2 * linux/fs/ext3/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/jbd.h>
18 #include <linux/ext3_fs.h>
19 #include <linux/ext3_jbd.h>
20 #include <linux/quotaops.h>
21 #include <linux/buffer_head.h>
24 * balloc.c contains the blocks allocation and deallocation routines
28 * The free blocks are managed by bitmaps. A file system contains several
29 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
30 * block for inodes, N blocks for the inode table and data blocks.
32 * The file system contains group descriptors which are located after the
33 * super block. Each descriptor contains the number of the bitmap block and
34 * the free blocks count in the block. The descriptors are loaded in memory
35 * when a file system is mounted (see ext3_fill_super).
39 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
42 * ext3_get_group_desc() -- load group descriptor from disk
44 * @block_group: given block group
45 * @bh: pointer to the buffer head to store the block
48 struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb,
49 unsigned int block_group,
50 struct buffer_head ** bh)
52 unsigned long group_desc;
54 struct ext3_group_desc * desc;
55 struct ext3_sb_info *sbi = EXT3_SB(sb);
57 if (block_group >= sbi->s_groups_count) {
58 ext3_error (sb, "ext3_get_group_desc",
59 "block_group >= groups_count - "
60 "block_group = %d, groups_count = %lu",
61 block_group, sbi->s_groups_count);
67 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
68 offset = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
69 if (!sbi->s_group_desc[group_desc]) {
70 ext3_error (sb, "ext3_get_group_desc",
71 "Group descriptor not loaded - "
72 "block_group = %d, group_desc = %lu, desc = %lu",
73 block_group, group_desc, offset);
77 desc = (struct ext3_group_desc *) sbi->s_group_desc[group_desc]->b_data;
79 *bh = sbi->s_group_desc[group_desc];
83 static int ext3_valid_block_bitmap(struct super_block *sb,
84 struct ext3_group_desc *desc,
85 unsigned int block_group,
86 struct buffer_head *bh)
89 ext3_grpblk_t next_zero_bit;
90 ext3_fsblk_t bitmap_blk;
91 ext3_fsblk_t group_first_block;
93 group_first_block = ext3_group_first_block_no(sb, block_group);
95 /* check whether block bitmap block number is set */
96 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
97 offset = bitmap_blk - group_first_block;
98 if (!ext3_test_bit(offset, bh->b_data))
99 /* bad block bitmap */
102 /* check whether the inode bitmap block number is set */
103 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
104 offset = bitmap_blk - group_first_block;
105 if (!ext3_test_bit(offset, bh->b_data))
106 /* bad block bitmap */
109 /* check whether the inode table block number is set */
110 bitmap_blk = le32_to_cpu(desc->bg_inode_table);
111 offset = bitmap_blk - group_first_block;
112 next_zero_bit = ext3_find_next_zero_bit(bh->b_data,
113 offset + EXT3_SB(sb)->s_itb_per_group,
115 if (next_zero_bit >= offset + EXT3_SB(sb)->s_itb_per_group)
116 /* good bitmap for inode tables */
120 ext3_error(sb, __FUNCTION__,
121 "Invalid block bitmap - "
122 "block_group = %d, block = %lu",
123 block_group, bitmap_blk);
128 * read_block_bitmap()
130 * @block_group: given block group
132 * Read the bitmap for a given block_group,and validate the
133 * bits for block/inode/inode tables are set in the bitmaps
135 * Return buffer_head on success or NULL in case of failure.
137 static struct buffer_head *
138 read_block_bitmap(struct super_block *sb, unsigned int block_group)
140 struct ext3_group_desc * desc;
141 struct buffer_head * bh = NULL;
142 ext3_fsblk_t bitmap_blk;
144 desc = ext3_get_group_desc(sb, block_group, NULL);
147 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
148 bh = sb_getblk(sb, bitmap_blk);
150 ext3_error(sb, __FUNCTION__,
151 "Cannot read block bitmap - "
152 "block_group = %d, block_bitmap = %u",
153 block_group, le32_to_cpu(desc->bg_block_bitmap));
156 if (likely(bh_uptodate_or_lock(bh)))
159 if (bh_submit_read(bh) < 0) {
161 ext3_error(sb, __FUNCTION__,
162 "Cannot read block bitmap - "
163 "block_group = %d, block_bitmap = %u",
164 block_group, le32_to_cpu(desc->bg_block_bitmap));
167 if (!ext3_valid_block_bitmap(sb, desc, block_group, bh)) {
174 * The reservation window structure operations
175 * --------------------------------------------
176 * Operations include:
177 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
179 * We use a red-black tree to represent per-filesystem reservation
185 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
186 * @rb_root: root of per-filesystem reservation rb tree
187 * @verbose: verbose mode
188 * @fn: function which wishes to dump the reservation map
190 * If verbose is turned on, it will print the whole block reservation
191 * windows(start, end). Otherwise, it will only print out the "bad" windows,
192 * those windows that overlap with their immediate neighbors.
195 static void __rsv_window_dump(struct rb_root *root, int verbose,
199 struct ext3_reserve_window_node *rsv, *prev;
207 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
209 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
211 printk("reservation window 0x%p "
212 "start: %lu, end: %lu\n",
213 rsv, rsv->rsv_start, rsv->rsv_end);
214 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
215 printk("Bad reservation %p (start >= end)\n",
219 if (prev && prev->rsv_end >= rsv->rsv_start) {
220 printk("Bad reservation %p (prev->end >= start)\n",
226 printk("Restarting reservation walk in verbose mode\n");
234 printk("Window map complete.\n");
237 #define rsv_window_dump(root, verbose) \
238 __rsv_window_dump((root), (verbose), __FUNCTION__)
240 #define rsv_window_dump(root, verbose) do {} while (0)
244 * goal_in_my_reservation()
245 * @rsv: inode's reservation window
246 * @grp_goal: given goal block relative to the allocation block group
247 * @group: the current allocation block group
248 * @sb: filesystem super block
250 * Test if the given goal block (group relative) is within the file's
251 * own block reservation window range.
253 * If the reservation window is outside the goal allocation group, return 0;
254 * grp_goal (given goal block) could be -1, which means no specific
255 * goal block. In this case, always return 1.
256 * If the goal block is within the reservation window, return 1;
257 * otherwise, return 0;
260 goal_in_my_reservation(struct ext3_reserve_window *rsv, ext3_grpblk_t grp_goal,
261 unsigned int group, struct super_block * sb)
263 ext3_fsblk_t group_first_block, group_last_block;
265 group_first_block = ext3_group_first_block_no(sb, group);
266 group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
268 if ((rsv->_rsv_start > group_last_block) ||
269 (rsv->_rsv_end < group_first_block))
271 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
272 || (grp_goal + group_first_block > rsv->_rsv_end)))
278 * search_reserve_window()
279 * @rb_root: root of reservation tree
280 * @goal: target allocation block
282 * Find the reserved window which includes the goal, or the previous one
283 * if the goal is not in any window.
284 * Returns NULL if there are no windows or if all windows start after the goal.
286 static struct ext3_reserve_window_node *
287 search_reserve_window(struct rb_root *root, ext3_fsblk_t goal)
289 struct rb_node *n = root->rb_node;
290 struct ext3_reserve_window_node *rsv;
296 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
298 if (goal < rsv->rsv_start)
300 else if (goal > rsv->rsv_end)
306 * We've fallen off the end of the tree: the goal wasn't inside
307 * any particular node. OK, the previous node must be to one
308 * side of the interval containing the goal. If it's the RHS,
309 * we need to back up one.
311 if (rsv->rsv_start > goal) {
312 n = rb_prev(&rsv->rsv_node);
313 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
319 * ext3_rsv_window_add() -- Insert a window to the block reservation rb tree.
321 * @rsv: reservation window to add
323 * Must be called with rsv_lock hold.
325 void ext3_rsv_window_add(struct super_block *sb,
326 struct ext3_reserve_window_node *rsv)
328 struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
329 struct rb_node *node = &rsv->rsv_node;
330 ext3_fsblk_t start = rsv->rsv_start;
332 struct rb_node ** p = &root->rb_node;
333 struct rb_node * parent = NULL;
334 struct ext3_reserve_window_node *this;
339 this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
341 if (start < this->rsv_start)
343 else if (start > this->rsv_end)
346 rsv_window_dump(root, 1);
351 rb_link_node(node, parent, p);
352 rb_insert_color(node, root);
356 * ext3_rsv_window_remove() -- unlink a window from the reservation rb tree
358 * @rsv: reservation window to remove
360 * Mark the block reservation window as not allocated, and unlink it
361 * from the filesystem reservation window rb tree. Must be called with
364 static void rsv_window_remove(struct super_block *sb,
365 struct ext3_reserve_window_node *rsv)
367 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
368 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
369 rsv->rsv_alloc_hit = 0;
370 rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
374 * rsv_is_empty() -- Check if the reservation window is allocated.
375 * @rsv: given reservation window to check
377 * returns 1 if the end block is EXT3_RESERVE_WINDOW_NOT_ALLOCATED.
379 static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
381 /* a valid reservation end block could not be 0 */
382 return rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
386 * ext3_init_block_alloc_info()
387 * @inode: file inode structure
389 * Allocate and initialize the reservation window structure, and
390 * link the window to the ext3 inode structure at last
392 * The reservation window structure is only dynamically allocated
393 * and linked to ext3 inode the first time the open file
394 * needs a new block. So, before every ext3_new_block(s) call, for
395 * regular files, we should check whether the reservation window
396 * structure exists or not. In the latter case, this function is called.
397 * Fail to do so will result in block reservation being turned off for that
400 * This function is called from ext3_get_blocks_handle(), also called
401 * when setting the reservation window size through ioctl before the file
402 * is open for write (needs block allocation).
404 * Needs truncate_mutex protection prior to call this function.
406 void ext3_init_block_alloc_info(struct inode *inode)
408 struct ext3_inode_info *ei = EXT3_I(inode);
409 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
410 struct super_block *sb = inode->i_sb;
412 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
414 struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node;
416 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
417 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
420 * if filesystem is mounted with NORESERVATION, the goal
421 * reservation window size is set to zero to indicate
422 * block reservation is off
424 if (!test_opt(sb, RESERVATION))
425 rsv->rsv_goal_size = 0;
427 rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS;
428 rsv->rsv_alloc_hit = 0;
429 block_i->last_alloc_logical_block = 0;
430 block_i->last_alloc_physical_block = 0;
432 ei->i_block_alloc_info = block_i;
436 * ext3_discard_reservation()
439 * Discard(free) block reservation window on last file close, or truncate
442 * It is being called in three cases:
443 * ext3_release_file(): last writer close the file
444 * ext3_clear_inode(): last iput(), when nobody link to this file.
445 * ext3_truncate(): when the block indirect map is about to change.
448 void ext3_discard_reservation(struct inode *inode)
450 struct ext3_inode_info *ei = EXT3_I(inode);
451 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
452 struct ext3_reserve_window_node *rsv;
453 spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
458 rsv = &block_i->rsv_window_node;
459 if (!rsv_is_empty(&rsv->rsv_window)) {
461 if (!rsv_is_empty(&rsv->rsv_window))
462 rsv_window_remove(inode->i_sb, rsv);
463 spin_unlock(rsv_lock);
468 * ext3_free_blocks_sb() -- Free given blocks and update quota
469 * @handle: handle to this transaction
471 * @block: start physcial block to free
472 * @count: number of blocks to free
473 * @pdquot_freed_blocks: pointer to quota
475 void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
476 ext3_fsblk_t block, unsigned long count,
477 unsigned long *pdquot_freed_blocks)
479 struct buffer_head *bitmap_bh = NULL;
480 struct buffer_head *gd_bh;
481 unsigned long block_group;
484 unsigned long overflow;
485 struct ext3_group_desc * desc;
486 struct ext3_super_block * es;
487 struct ext3_sb_info *sbi;
489 ext3_grpblk_t group_freed;
491 *pdquot_freed_blocks = 0;
494 if (block < le32_to_cpu(es->s_first_data_block) ||
495 block + count < block ||
496 block + count > le32_to_cpu(es->s_blocks_count)) {
497 ext3_error (sb, "ext3_free_blocks",
498 "Freeing blocks not in datazone - "
499 "block = "E3FSBLK", count = %lu", block, count);
503 ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
507 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
508 EXT3_BLOCKS_PER_GROUP(sb);
509 bit = (block - le32_to_cpu(es->s_first_data_block)) %
510 EXT3_BLOCKS_PER_GROUP(sb);
512 * Check to see if we are freeing blocks across a group
515 if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
516 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
520 bitmap_bh = read_block_bitmap(sb, block_group);
523 desc = ext3_get_group_desc (sb, block_group, &gd_bh);
527 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
528 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
529 in_range (block, le32_to_cpu(desc->bg_inode_table),
530 sbi->s_itb_per_group) ||
531 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
532 sbi->s_itb_per_group)) {
533 ext3_error (sb, "ext3_free_blocks",
534 "Freeing blocks in system zones - "
535 "Block = "E3FSBLK", count = %lu",
541 * We are about to start releasing blocks in the bitmap,
542 * so we need undo access.
544 /* @@@ check errors */
545 BUFFER_TRACE(bitmap_bh, "getting undo access");
546 err = ext3_journal_get_undo_access(handle, bitmap_bh);
551 * We are about to modify some metadata. Call the journal APIs
552 * to unshare ->b_data if a currently-committing transaction is
555 BUFFER_TRACE(gd_bh, "get_write_access");
556 err = ext3_journal_get_write_access(handle, gd_bh);
560 jbd_lock_bh_state(bitmap_bh);
562 for (i = 0, group_freed = 0; i < count; i++) {
564 * An HJ special. This is expensive...
566 #ifdef CONFIG_JBD_DEBUG
567 jbd_unlock_bh_state(bitmap_bh);
569 struct buffer_head *debug_bh;
570 debug_bh = sb_find_get_block(sb, block + i);
572 BUFFER_TRACE(debug_bh, "Deleted!");
573 if (!bh2jh(bitmap_bh)->b_committed_data)
574 BUFFER_TRACE(debug_bh,
575 "No commited data in bitmap");
576 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
580 jbd_lock_bh_state(bitmap_bh);
582 if (need_resched()) {
583 jbd_unlock_bh_state(bitmap_bh);
585 jbd_lock_bh_state(bitmap_bh);
587 /* @@@ This prevents newly-allocated data from being
588 * freed and then reallocated within the same
591 * Ideally we would want to allow that to happen, but to
592 * do so requires making journal_forget() capable of
593 * revoking the queued write of a data block, which
594 * implies blocking on the journal lock. *forget()
595 * cannot block due to truncate races.
597 * Eventually we can fix this by making journal_forget()
598 * return a status indicating whether or not it was able
599 * to revoke the buffer. On successful revoke, it is
600 * safe not to set the allocation bit in the committed
601 * bitmap, because we know that there is no outstanding
602 * activity on the buffer any more and so it is safe to
605 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
606 J_ASSERT_BH(bitmap_bh,
607 bh2jh(bitmap_bh)->b_committed_data != NULL);
608 ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
609 bh2jh(bitmap_bh)->b_committed_data);
612 * We clear the bit in the bitmap after setting the committed
613 * data bit, because this is the reverse order to that which
614 * the allocator uses.
616 BUFFER_TRACE(bitmap_bh, "clear bit");
617 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
618 bit + i, bitmap_bh->b_data)) {
619 jbd_unlock_bh_state(bitmap_bh);
620 ext3_error(sb, __FUNCTION__,
621 "bit already cleared for block "E3FSBLK,
623 jbd_lock_bh_state(bitmap_bh);
624 BUFFER_TRACE(bitmap_bh, "bit already cleared");
629 jbd_unlock_bh_state(bitmap_bh);
631 spin_lock(sb_bgl_lock(sbi, block_group));
632 le16_add_cpu(&desc->bg_free_blocks_count, group_freed);
633 spin_unlock(sb_bgl_lock(sbi, block_group));
634 percpu_counter_add(&sbi->s_freeblocks_counter, count);
636 /* We dirtied the bitmap block */
637 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
638 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
640 /* And the group descriptor block */
641 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
642 ret = ext3_journal_dirty_metadata(handle, gd_bh);
644 *pdquot_freed_blocks += group_freed;
646 if (overflow && !err) {
654 ext3_std_error(sb, err);
659 * ext3_free_blocks() -- Free given blocks and update quota
660 * @handle: handle for this transaction
662 * @block: start physical block to free
663 * @count: number of blocks to count
665 void ext3_free_blocks(handle_t *handle, struct inode *inode,
666 ext3_fsblk_t block, unsigned long count)
668 struct super_block * sb;
669 unsigned long dquot_freed_blocks;
673 printk ("ext3_free_blocks: nonexistent device");
676 ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
677 if (dquot_freed_blocks)
678 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
683 * ext3_test_allocatable()
684 * @nr: given allocation block group
685 * @bh: bufferhead contains the bitmap of the given block group
687 * For ext3 allocations, we must not reuse any blocks which are
688 * allocated in the bitmap buffer's "last committed data" copy. This
689 * prevents deletes from freeing up the page for reuse until we have
690 * committed the delete transaction.
692 * If we didn't do this, then deleting something and reallocating it as
693 * data would allow the old block to be overwritten before the
694 * transaction committed (because we force data to disk before commit).
695 * This would lead to corruption if we crashed between overwriting the
696 * data and committing the delete.
698 * @@@ We may want to make this allocation behaviour conditional on
699 * data-writes at some point, and disable it for metadata allocations or
702 static int ext3_test_allocatable(ext3_grpblk_t nr, struct buffer_head *bh)
705 struct journal_head *jh = bh2jh(bh);
707 if (ext3_test_bit(nr, bh->b_data))
710 jbd_lock_bh_state(bh);
711 if (!jh->b_committed_data)
714 ret = !ext3_test_bit(nr, jh->b_committed_data);
715 jbd_unlock_bh_state(bh);
720 * bitmap_search_next_usable_block()
721 * @start: the starting block (group relative) of the search
722 * @bh: bufferhead contains the block group bitmap
723 * @maxblocks: the ending block (group relative) of the reservation
725 * The bitmap search --- search forward alternately through the actual
726 * bitmap on disk and the last-committed copy in journal, until we find a
727 * bit free in both bitmaps.
730 bitmap_search_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
731 ext3_grpblk_t maxblocks)
734 struct journal_head *jh = bh2jh(bh);
736 while (start < maxblocks) {
737 next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
738 if (next >= maxblocks)
740 if (ext3_test_allocatable(next, bh))
742 jbd_lock_bh_state(bh);
743 if (jh->b_committed_data)
744 start = ext3_find_next_zero_bit(jh->b_committed_data,
746 jbd_unlock_bh_state(bh);
752 * find_next_usable_block()
753 * @start: the starting block (group relative) to find next
754 * allocatable block in bitmap.
755 * @bh: bufferhead contains the block group bitmap
756 * @maxblocks: the ending block (group relative) for the search
758 * Find an allocatable block in a bitmap. We honor both the bitmap and
759 * its last-committed copy (if that exists), and perform the "most
760 * appropriate allocation" algorithm of looking for a free block near
761 * the initial goal; then for a free byte somewhere in the bitmap; then
762 * for any free bit in the bitmap.
765 find_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
766 ext3_grpblk_t maxblocks)
768 ext3_grpblk_t here, next;
773 * The goal was occupied; search forward for a free
774 * block within the next XX blocks.
776 * end_goal is more or less random, but it has to be
777 * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
778 * next 64-bit boundary is simple..
780 ext3_grpblk_t end_goal = (start + 63) & ~63;
781 if (end_goal > maxblocks)
782 end_goal = maxblocks;
783 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
784 if (here < end_goal && ext3_test_allocatable(here, bh))
786 ext3_debug("Bit not found near goal\n");
793 p = ((char *)bh->b_data) + (here >> 3);
794 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
795 next = (r - ((char *)bh->b_data)) << 3;
797 if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
801 * The bitmap search --- search forward alternately through the actual
802 * bitmap and the last-committed copy until we find a bit free in
805 here = bitmap_search_next_usable_block(here, bh, maxblocks);
811 * @block: the free block (group relative) to allocate
812 * @bh: the bufferhead containts the block group bitmap
814 * We think we can allocate this block in this bitmap. Try to set the bit.
815 * If that succeeds then check that nobody has allocated and then freed the
816 * block since we saw that is was not marked in b_committed_data. If it _was_
817 * allocated and freed then clear the bit in the bitmap again and return
821 claim_block(spinlock_t *lock, ext3_grpblk_t block, struct buffer_head *bh)
823 struct journal_head *jh = bh2jh(bh);
826 if (ext3_set_bit_atomic(lock, block, bh->b_data))
828 jbd_lock_bh_state(bh);
829 if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
830 ext3_clear_bit_atomic(lock, block, bh->b_data);
835 jbd_unlock_bh_state(bh);
840 * ext3_try_to_allocate()
842 * @handle: handle to this transaction
843 * @group: given allocation block group
844 * @bitmap_bh: bufferhead holds the block bitmap
845 * @grp_goal: given target block within the group
846 * @count: target number of blocks to allocate
847 * @my_rsv: reservation window
849 * Attempt to allocate blocks within a give range. Set the range of allocation
850 * first, then find the first free bit(s) from the bitmap (within the range),
851 * and at last, allocate the blocks by claiming the found free bit as allocated.
853 * To set the range of this allocation:
854 * if there is a reservation window, only try to allocate block(s) from the
855 * file's own reservation window;
856 * Otherwise, the allocation range starts from the give goal block, ends at
857 * the block group's last block.
859 * If we failed to allocate the desired block then we may end up crossing to a
860 * new bitmap. In that case we must release write access to the old one via
861 * ext3_journal_release_buffer(), else we'll run out of credits.
864 ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
865 struct buffer_head *bitmap_bh, ext3_grpblk_t grp_goal,
866 unsigned long *count, struct ext3_reserve_window *my_rsv)
868 ext3_fsblk_t group_first_block;
869 ext3_grpblk_t start, end;
870 unsigned long num = 0;
872 /* we do allocation within the reservation window if we have a window */
874 group_first_block = ext3_group_first_block_no(sb, group);
875 if (my_rsv->_rsv_start >= group_first_block)
876 start = my_rsv->_rsv_start - group_first_block;
878 /* reservation window cross group boundary */
880 end = my_rsv->_rsv_end - group_first_block + 1;
881 if (end > EXT3_BLOCKS_PER_GROUP(sb))
882 /* reservation window crosses group boundary */
883 end = EXT3_BLOCKS_PER_GROUP(sb);
884 if ((start <= grp_goal) && (grp_goal < end))
893 end = EXT3_BLOCKS_PER_GROUP(sb);
896 BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
899 if (grp_goal < 0 || !ext3_test_allocatable(grp_goal, bitmap_bh)) {
900 grp_goal = find_next_usable_block(start, bitmap_bh, end);
906 for (i = 0; i < 7 && grp_goal > start &&
907 ext3_test_allocatable(grp_goal - 1,
915 if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group),
916 grp_goal, bitmap_bh)) {
918 * The block was allocated by another thread, or it was
919 * allocated and then freed by another thread
929 while (num < *count && grp_goal < end
930 && ext3_test_allocatable(grp_goal, bitmap_bh)
931 && claim_block(sb_bgl_lock(EXT3_SB(sb), group),
932 grp_goal, bitmap_bh)) {
937 return grp_goal - num;
944 * find_next_reservable_window():
945 * find a reservable space within the given range.
946 * It does not allocate the reservation window for now:
947 * alloc_new_reservation() will do the work later.
949 * @search_head: the head of the searching list;
950 * This is not necessarily the list head of the whole filesystem
952 * We have both head and start_block to assist the search
953 * for the reservable space. The list starts from head,
954 * but we will shift to the place where start_block is,
955 * then start from there, when looking for a reservable space.
957 * @size: the target new reservation window size
959 * @group_first_block: the first block we consider to start
960 * the real search from
963 * the maximum block number that our goal reservable space
964 * could start from. This is normally the last block in this
965 * group. The search will end when we found the start of next
966 * possible reservable space is out of this boundary.
967 * This could handle the cross boundary reservation window
970 * basically we search from the given range, rather than the whole
971 * reservation double linked list, (start_block, last_block)
972 * to find a free region that is of my size and has not
976 static int find_next_reservable_window(
977 struct ext3_reserve_window_node *search_head,
978 struct ext3_reserve_window_node *my_rsv,
979 struct super_block * sb,
980 ext3_fsblk_t start_block,
981 ext3_fsblk_t last_block)
983 struct rb_node *next;
984 struct ext3_reserve_window_node *rsv, *prev;
986 int size = my_rsv->rsv_goal_size;
988 /* TODO: make the start of the reservation window byte-aligned */
989 /* cur = *start_block & ~7;*/
996 if (cur <= rsv->rsv_end)
997 cur = rsv->rsv_end + 1;
1000 * in the case we could not find a reservable space
1001 * that is what is expected, during the re-search, we could
1002 * remember what's the largest reservable space we could have
1003 * and return that one.
1005 * For now it will fail if we could not find the reservable
1006 * space with expected-size (or more)...
1008 if (cur > last_block)
1009 return -1; /* fail */
1012 next = rb_next(&rsv->rsv_node);
1013 rsv = rb_entry(next,struct ext3_reserve_window_node,rsv_node);
1016 * Reached the last reservation, we can just append to the
1022 if (cur + size <= rsv->rsv_start) {
1024 * Found a reserveable space big enough. We could
1025 * have a reservation across the group boundary here
1031 * we come here either :
1032 * when we reach the end of the whole list,
1033 * and there is empty reservable space after last entry in the list.
1034 * append it to the end of the list.
1036 * or we found one reservable space in the middle of the list,
1037 * return the reservation window that we could append to.
1041 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
1042 rsv_window_remove(sb, my_rsv);
1045 * Let's book the whole avaliable window for now. We will check the
1046 * disk bitmap later and then, if there are free blocks then we adjust
1047 * the window size if it's larger than requested.
1048 * Otherwise, we will remove this node from the tree next time
1049 * call find_next_reservable_window.
1051 my_rsv->rsv_start = cur;
1052 my_rsv->rsv_end = cur + size - 1;
1053 my_rsv->rsv_alloc_hit = 0;
1056 ext3_rsv_window_add(sb, my_rsv);
1062 * alloc_new_reservation()--allocate a new reservation window
1064 * To make a new reservation, we search part of the filesystem
1065 * reservation list (the list that inside the group). We try to
1066 * allocate a new reservation window near the allocation goal,
1067 * or the beginning of the group, if there is no goal.
1069 * We first find a reservable space after the goal, then from
1070 * there, we check the bitmap for the first free block after
1071 * it. If there is no free block until the end of group, then the
1072 * whole group is full, we failed. Otherwise, check if the free
1073 * block is inside the expected reservable space, if so, we
1075 * If the first free block is outside the reservable space, then
1076 * start from the first free block, we search for next available
1079 * on succeed, a new reservation will be found and inserted into the list
1080 * It contains at least one free block, and it does not overlap with other
1081 * reservation windows.
1083 * failed: we failed to find a reservation window in this group
1085 * @rsv: the reservation
1087 * @grp_goal: The goal (group-relative). It is where the search for a
1088 * free reservable space should start from.
1089 * if we have a grp_goal(grp_goal >0 ), then start from there,
1090 * no grp_goal(grp_goal = -1), we start from the first block
1093 * @sb: the super block
1094 * @group: the group we are trying to allocate in
1095 * @bitmap_bh: the block group block bitmap
1098 static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
1099 ext3_grpblk_t grp_goal, struct super_block *sb,
1100 unsigned int group, struct buffer_head *bitmap_bh)
1102 struct ext3_reserve_window_node *search_head;
1103 ext3_fsblk_t group_first_block, group_end_block, start_block;
1104 ext3_grpblk_t first_free_block;
1105 struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
1108 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1110 group_first_block = ext3_group_first_block_no(sb, group);
1111 group_end_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1114 start_block = group_first_block;
1116 start_block = grp_goal + group_first_block;
1118 size = my_rsv->rsv_goal_size;
1120 if (!rsv_is_empty(&my_rsv->rsv_window)) {
1122 * if the old reservation is cross group boundary
1123 * and if the goal is inside the old reservation window,
1124 * we will come here when we just failed to allocate from
1125 * the first part of the window. We still have another part
1126 * that belongs to the next group. In this case, there is no
1127 * point to discard our window and try to allocate a new one
1128 * in this group(which will fail). we should
1129 * keep the reservation window, just simply move on.
1131 * Maybe we could shift the start block of the reservation
1132 * window to the first block of next group.
1135 if ((my_rsv->rsv_start <= group_end_block) &&
1136 (my_rsv->rsv_end > group_end_block) &&
1137 (start_block >= my_rsv->rsv_start))
1140 if ((my_rsv->rsv_alloc_hit >
1141 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
1143 * if the previously allocation hit ratio is
1144 * greater than 1/2, then we double the size of
1145 * the reservation window the next time,
1146 * otherwise we keep the same size window
1149 if (size > EXT3_MAX_RESERVE_BLOCKS)
1150 size = EXT3_MAX_RESERVE_BLOCKS;
1151 my_rsv->rsv_goal_size= size;
1155 spin_lock(rsv_lock);
1157 * shift the search start to the window near the goal block
1159 search_head = search_reserve_window(fs_rsv_root, start_block);
1162 * find_next_reservable_window() simply finds a reservable window
1163 * inside the given range(start_block, group_end_block).
1165 * To make sure the reservation window has a free bit inside it, we
1166 * need to check the bitmap after we found a reservable window.
1169 ret = find_next_reservable_window(search_head, my_rsv, sb,
1170 start_block, group_end_block);
1173 if (!rsv_is_empty(&my_rsv->rsv_window))
1174 rsv_window_remove(sb, my_rsv);
1175 spin_unlock(rsv_lock);
1180 * On success, find_next_reservable_window() returns the
1181 * reservation window where there is a reservable space after it.
1182 * Before we reserve this reservable space, we need
1183 * to make sure there is at least a free block inside this region.
1185 * searching the first free bit on the block bitmap and copy of
1186 * last committed bitmap alternatively, until we found a allocatable
1187 * block. Search start from the start block of the reservable space
1190 spin_unlock(rsv_lock);
1191 first_free_block = bitmap_search_next_usable_block(
1192 my_rsv->rsv_start - group_first_block,
1193 bitmap_bh, group_end_block - group_first_block + 1);
1195 if (first_free_block < 0) {
1197 * no free block left on the bitmap, no point
1198 * to reserve the space. return failed.
1200 spin_lock(rsv_lock);
1201 if (!rsv_is_empty(&my_rsv->rsv_window))
1202 rsv_window_remove(sb, my_rsv);
1203 spin_unlock(rsv_lock);
1204 return -1; /* failed */
1207 start_block = first_free_block + group_first_block;
1209 * check if the first free block is within the
1210 * free space we just reserved
1212 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1213 return 0; /* success */
1215 * if the first free bit we found is out of the reservable space
1216 * continue search for next reservable space,
1217 * start from where the free block is,
1218 * we also shift the list head to where we stopped last time
1220 search_head = my_rsv;
1221 spin_lock(rsv_lock);
1226 * try_to_extend_reservation()
1227 * @my_rsv: given reservation window
1229 * @size: the delta to extend
1231 * Attempt to expand the reservation window large enough to have
1232 * required number of free blocks
1234 * Since ext3_try_to_allocate() will always allocate blocks within
1235 * the reservation window range, if the window size is too small,
1236 * multiple blocks allocation has to stop at the end of the reservation
1237 * window. To make this more efficient, given the total number of
1238 * blocks needed and the current size of the window, we try to
1239 * expand the reservation window size if necessary on a best-effort
1240 * basis before ext3_new_blocks() tries to allocate blocks,
1242 static void try_to_extend_reservation(struct ext3_reserve_window_node *my_rsv,
1243 struct super_block *sb, int size)
1245 struct ext3_reserve_window_node *next_rsv;
1246 struct rb_node *next;
1247 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1249 if (!spin_trylock(rsv_lock))
1252 next = rb_next(&my_rsv->rsv_node);
1255 my_rsv->rsv_end += size;
1257 next_rsv = rb_entry(next, struct ext3_reserve_window_node, rsv_node);
1259 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1260 my_rsv->rsv_end += size;
1262 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1264 spin_unlock(rsv_lock);
1268 * ext3_try_to_allocate_with_rsv()
1270 * @handle: handle to this transaction
1271 * @group: given allocation block group
1272 * @bitmap_bh: bufferhead holds the block bitmap
1273 * @grp_goal: given target block within the group
1274 * @count: target number of blocks to allocate
1275 * @my_rsv: reservation window
1276 * @errp: pointer to store the error code
1278 * This is the main function used to allocate a new block and its reservation
1281 * Each time when a new block allocation is need, first try to allocate from
1282 * its own reservation. If it does not have a reservation window, instead of
1283 * looking for a free bit on bitmap first, then look up the reservation list to
1284 * see if it is inside somebody else's reservation window, we try to allocate a
1285 * reservation window for it starting from the goal first. Then do the block
1286 * allocation within the reservation window.
1288 * This will avoid keeping on searching the reservation list again and
1289 * again when somebody is looking for a free block (without
1290 * reservation), and there are lots of free blocks, but they are all
1293 * We use a red-black tree for the per-filesystem reservation list.
1296 static ext3_grpblk_t
1297 ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1298 unsigned int group, struct buffer_head *bitmap_bh,
1299 ext3_grpblk_t grp_goal,
1300 struct ext3_reserve_window_node * my_rsv,
1301 unsigned long *count, int *errp)
1303 ext3_fsblk_t group_first_block, group_last_block;
1304 ext3_grpblk_t ret = 0;
1306 unsigned long num = *count;
1311 * Make sure we use undo access for the bitmap, because it is critical
1312 * that we do the frozen_data COW on bitmap buffers in all cases even
1313 * if the buffer is in BJ_Forget state in the committing transaction.
1315 BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1316 fatal = ext3_journal_get_undo_access(handle, bitmap_bh);
1323 * we don't deal with reservation when
1324 * filesystem is mounted without reservation
1325 * or the file is not a regular file
1326 * or last attempt to allocate a block with reservation turned on failed
1328 if (my_rsv == NULL ) {
1329 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1330 grp_goal, count, NULL);
1334 * grp_goal is a group relative block number (if there is a goal)
1335 * 0 <= grp_goal < EXT3_BLOCKS_PER_GROUP(sb)
1336 * first block is a filesystem wide block number
1337 * first block is the block number of the first block in this group
1339 group_first_block = ext3_group_first_block_no(sb, group);
1340 group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1343 * Basically we will allocate a new block from inode's reservation
1346 * We need to allocate a new reservation window, if:
1347 * a) inode does not have a reservation window; or
1348 * b) last attempt to allocate a block from existing reservation
1350 * c) we come here with a goal and with a reservation window
1352 * We do not need to allocate a new reservation window if we come here
1353 * at the beginning with a goal and the goal is inside the window, or
1354 * we don't have a goal but already have a reservation window.
1355 * then we could go to allocate from the reservation window directly.
1358 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1359 !goal_in_my_reservation(&my_rsv->rsv_window,
1360 grp_goal, group, sb)) {
1361 if (my_rsv->rsv_goal_size < *count)
1362 my_rsv->rsv_goal_size = *count;
1363 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1368 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1369 grp_goal, group, sb))
1371 } else if (grp_goal >= 0) {
1372 int curr = my_rsv->rsv_end -
1373 (grp_goal + group_first_block) + 1;
1376 try_to_extend_reservation(my_rsv, sb,
1380 if ((my_rsv->rsv_start > group_last_block) ||
1381 (my_rsv->rsv_end < group_first_block)) {
1382 rsv_window_dump(&EXT3_SB(sb)->s_rsv_window_root, 1);
1385 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1386 grp_goal, &num, &my_rsv->rsv_window);
1388 my_rsv->rsv_alloc_hit += num;
1390 break; /* succeed */
1396 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1398 fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1406 BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1407 ext3_journal_release_buffer(handle, bitmap_bh);
1412 * ext3_has_free_blocks()
1413 * @sbi: in-core super block structure.
1415 * Check if filesystem has at least 1 free block available for allocation.
1417 static int ext3_has_free_blocks(struct ext3_sb_info *sbi)
1419 ext3_fsblk_t free_blocks, root_blocks;
1421 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1422 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1423 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1424 sbi->s_resuid != current->fsuid &&
1425 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1432 * ext3_should_retry_alloc()
1434 * @retries number of attemps has been made
1436 * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1437 * it is profitable to retry the operation, this function will wait
1438 * for the current or commiting transaction to complete, and then
1441 * if the total number of retries exceed three times, return FALSE.
1443 int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1445 if (!ext3_has_free_blocks(EXT3_SB(sb)) || (*retries)++ > 3)
1448 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1450 return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1454 * ext3_new_blocks() -- core block(s) allocation function
1455 * @handle: handle to this transaction
1456 * @inode: file inode
1457 * @goal: given target block(filesystem wide)
1458 * @count: target number of blocks to allocate
1461 * ext3_new_blocks uses a goal block to assist allocation. It tries to
1462 * allocate block(s) from the block group contains the goal block first. If that
1463 * fails, it will try to allocate block(s) from other block groups without
1464 * any specific goal block.
1467 ext3_fsblk_t ext3_new_blocks(handle_t *handle, struct inode *inode,
1468 ext3_fsblk_t goal, unsigned long *count, int *errp)
1470 struct buffer_head *bitmap_bh = NULL;
1471 struct buffer_head *gdp_bh;
1474 ext3_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1475 ext3_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1476 ext3_fsblk_t ret_block; /* filesyetem-wide allocated block */
1477 int bgi; /* blockgroup iteration index */
1479 int performed_allocation = 0;
1480 ext3_grpblk_t free_blocks; /* number of free blocks in a group */
1481 struct super_block *sb;
1482 struct ext3_group_desc *gdp;
1483 struct ext3_super_block *es;
1484 struct ext3_sb_info *sbi;
1485 struct ext3_reserve_window_node *my_rsv = NULL;
1486 struct ext3_block_alloc_info *block_i;
1487 unsigned short windowsz = 0;
1489 static int goal_hits, goal_attempts;
1491 unsigned long ngroups;
1492 unsigned long num = *count;
1497 printk("ext3_new_block: nonexistent device");
1502 * Check quota for allocation of this block.
1504 if (DQUOT_ALLOC_BLOCK(inode, num)) {
1510 es = EXT3_SB(sb)->s_es;
1511 ext3_debug("goal=%lu.\n", goal);
1513 * Allocate a block from reservation only when
1514 * filesystem is mounted with reservation(default,-o reservation), and
1515 * it's a regular file, and
1516 * the desired window size is greater than 0 (One could use ioctl
1517 * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1518 * reservation on that particular file)
1520 block_i = EXT3_I(inode)->i_block_alloc_info;
1521 if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1522 my_rsv = &block_i->rsv_window_node;
1524 if (!ext3_has_free_blocks(sbi)) {
1530 * First, test whether the goal block is free.
1532 if (goal < le32_to_cpu(es->s_first_data_block) ||
1533 goal >= le32_to_cpu(es->s_blocks_count))
1534 goal = le32_to_cpu(es->s_first_data_block);
1535 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1536 EXT3_BLOCKS_PER_GROUP(sb);
1537 goal_group = group_no;
1539 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1543 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1545 * if there is not enough free blocks to make a new resevation
1546 * turn off reservation for this allocation
1548 if (my_rsv && (free_blocks < windowsz)
1549 && (rsv_is_empty(&my_rsv->rsv_window)))
1552 if (free_blocks > 0) {
1553 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1554 EXT3_BLOCKS_PER_GROUP(sb));
1555 bitmap_bh = read_block_bitmap(sb, group_no);
1558 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1559 group_no, bitmap_bh, grp_target_blk,
1560 my_rsv, &num, &fatal);
1563 if (grp_alloc_blk >= 0)
1567 ngroups = EXT3_SB(sb)->s_groups_count;
1571 * Now search the rest of the groups. We assume that
1572 * group_no and gdp correctly point to the last group visited.
1574 for (bgi = 0; bgi < ngroups; bgi++) {
1576 if (group_no >= ngroups)
1578 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1581 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1583 * skip this group if the number of
1584 * free blocks is less than half of the reservation
1587 if (free_blocks <= (windowsz/2))
1591 bitmap_bh = read_block_bitmap(sb, group_no);
1595 * try to allocate block(s) from this group, without a goal(-1).
1597 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1598 group_no, bitmap_bh, -1, my_rsv,
1602 if (grp_alloc_blk >= 0)
1606 * We may end up a bogus ealier ENOSPC error due to
1607 * filesystem is "full" of reservations, but
1608 * there maybe indeed free blocks avaliable on disk
1609 * In this case, we just forget about the reservations
1610 * just do block allocation as without reservations.
1615 group_no = goal_group;
1618 /* No space left on the device */
1624 ext3_debug("using block group %d(%d)\n",
1625 group_no, gdp->bg_free_blocks_count);
1627 BUFFER_TRACE(gdp_bh, "get_write_access");
1628 fatal = ext3_journal_get_write_access(handle, gdp_bh);
1632 ret_block = grp_alloc_blk + ext3_group_first_block_no(sb, group_no);
1634 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1635 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1636 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1637 EXT3_SB(sb)->s_itb_per_group) ||
1638 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1639 EXT3_SB(sb)->s_itb_per_group)) {
1640 ext3_error(sb, "ext3_new_block",
1641 "Allocating block in system zone - "
1642 "blocks from "E3FSBLK", length %lu",
1647 performed_allocation = 1;
1649 #ifdef CONFIG_JBD_DEBUG
1651 struct buffer_head *debug_bh;
1653 /* Record bitmap buffer state in the newly allocated block */
1654 debug_bh = sb_find_get_block(sb, ret_block);
1656 BUFFER_TRACE(debug_bh, "state when allocated");
1657 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1661 jbd_lock_bh_state(bitmap_bh);
1662 spin_lock(sb_bgl_lock(sbi, group_no));
1663 if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1666 for (i = 0; i < num; i++) {
1667 if (ext3_test_bit(grp_alloc_blk+i,
1668 bh2jh(bitmap_bh)->b_committed_data)) {
1669 printk("%s: block was unexpectedly set in "
1670 "b_committed_data\n", __FUNCTION__);
1674 ext3_debug("found bit %d\n", grp_alloc_blk);
1675 spin_unlock(sb_bgl_lock(sbi, group_no));
1676 jbd_unlock_bh_state(bitmap_bh);
1679 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1680 ext3_error(sb, "ext3_new_block",
1681 "block("E3FSBLK") >= blocks count(%d) - "
1682 "block_group = %d, es == %p ", ret_block,
1683 le32_to_cpu(es->s_blocks_count), group_no, es);
1688 * It is up to the caller to add the new buffer to a journal
1689 * list of some description. We don't know in advance whether
1690 * the caller wants to use it as metadata or data.
1692 ext3_debug("allocating block %lu. Goal hits %d of %d.\n",
1693 ret_block, goal_hits, goal_attempts);
1695 spin_lock(sb_bgl_lock(sbi, group_no));
1696 le16_add_cpu(&gdp->bg_free_blocks_count, -num);
1697 spin_unlock(sb_bgl_lock(sbi, group_no));
1698 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1700 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1701 err = ext3_journal_dirty_metadata(handle, gdp_bh);
1711 DQUOT_FREE_BLOCK(inode, *count-num);
1720 ext3_std_error(sb, fatal);
1723 * Undo the block allocation
1725 if (!performed_allocation)
1726 DQUOT_FREE_BLOCK(inode, *count);
1731 ext3_fsblk_t ext3_new_block(handle_t *handle, struct inode *inode,
1732 ext3_fsblk_t goal, int *errp)
1734 unsigned long count = 1;
1736 return ext3_new_blocks(handle, inode, goal, &count, errp);
1740 * ext3_count_free_blocks() -- count filesystem free blocks
1743 * Adds up the number of free blocks from each block group.
1745 ext3_fsblk_t ext3_count_free_blocks(struct super_block *sb)
1747 ext3_fsblk_t desc_count;
1748 struct ext3_group_desc *gdp;
1750 unsigned long ngroups = EXT3_SB(sb)->s_groups_count;
1752 struct ext3_super_block *es;
1753 ext3_fsblk_t bitmap_count;
1755 struct buffer_head *bitmap_bh = NULL;
1757 es = EXT3_SB(sb)->s_es;
1763 for (i = 0; i < ngroups; i++) {
1764 gdp = ext3_get_group_desc(sb, i, NULL);
1767 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1769 bitmap_bh = read_block_bitmap(sb, i);
1770 if (bitmap_bh == NULL)
1773 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1774 printk("group %d: stored = %d, counted = %lu\n",
1775 i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1779 printk("ext3_count_free_blocks: stored = "E3FSBLK
1780 ", computed = "E3FSBLK", "E3FSBLK"\n",
1781 le32_to_cpu(es->s_free_blocks_count),
1782 desc_count, bitmap_count);
1783 return bitmap_count;
1787 for (i = 0; i < ngroups; i++) {
1788 gdp = ext3_get_group_desc(sb, i, NULL);
1791 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1798 static inline int test_root(int a, int b)
1807 static int ext3_group_sparse(int group)
1813 return (test_root(group, 7) || test_root(group, 5) ||
1814 test_root(group, 3));
1818 * ext3_bg_has_super - number of blocks used by the superblock in group
1819 * @sb: superblock for filesystem
1820 * @group: group number to check
1822 * Return the number of blocks used by the superblock (primary or backup)
1823 * in this group. Currently this will be only 0 or 1.
1825 int ext3_bg_has_super(struct super_block *sb, int group)
1827 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1828 EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1829 !ext3_group_sparse(group))
1834 static unsigned long ext3_bg_num_gdb_meta(struct super_block *sb, int group)
1836 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1837 unsigned long first = metagroup * EXT3_DESC_PER_BLOCK(sb);
1838 unsigned long last = first + EXT3_DESC_PER_BLOCK(sb) - 1;
1840 if (group == first || group == first + 1 || group == last)
1845 static unsigned long ext3_bg_num_gdb_nometa(struct super_block *sb, int group)
1847 return ext3_bg_has_super(sb, group) ? EXT3_SB(sb)->s_gdb_count : 0;
1851 * ext3_bg_num_gdb - number of blocks used by the group table in group
1852 * @sb: superblock for filesystem
1853 * @group: group number to check
1855 * Return the number of blocks used by the group descriptor table
1856 * (primary or backup) in this group. In the future there may be a
1857 * different number of descriptor blocks in each group.
1859 unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1861 unsigned long first_meta_bg =
1862 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_meta_bg);
1863 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1865 if (!EXT3_HAS_INCOMPAT_FEATURE(sb,EXT3_FEATURE_INCOMPAT_META_BG) ||
1866 metagroup < first_meta_bg)
1867 return ext3_bg_num_gdb_nometa(sb,group);
1869 return ext3_bg_num_gdb_meta(sb,group);