
738 * next 64-bit boundary is simple..
740 ext4_grpblk_t end_goal = (start + 63) & ~63;
741 if (end_goal > maxblocks)
742 end_goal = maxblocks;
743 here = ext4_find_next_zero_bit(bh->b_data, end_goal, start);
744 if (here < end_goal && ext4_test_allocatable(here, bh))
746 ext4_debug("Bit not found near goal\n");
753 p = ((char *)bh->b_data) + (here >> 3);
754 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
755 next = (r - ((char *)bh->b_data)) << 3;
757 if (next < maxblocks && next >= start && ext4_test_allocatable(next, bh))
761 * The bitmap search --- search forward alternately through the actual
762 * bitmap and the last-committed copy until we find a bit free in
765 here = bitmap_search_next_usable_block(here, bh, maxblocks);
771 * @block: the free block (group relative) to allocate
772 * @bh: the bufferhead containts the block group bitmap
774 * We think we can allocate this block in this bitmap. Try to set the bit.
775 * If that succeeds then check that nobody has allocated and then freed the
776 * block since we saw that is was not marked in b_committed_data. If it _was_
777 * allocated and freed then clear the bit in the bitmap again and return
781 claim_block(spinlock_t *lock, ext4_grpblk_t block, struct buffer_head *bh)
783 struct journal_head *jh = bh2jh(bh);
786 if (ext4_set_bit_atomic(lock, block, bh->b_data))
788 jbd_lock_bh_state(bh);
789 if (jh->b_committed_data && ext4_test_bit(block,jh->b_committed_data)) {
790 ext4_clear_bit_atomic(lock, block, bh->b_data);
795 jbd_unlock_bh_state(bh);
800 * ext4_try_to_allocate()
802 * @handle: handle to this transaction
803 * @group: given allocation block group
804 * @bitmap_bh: bufferhead holds the block bitmap
805 * @grp_goal: given target block within the group
806 * @count: target number of blocks to allocate
807 * @my_rsv: reservation window
809 * Attempt to allocate blocks within a give range. Set the range of allocation
810 * first, then find the first free bit(s) from the bitmap (within the range),
811 * and at last, allocate the blocks by claiming the found free bit as allocated.
813 * To set the range of this allocation:
814 * if there is a reservation window, only try to allocate block(s) from the
815 * file's own reservation window;
816 * Otherwise, the allocation range starts from the give goal block, ends at
817 * the block group's last block.
819 * If we failed to allocate the desired block then we may end up crossing to a
820 * new bitmap. In that case we must release write access to the old one via
821 * ext4_journal_release_buffer(), else we'll run out of credits.
824 ext4_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
825 struct buffer_head *bitmap_bh, ext4_grpblk_t grp_goal,
826 unsigned long *count, struct ext4_reserve_window *my_rsv)
828 ext4_fsblk_t group_first_block;
829 ext4_grpblk_t start, end;
830 unsigned long num = 0;
832 /* we do allocation within the reservation window if we have a window */
834 group_first_block = ext4_group_first_block_no(sb, group);
835 if (my_rsv->_rsv_start >= group_first_block)
836 start = my_rsv->_rsv_start - group_first_block;
838 /* reservation window cross group boundary */
840 end = my_rsv->_rsv_end - group_first_block + 1;
841 if (end > EXT4_BLOCKS_PER_GROUP(sb))
842 /* reservation window crosses group boundary */
843 end = EXT4_BLOCKS_PER_GROUP(sb);
844 if ((start <= grp_goal) && (grp_goal < end))
853 end = EXT4_BLOCKS_PER_GROUP(sb);
856 BUG_ON(start > EXT4_BLOCKS_PER_GROUP(sb));
859 if (grp_goal < 0 || !ext4_test_allocatable(grp_goal, bitmap_bh)) {
860 grp_goal = find_next_usable_block(start, bitmap_bh, end);
866 for (i = 0; i < 7 && grp_goal > start &&
867 ext4_test_allocatable(grp_goal - 1,
875 if (!claim_block(sb_bgl_lock(EXT4_SB(sb), group),
876 grp_goal, bitmap_bh)) {
878 * The block was allocated by another thread, or it was
879 * allocated and then freed by another thread
889 while (num < *count && grp_goal < end
890 && ext4_test_allocatable(grp_goal, bitmap_bh)
891 && claim_block(sb_bgl_lock(EXT4_SB(sb), group),
892 grp_goal, bitmap_bh)) {
897 return grp_goal - num;
904 * find_next_reservable_window():
905 * find a reservable space within the given range.
906 * It does not allocate the reservation window for now:
907 * alloc_new_reservation() will do the work later.
909 * @search_head: the head of the searching list;
910 * This is not necessarily the list head of the whole filesystem
912 * We have both head and start_block to assist the search
913 * for the reservable space. The list starts from head,
914 * but we will shift to the place where start_block is,
915 * then start from there, when looking for a reservable space.
917 * @size: the target new reservation window size
919 * @group_first_block: the first block we consider to start
920 * the real search from
923 * the maximum block number that our goal reservable space
924 * could start from. This is normally the last block in this
925 * group. The search will end when we found the start of next
926 * possible reservable space is out of this boundary.
927 * This could handle the cross boundary reservation window
930 * basically we search from the given range, rather than the whole
931 * reservation double linked list, (start_block, last_block)
932 * to find a free region that is of my size and has not
936 static int find_next_reservable_window(
937 struct ext4_reserve_window_node *search_head,
938 struct ext4_reserve_window_node *my_rsv,
939 struct super_block * sb,
940 ext4_fsblk_t start_block,
941 ext4_fsblk_t last_block)
943 struct rb_node *next;
944 struct ext4_reserve_window_node *rsv, *prev;
946 int size = my_rsv->rsv_goal_size;
948 /* TODO: make the start of the reservation window byte-aligned */
949 /* cur = *start_block & ~7;*/
956 if (cur <= rsv->rsv_end)
957 cur = rsv->rsv_end + 1;
960 * in the case we could not find a reservable space
961 * that is what is expected, during the re-search, we could
962 * remember what's the largest reservable space we could have
963 * and return that one.
965 * For now it will fail if we could not find the reservable
966 * space with expected-size (or more)...
968 if (cur > last_block)
969 return -1; /* fail */
972 next = rb_next(&rsv->rsv_node);
973 rsv = rb_entry(next,struct ext4_reserve_window_node,rsv_node);
976 * Reached the last reservation, we can just append to the
982 if (cur + size <= rsv->rsv_start) {
984 * Found a reserveable space big enough. We could
985 * have a reservation across the group boundary here
991 * we come here either :
992 * when we reach the end of the whole list,
993 * and there is empty reservable space after last entry in the list.
994 * append it to the end of the list.
996 * or we found one reservable space in the middle of the list,
997 * return the reservation window that we could append to.
1001 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
1002 rsv_window_remove(sb, my_rsv);
1005 * Let's book the whole avaliable window for now. We will check the
1006 * disk bitmap later and then, if there are free blocks then we adjust
1007 * the window size if it's larger than requested.
1008 * Otherwise, we will remove this node from the tree next time
1009 * call find_next_reservable_window.
1011 my_rsv->rsv_start = cur;
1012 my_rsv->rsv_end = cur + size - 1;
1013 my_rsv->rsv_alloc_hit = 0;
1016 ext4_rsv_window_add(sb, my_rsv);
1022 * alloc_new_reservation()--allocate a new reservation window
1024 * To make a new reservation, we search part of the filesystem
1025 * reservation list (the list that inside the group). We try to
1026 * allocate a new reservation window near the allocation goal,
1027 * or the beginning of the group, if there is no goal.
1029 * We first find a reservable space after the goal, then from
1030 * there, we check the bitmap for the first free block after
1031 * it. If there is no free block until the end of group, then the
1032 * whole group is full, we failed. Otherwise, check if the free
1033 * block is inside the expected reservable space, if so, we
1035 * If the first free block is outside the reservable space, then
1036 * start from the first free block, we search for next available
1039 * on succeed, a new reservation will be found and inserted into the list
1040 * It contains at least one free block, and it does not overlap with other
1041 * reservation windows.
1043 * failed: we failed to find a reservation window in this group
1045 * @rsv: the reservation
1047 * @grp_goal: The goal (group-relative). It is where the search for a
1048 * free reservable space should start from.
1049 * if we have a grp_goal(grp_goal >0 ), then start from there,
1050 * no grp_goal(grp_goal = -1), we start from the first block
1053 * @sb: the super block
1054 * @group: the group we are trying to allocate in
1055 * @bitmap_bh: the block group block bitmap
1058 static int alloc_new_reservation(struct ext4_reserve_window_node *my_rsv,
1059 ext4_grpblk_t grp_goal, struct super_block *sb,
1060 unsigned int group, struct buffer_head *bitmap_bh)
1062 struct ext4_reserve_window_node *search_head;
1063 ext4_fsblk_t group_first_block, group_end_block, start_block;
1064 ext4_grpblk_t first_free_block;
1065 struct rb_root *fs_rsv_root = &EXT4_SB(sb)->s_rsv_window_root;
1068 spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock;
1070 group_first_block = ext4_group_first_block_no(sb, group);
1071 group_end_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1074 start_block = group_first_block;
1076 start_block = grp_goal + group_first_block;
1078 size = my_rsv->rsv_goal_size;
1080 if (!rsv_is_empty(&my_rsv->rsv_window)) {
1082 * if the old reservation is cross group boundary
1083 * and if the goal is inside the old reservation window,
1084 * we will come here when we just failed to allocate from
1085 * the first part of the window. We still have another part
1086 * that belongs to the next group. In this case, there is no
1087 * point to discard our window and try to allocate a new one
1088 * in this group(which will fail). we should
1089 * keep the reservation window, just simply move on.
1091 * Maybe we could shift the start block of the reservation
1092 * window to the first block of next group.
1095 if ((my_rsv->rsv_start <= group_end_block) &&
1096 (my_rsv->rsv_end > group_end_block) &&
1097 (start_block >= my_rsv->rsv_start))
1100 if ((my_rsv->rsv_alloc_hit >
1101 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
1103 * if the previously allocation hit ratio is
1104 * greater than 1/2, then we double the size of
1105 * the reservation window the next time,
1106 * otherwise we keep the same size window
1109 if (size > EXT4_MAX_RESERVE_BLOCKS)
1110 size = EXT4_MAX_RESERVE_BLOCKS;
1111 my_rsv->rsv_goal_size= size;
1115 spin_lock(rsv_lock);
1117 * shift the search start to the window near the goal block
1119 search_head = search_reserve_window(fs_rsv_root, start_block);
1122 * find_next_reservable_window() simply finds a reservable window
1123 * inside the given range(start_block, group_end_block).
1125 * To make sure the reservation window has a free bit inside it, we
1126 * need to check the bitmap after we found a reservable window.
1129 ret = find_next_reservable_window(search_head, my_rsv, sb,
1130 start_block, group_end_block);
1133 if (!rsv_is_empty(&my_rsv->rsv_window))
1134 rsv_window_remove(sb, my_rsv);
1135 spin_unlock(rsv_lock);
1140 * On success, find_next_reservable_window() returns the
1141 * reservation window where there is a reservable space after it.
1142 * Before we reserve this reservable space, we need
1143 * to make sure there is at least a free block inside this region.
1145 * searching the first free bit on the block bitmap and copy of
1146 * last committed bitmap alternatively, until we found a allocatable
1147 * block. Search start from the start block of the reservable space
1150 spin_unlock(rsv_lock);
1151 first_free_block = bitmap_search_next_usable_block(
1152 my_rsv->rsv_start - group_first_block,
1153 bitmap_bh, group_end_block - group_first_block + 1);
1155 if (first_free_block < 0) {
1157 * no free block left on the bitmap, no point
1158 * to reserve the space. return failed.
1160 spin_lock(rsv_lock);
1161 if (!rsv_is_empty(&my_rsv->rsv_window))
1162 rsv_window_remove(sb, my_rsv);
1163 spin_unlock(rsv_lock);
1164 return -1; /* failed */
1167 start_block = first_free_block + group_first_block;
1169 * check if the first free block is within the
1170 * free space we just reserved
1172 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1173 return 0; /* success */
1175 * if the first free bit we found is out of the reservable space
1176 * continue search for next reservable space,
1177 * start from where the free block is,
1178 * we also shift the list head to where we stopped last time
1180 search_head = my_rsv;
1181 spin_lock(rsv_lock);
1186 * try_to_extend_reservation()
1187 * @my_rsv: given reservation window
1189 * @size: the delta to extend
1191 * Attempt to expand the reservation window large enough to have
1192 * required number of free blocks
1194 * Since ext4_try_to_allocate() will always allocate blocks within
1195 * the reservation window range, if the window size is too small,
1196 * multiple blocks allocation has to stop at the end of the reservation
1197 * window. To make this more efficient, given the total number of
1198 * blocks needed and the current size of the window, we try to
1199 * expand the reservation window size if necessary on a best-effort
1200 * basis before ext4_new_blocks() tries to allocate blocks,
1202 static void try_to_extend_reservation(struct ext4_reserve_window_node *my_rsv,
1203 struct super_block *sb, int size)
1205 struct ext4_reserve_window_node *next_rsv;
1206 struct rb_node *next;
1207 spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock;
1209 if (!spin_trylock(rsv_lock))
1212 next = rb_next(&my_rsv->rsv_node);
1215 my_rsv->rsv_end += size;
1217 next_rsv = rb_entry(next, struct ext4_reserve_window_node, rsv_node);
1219 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1220 my_rsv->rsv_end += size;
1222 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1224 spin_unlock(rsv_lock);
1228 * ext4_try_to_allocate_with_rsv()
1230 * @handle: handle to this transaction
1231 * @group: given allocation block group
1232 * @bitmap_bh: bufferhead holds the block bitmap
1233 * @grp_goal: given target block within the group
1234 * @count: target number of blocks to allocate
1235 * @my_rsv: reservation window
1236 * @errp: pointer to store the error code
1238 * This is the main function used to allocate a new block and its reservation
1241 * Each time when a new block allocation is need, first try to allocate from
1242 * its own reservation. If it does not have a reservation window, instead of
1243 * looking for a free bit on bitmap first, then look up the reservation list to
1244 * see if it is inside somebody else's reservation window, we try to allocate a
1245 * reservation window for it starting from the goal first. Then do the block
1246 * allocation within the reservation window.
1248 * This will avoid keeping on searching the reservation list again and
1249 * again when somebody is looking for a free block (without
1250 * reservation), and there are lots of free blocks, but they are all
1253 * We use a red-black tree for the per-filesystem reservation list.
1256 static ext4_grpblk_t
1257 ext4_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1258 unsigned int group, struct buffer_head *bitmap_bh,
1259 ext4_grpblk_t grp_goal,
1260 struct ext4_reserve_window_node * my_rsv,
1261 unsigned long *count, int *errp)
1263 ext4_fsblk_t group_first_block, group_last_block;
1264 ext4_grpblk_t ret = 0;
1266 unsigned long num = *count;
1271 * Make sure we use undo access for the bitmap, because it is critical
1272 * that we do the frozen_data COW on bitmap buffers in all cases even
1273 * if the buffer is in BJ_Forget state in the committing transaction.
1275 BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1276 fatal = ext4_journal_get_undo_access(handle, bitmap_bh);
1283 * we don't deal with reservation when
1284 * filesystem is mounted without reservation
1285 * or the file is not a regular file
1286 * or last attempt to allocate a block with reservation turned on failed
1288 if (my_rsv == NULL ) {
1289 ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh,
1290 grp_goal, count, NULL);
1294 * grp_goal is a group relative block number (if there is a goal)
1295 * 0 <= grp_goal < EXT4_BLOCKS_PER_GROUP(sb)
1296 * first block is a filesystem wide block number
1297 * first block is the block number of the first block in this group
1299 group_first_block = ext4_group_first_block_no(sb, group);
1300 group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1303 * Basically we will allocate a new block from inode's reservation
1306 * We need to allocate a new reservation window, if:
1307 * a) inode does not have a reservation window; or
1308 * b) last attempt to allocate a block from existing reservation
1310 * c) we come here with a goal and with a reservation window
1312 * We do not need to allocate a new reservation window if we come here
1313 * at the beginning with a goal and the goal is inside the window, or
1314 * we don't have a goal but already have a reservation window.
1315 * then we could go to allocate from the reservation window directly.
1318 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1319 !goal_in_my_reservation(&my_rsv->rsv_window,
1320 grp_goal, group, sb)) {
1321 if (my_rsv->rsv_goal_size < *count)
1322 my_rsv->rsv_goal_size = *count;
1323 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1328 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1329 grp_goal, group, sb))
1331 } else if (grp_goal >= 0) {
1332 int curr = my_rsv->rsv_end -
1333 (grp_goal + group_first_block) + 1;
1336 try_to_extend_reservation(my_rsv, sb,
1340 if ((my_rsv->rsv_start > group_last_block) ||
1341 (my_rsv->rsv_end < group_first_block)) {
1342 rsv_window_dump(&EXT4_SB(sb)->s_rsv_window_root, 1);
1345 ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh,
1346 grp_goal, &num, &my_rsv->rsv_window);
1348 my_rsv->rsv_alloc_hit += num;
1350 break; /* succeed */
1356 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1358 fatal = ext4_journal_dirty_metadata(handle, bitmap_bh);
1366 BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1367 ext4_journal_release_buffer(handle, bitmap_bh);
1372 * ext4_has_free_blocks()
1373 * @sbi: in-core super block structure.
1375 * Check if filesystem has at least 1 free block available for allocation.
1377 static int ext4_has_free_blocks(struct super_block *sb)
1379 struct ext4_sb_info *sbi = EXT4_SB(sb);
1380 ext4_fsblk_t free_blocks, root_blocks;
1383 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1384 root_blocks = ext4_r_blocks_count(sbi->s_es);
1386 vxdprintk(VXD_CBIT(dlim, 3),
1387 "ext4_has_free_blocks(%p): free=%llu, root=%llu",
1388 sb, free_blocks, root_blocks);
1390 DLIMIT_ADJUST_BLOCK(sb, dx_current_tag(), &free_blocks, &root_blocks);
1392 cond = (free_blocks < root_blocks + 1 &&
1393 !capable(CAP_SYS_RESOURCE) &&
1394 sbi->s_resuid != current->fsuid &&
1395 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid)));
1397 vxdprintk(VXD_CBIT(dlim, 3),
1398 "ext4_has_free_blocks(%p): %llu<%llu+1, %c, %u!=%u r=%d",
1399 sb, free_blocks, root_blocks,
1400 !capable(CAP_SYS_RESOURCE)?'1':'0',
1401 sbi->s_resuid, current->fsuid, cond?0:1);
1403 return (cond ? 0 : 1);
1407 * ext4_should_retry_alloc()
1409 * @retries number of attemps has been made
1411 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
1412 * it is profitable to retry the operation, this function will wait
1413 * for the current or commiting transaction to complete, and then
1416 * if the total number of retries exceed three times, return FALSE.
1418 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
1420 if (!ext4_has_free_blocks(sb) || (*retries)++ > 3)
1423 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1425 return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
1429 * ext4_new_blocks() -- core block(s) allocation function
1430 * @handle: handle to this transaction
1431 * @inode: file inode
1432 * @goal: given target block(filesystem wide)
1433 * @count: target number of blocks to allocate
1436 * ext4_new_blocks uses a goal block to assist allocation. It tries to
1437 * allocate block(s) from the block group contains the goal block first. If that
1438 * fails, it will try to allocate block(s) from other block groups without
1439 * any specific goal block.
1442 ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode,
1443 ext4_fsblk_t goal, unsigned long *count, int *errp)
1445 struct buffer_head *bitmap_bh = NULL;
1446 struct buffer_head *gdp_bh;
1447 unsigned long group_no;
1449 ext4_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1450 ext4_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1451 ext4_fsblk_t ret_block; /* filesyetem-wide allocated block */
1452 int bgi; /* blockgroup iteration index */
1454 int performed_allocation = 0;
1455 ext4_grpblk_t free_blocks; /* number of free blocks in a group */
1456 struct super_block *sb;
1457 struct ext4_group_desc *gdp;
1458 struct ext4_super_block *es;
1459 struct ext4_sb_info *sbi;
1460 struct ext4_reserve_window_node *my_rsv = NULL;
1461 struct ext4_block_alloc_info *block_i;
1462 unsigned short windowsz = 0;
1464 static int goal_hits, goal_attempts;
1466 unsigned long ngroups;
1467 unsigned long num = *count;
1472 printk("ext4_new_block: nonexistent device");
1477 * Check quota for allocation of this block.
1479 if (DQUOT_ALLOC_BLOCK(inode, num)) {
1483 if (DLIMIT_ALLOC_BLOCK(inode, 1))
1487 es = EXT4_SB(sb)->s_es;
1488 ext4_debug("goal=%lu.\n", goal);
1490 * Allocate a block from reservation only when
1491 * filesystem is mounted with reservation(default,-o reservation), and
1492 * it's a regular file, and
1493 * the desired window size is greater than 0 (One could use ioctl
1494 * command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off
1495 * reservation on that particular file)
1497 block_i = EXT4_I(inode)->i_block_alloc_info;
1498 if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1499 my_rsv = &block_i->rsv_window_node;
1501 if (!ext4_has_free_blocks(sb)) {
1507 * First, test whether the goal block is free.
1509 if (goal < le32_to_cpu(es->s_first_data_block) ||
1510 goal >= ext4_blocks_count(es))
1511 goal = le32_to_cpu(es->s_first_data_block);
1512 ext4_get_group_no_and_offset(sb, goal, &group_no, &grp_target_blk);
1513 goal_group = group_no;
1515 gdp = ext4_get_group_desc(sb, group_no, &gdp_bh);
1519 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1521 * if there is not enough free blocks to make a new resevation
1522 * turn off reservation for this allocation
1524 if (my_rsv && (free_blocks < windowsz)
1525 && (rsv_is_empty(&my_rsv->rsv_window)))
1528 if (free_blocks > 0) {
1529 bitmap_bh = read_block_bitmap(sb, group_no);
1532 grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle,
1533 group_no, bitmap_bh, grp_target_blk,
1534 my_rsv, &num, &fatal);
1537 if (grp_alloc_blk >= 0)
1541 ngroups = EXT4_SB(sb)->s_groups_count;
1545 * Now search the rest of the groups. We assume that
1546 * i and gdp correctly point to the last group visited.
1548 for (bgi = 0; bgi < ngroups; bgi++) {
1550 if (group_no >= ngroups)
1552 gdp = ext4_get_group_desc(sb, group_no, &gdp_bh);
1555 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1557 * skip this group if the number of
1558 * free blocks is less than half of the reservation
1561 if (free_blocks <= (windowsz/2))
1565 bitmap_bh = read_block_bitmap(sb, group_no);
1569 * try to allocate block(s) from this group, without a goal(-1).
1571 grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle,
1572 group_no, bitmap_bh, -1, my_rsv,
1576 if (grp_alloc_blk >= 0)
1580 * We may end up a bogus ealier ENOSPC error due to
1581 * filesystem is "full" of reservations, but
1582 * there maybe indeed free blocks avaliable on disk
1583 * In this case, we just forget about the reservations
1584 * just do block allocation as without reservations.
1589 group_no = goal_group;
1592 /* No space left on the device */
1598 ext4_debug("using block group %d(%d)\n",
1599 group_no, gdp->bg_free_blocks_count);
1601 BUFFER_TRACE(gdp_bh, "get_write_access");
1602 fatal = ext4_journal_get_write_access(handle, gdp_bh);
1606 ret_block = grp_alloc_blk + ext4_group_first_block_no(sb, group_no);
1608 if (in_range(ext4_block_bitmap(sb, gdp), ret_block, num) ||
1609 in_range(ext4_block_bitmap(sb, gdp), ret_block, num) ||
1610 in_range(ret_block, ext4_inode_table(sb, gdp),
1611 EXT4_SB(sb)->s_itb_per_group) ||
1612 in_range(ret_block + num - 1, ext4_inode_table(sb, gdp),
1613 EXT4_SB(sb)->s_itb_per_group))
1614 ext4_error(sb, "ext4_new_block",
1615 "Allocating block in system zone - "
1616 "blocks from %llu, length %lu",
1619 performed_allocation = 1;
1621 #ifdef CONFIG_JBD_DEBUG
1623 struct buffer_head *debug_bh;
1625 /* Record bitmap buffer state in the newly allocated block */
1626 debug_bh = sb_find_get_block(sb, ret_block);
1628 BUFFER_TRACE(debug_bh, "state when allocated");
1629 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1633 jbd_lock_bh_state(bitmap_bh);
1634 spin_lock(sb_bgl_lock(sbi, group_no));
1635 if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1638 for (i = 0; i < num; i++) {
1639 if (ext4_test_bit(grp_alloc_blk+i,
1640 bh2jh(bitmap_bh)->b_committed_data)) {
1641 printk("%s: block was unexpectedly set in "
1642 "b_committed_data\n", __FUNCTION__);
1646 ext4_debug("found bit %d\n", grp_alloc_blk);
1647 spin_unlock(sb_bgl_lock(sbi, group_no));
1648 jbd_unlock_bh_state(bitmap_bh);
1651 if (ret_block + num - 1 >= ext4_blocks_count(es)) {
1652 ext4_error(sb, "ext4_new_block",
1653 "block(%llu) >= blocks count(%llu) - "
1654 "block_group = %lu, es == %p ", ret_block,
1655 ext4_blocks_count(es), group_no, es);
1660 * It is up to the caller to add the new buffer to a journal
1661 * list of some description. We don't know in advance whether
1662 * the caller wants to use it as metadata or data.
1664 ext4_debug("allocating block %lu. Goal hits %d of %d.\n",
1665 ret_block, goal_hits, goal_attempts);
1667 spin_lock(sb_bgl_lock(sbi, group_no));
1668 gdp->bg_free_blocks_count =
1669 cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count)-num);
1670 spin_unlock(sb_bgl_lock(sbi, group_no));
1671 percpu_counter_mod(&sbi->s_freeblocks_counter, -num);
1673 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1674 err = ext4_journal_dirty_metadata(handle, gdp_bh);
1684 DQUOT_FREE_BLOCK(inode, *count-num);
1691 if (!performed_allocation)
1692 DLIMIT_FREE_BLOCK(inode, 1);
1696 ext4_std_error(sb, fatal);
1699 * Undo the block allocation
1701 if (!performed_allocation)
1702 DQUOT_FREE_BLOCK(inode, *count);
1707 ext4_fsblk_t ext4_new_block(handle_t *handle, struct inode *inode,
1708 ext4_fsblk_t goal, int *errp)
1710 unsigned long count = 1;
1712 return ext4_new_blocks(handle, inode, goal, &count, errp);
1716 * ext4_count_free_blocks() -- count filesystem free blocks
1719 * Adds up the number of free blocks from each block group.
1721 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
1723 ext4_fsblk_t desc_count;
1724 struct ext4_group_desc *gdp;
1726 unsigned long ngroups = EXT4_SB(sb)->s_groups_count;
1728 struct ext4_super_block *es;
1729 ext4_fsblk_t bitmap_count;
1731 struct buffer_head *bitmap_bh = NULL;
1733 es = EXT4_SB(sb)->s_es;
1739 for (i = 0; i < ngroups; i++) {
1740 gdp = ext4_get_group_desc(sb, i, NULL);
1743 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1745 bitmap_bh = read_block_bitmap(sb, i);
1746 if (bitmap_bh == NULL)
1749 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
1750 printk("group %d: stored = %d, counted = %lu\n",
1751 i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1755 printk("ext4_count_free_blocks: stored = %llu"
1756 ", computed = %llu, %llu\n",
1757 EXT4_FREE_BLOCKS_COUNT(es),
1758 desc_count, bitmap_count);
1759 return bitmap_count;
1763 for (i = 0; i < ngroups; i++) {
1764 gdp = ext4_get_group_desc(sb, i, NULL);
1767 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1775 block_in_use(ext4_fsblk_t block, struct super_block *sb, unsigned char *map)
1777 ext4_grpblk_t offset;
1779 ext4_get_group_no_and_offset(sb, block, NULL, &offset);
1780 return ext4_test_bit (offset, map);
1783 static inline int test_root(int a, int b)
1792 static int ext4_group_sparse(int group)
1798 return (test_root(group, 7) || test_root(group, 5) ||
1799 test_root(group, 3));
1803 * ext4_bg_has_super - number of blocks used by the superblock in group
1804 * @sb: superblock for filesystem
1805 * @group: group number to check
1807 * Return the number of blocks used by the superblock (primary or backup)
1808 * in this group. Currently this will be only 0 or 1.
1810 int ext4_bg_has_super(struct super_block *sb, int group)
1812 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1813 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1814 !ext4_group_sparse(group))
1819 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, int group)
1821 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
1822 unsigned long first = metagroup * EXT4_DESC_PER_BLOCK(sb);
1823 unsigned long last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
1825 if (group == first || group == first + 1 || group == last)
1830 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, int group)
1832 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1833 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1834 !ext4_group_sparse(group))
1836 return EXT4_SB(sb)->s_gdb_count;
1840 * ext4_bg_num_gdb - number of blocks used by the group table in group
1841 * @sb: superblock for filesystem
1842 * @group: group number to check
1844 * Return the number of blocks used by the group descriptor table
1845 * (primary or backup) in this group. In the future there may be a
1846 * different number of descriptor blocks in each group.
1848 unsigned long ext4_bg_num_gdb(struct super_block *sb, int group)
1850 unsigned long first_meta_bg =
1851 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
1852 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
1854 if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
1855 metagroup < first_meta_bg)
1856 return ext4_bg_num_gdb_nometa(sb,group);
1858 return ext4_bg_num_gdb_meta(sb,group);