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, __func__,
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, __func__,
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, __func__,
162 "Cannot read block bitmap - "
163 "block_group = %d, block_bitmap = %u",
164 block_group, le32_to_cpu(desc->bg_block_bitmap));
167 ext3_valid_block_bitmap(sb, desc, block_group, bh);
169 * file system mounted not to panic on error, continue with corrupt
175 * The reservation window structure operations
176 * --------------------------------------------
177 * Operations include:
178 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
180 * We use a red-black tree to represent per-filesystem reservation
186 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
187 * @rb_root: root of per-filesystem reservation rb tree
188 * @verbose: verbose mode
189 * @fn: function which wishes to dump the reservation map
191 * If verbose is turned on, it will print the whole block reservation
192 * windows(start, end). Otherwise, it will only print out the "bad" windows,
193 * those windows that overlap with their immediate neighbors.
196 static void __rsv_window_dump(struct rb_root *root, int verbose,
200 struct ext3_reserve_window_node *rsv, *prev;
208 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
210 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
212 printk("reservation window 0x%p "
213 "start: %lu, end: %lu\n",
214 rsv, rsv->rsv_start, rsv->rsv_end);
215 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
216 printk("Bad reservation %p (start >= end)\n",
220 if (prev && prev->rsv_end >= rsv->rsv_start) {
221 printk("Bad reservation %p (prev->end >= start)\n",
227 printk("Restarting reservation walk in verbose mode\n");
235 printk("Window map complete.\n");
238 #define rsv_window_dump(root, verbose) \
239 __rsv_window_dump((root), (verbose), __func__)
241 #define rsv_window_dump(root, verbose) do {} while (0)
245 * goal_in_my_reservation()
246 * @rsv: inode's reservation window
247 * @grp_goal: given goal block relative to the allocation block group
248 * @group: the current allocation block group
249 * @sb: filesystem super block
251 * Test if the given goal block (group relative) is within the file's
252 * own block reservation window range.
254 * If the reservation window is outside the goal allocation group, return 0;
255 * grp_goal (given goal block) could be -1, which means no specific
256 * goal block. In this case, always return 1.
257 * If the goal block is within the reservation window, return 1;
258 * otherwise, return 0;
261 goal_in_my_reservation(struct ext3_reserve_window *rsv, ext3_grpblk_t grp_goal,
262 unsigned int group, struct super_block * sb)
264 ext3_fsblk_t group_first_block, group_last_block;
266 group_first_block = ext3_group_first_block_no(sb, group);
267 group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
269 if ((rsv->_rsv_start > group_last_block) ||
270 (rsv->_rsv_end < group_first_block))
272 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
273 || (grp_goal + group_first_block > rsv->_rsv_end)))
279 * search_reserve_window()
280 * @rb_root: root of reservation tree
281 * @goal: target allocation block
283 * Find the reserved window which includes the goal, or the previous one
284 * if the goal is not in any window.
285 * Returns NULL if there are no windows or if all windows start after the goal.
287 static struct ext3_reserve_window_node *
288 search_reserve_window(struct rb_root *root, ext3_fsblk_t goal)
290 struct rb_node *n = root->rb_node;
291 struct ext3_reserve_window_node *rsv;
297 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
299 if (goal < rsv->rsv_start)
301 else if (goal > rsv->rsv_end)
307 * We've fallen off the end of the tree: the goal wasn't inside
308 * any particular node. OK, the previous node must be to one
309 * side of the interval containing the goal. If it's the RHS,
310 * we need to back up one.
312 if (rsv->rsv_start > goal) {
313 n = rb_prev(&rsv->rsv_node);
314 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
320 * ext3_rsv_window_add() -- Insert a window to the block reservation rb tree.
322 * @rsv: reservation window to add
324 * Must be called with rsv_lock hold.
326 void ext3_rsv_window_add(struct super_block *sb,
327 struct ext3_reserve_window_node *rsv)
329 struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
330 struct rb_node *node = &rsv->rsv_node;
331 ext3_fsblk_t start = rsv->rsv_start;
333 struct rb_node ** p = &root->rb_node;
334 struct rb_node * parent = NULL;
335 struct ext3_reserve_window_node *this;
340 this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
342 if (start < this->rsv_start)
344 else if (start > this->rsv_end)
347 rsv_window_dump(root, 1);
352 rb_link_node(node, parent, p);
353 rb_insert_color(node, root);
357 * ext3_rsv_window_remove() -- unlink a window from the reservation rb tree
359 * @rsv: reservation window to remove
361 * Mark the block reservation window as not allocated, and unlink it
362 * from the filesystem reservation window rb tree. Must be called with
365 static void rsv_window_remove(struct super_block *sb,
366 struct ext3_reserve_window_node *rsv)
368 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
369 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
370 rsv->rsv_alloc_hit = 0;
371 rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
375 * rsv_is_empty() -- Check if the reservation window is allocated.
376 * @rsv: given reservation window to check
378 * returns 1 if the end block is EXT3_RESERVE_WINDOW_NOT_ALLOCATED.
380 static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
382 /* a valid reservation end block could not be 0 */
383 return rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
387 * ext3_init_block_alloc_info()
388 * @inode: file inode structure
390 * Allocate and initialize the reservation window structure, and
391 * link the window to the ext3 inode structure at last
393 * The reservation window structure is only dynamically allocated
394 * and linked to ext3 inode the first time the open file
395 * needs a new block. So, before every ext3_new_block(s) call, for
396 * regular files, we should check whether the reservation window
397 * structure exists or not. In the latter case, this function is called.
398 * Fail to do so will result in block reservation being turned off for that
401 * This function is called from ext3_get_blocks_handle(), also called
402 * when setting the reservation window size through ioctl before the file
403 * is open for write (needs block allocation).
405 * Needs truncate_mutex protection prior to call this function.
407 void ext3_init_block_alloc_info(struct inode *inode)
409 struct ext3_inode_info *ei = EXT3_I(inode);
410 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
411 struct super_block *sb = inode->i_sb;
413 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
415 struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node;
417 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
418 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
421 * if filesystem is mounted with NORESERVATION, the goal
422 * reservation window size is set to zero to indicate
423 * block reservation is off
425 if (!test_opt(sb, RESERVATION))
426 rsv->rsv_goal_size = 0;
428 rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS;
429 rsv->rsv_alloc_hit = 0;
430 block_i->last_alloc_logical_block = 0;
431 block_i->last_alloc_physical_block = 0;
433 ei->i_block_alloc_info = block_i;
437 * ext3_discard_reservation()
440 * Discard(free) block reservation window on last file close, or truncate
443 * It is being called in three cases:
444 * ext3_release_file(): last writer close the file
445 * ext3_clear_inode(): last iput(), when nobody link to this file.
446 * ext3_truncate(): when the block indirect map is about to change.
449 void ext3_discard_reservation(struct inode *inode)
451 struct ext3_inode_info *ei = EXT3_I(inode);
452 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
453 struct ext3_reserve_window_node *rsv;
454 spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
459 rsv = &block_i->rsv_window_node;
460 if (!rsv_is_empty(&rsv->rsv_window)) {
462 if (!rsv_is_empty(&rsv->rsv_window))
463 rsv_window_remove(inode->i_sb, rsv);
464 spin_unlock(rsv_lock);
469 * ext3_free_blocks_sb() -- Free given blocks and update quota
470 * @handle: handle to this transaction
472 * @block: start physcial block to free
473 * @count: number of blocks to free
474 * @pdquot_freed_blocks: pointer to quota
476 void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
477 ext3_fsblk_t block, unsigned long count,
478 unsigned long *pdquot_freed_blocks)
480 struct buffer_head *bitmap_bh = NULL;
481 struct buffer_head *gd_bh;
482 unsigned long block_group;
485 unsigned long overflow;
486 struct ext3_group_desc * desc;
487 struct ext3_super_block * es;
488 struct ext3_sb_info *sbi;
490 ext3_grpblk_t group_freed;
492 *pdquot_freed_blocks = 0;
495 if (block < le32_to_cpu(es->s_first_data_block) ||
496 block + count < block ||
497 block + count > le32_to_cpu(es->s_blocks_count)) {
498 ext3_error (sb, "ext3_free_blocks",
499 "Freeing blocks not in datazone - "
500 "block = "E3FSBLK", count = %lu", block, count);
504 ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
508 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
509 EXT3_BLOCKS_PER_GROUP(sb);
510 bit = (block - le32_to_cpu(es->s_first_data_block)) %
511 EXT3_BLOCKS_PER_GROUP(sb);
513 * Check to see if we are freeing blocks across a group
516 if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
517 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
521 bitmap_bh = read_block_bitmap(sb, block_group);
524 desc = ext3_get_group_desc (sb, block_group, &gd_bh);
528 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
529 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
530 in_range (block, le32_to_cpu(desc->bg_inode_table),
531 sbi->s_itb_per_group) ||
532 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
533 sbi->s_itb_per_group)) {
534 ext3_error (sb, "ext3_free_blocks",
535 "Freeing blocks in system zones - "
536 "Block = "E3FSBLK", count = %lu",
542 * We are about to start releasing blocks in the bitmap,
543 * so we need undo access.
545 /* @@@ check errors */
546 BUFFER_TRACE(bitmap_bh, "getting undo access");
547 err = ext3_journal_get_undo_access(handle, bitmap_bh);
552 * We are about to modify some metadata. Call the journal APIs
553 * to unshare ->b_data if a currently-committing transaction is
556 BUFFER_TRACE(gd_bh, "get_write_access");
557 err = ext3_journal_get_write_access(handle, gd_bh);
561 jbd_lock_bh_state(bitmap_bh);
563 for (i = 0, group_freed = 0; i < count; i++) {
565 * An HJ special. This is expensive...
567 #ifdef CONFIG_JBD_DEBUG
568 jbd_unlock_bh_state(bitmap_bh);
570 struct buffer_head *debug_bh;
571 debug_bh = sb_find_get_block(sb, block + i);
573 BUFFER_TRACE(debug_bh, "Deleted!");
574 if (!bh2jh(bitmap_bh)->b_committed_data)
575 BUFFER_TRACE(debug_bh,
576 "No commited data in bitmap");
577 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
581 jbd_lock_bh_state(bitmap_bh);
583 if (need_resched()) {
584 jbd_unlock_bh_state(bitmap_bh);
586 jbd_lock_bh_state(bitmap_bh);
588 /* @@@ This prevents newly-allocated data from being
589 * freed and then reallocated within the same
592 * Ideally we would want to allow that to happen, but to
593 * do so requires making journal_forget() capable of
594 * revoking the queued write of a data block, which
595 * implies blocking on the journal lock. *forget()
596 * cannot block due to truncate races.
598 * Eventually we can fix this by making journal_forget()
599 * return a status indicating whether or not it was able
600 * to revoke the buffer. On successful revoke, it is
601 * safe not to set the allocation bit in the committed
602 * bitmap, because we know that there is no outstanding
603 * activity on the buffer any more and so it is safe to
606 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
607 J_ASSERT_BH(bitmap_bh,
608 bh2jh(bitmap_bh)->b_committed_data != NULL);
609 ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
610 bh2jh(bitmap_bh)->b_committed_data);
613 * We clear the bit in the bitmap after setting the committed
614 * data bit, because this is the reverse order to that which
615 * the allocator uses.
617 BUFFER_TRACE(bitmap_bh, "clear bit");
618 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
619 bit + i, bitmap_bh->b_data)) {
620 jbd_unlock_bh_state(bitmap_bh);
621 ext3_error(sb, __func__,
622 "bit already cleared for block "E3FSBLK,
624 jbd_lock_bh_state(bitmap_bh);
625 BUFFER_TRACE(bitmap_bh, "bit already cleared");
630 jbd_unlock_bh_state(bitmap_bh);
632 spin_lock(sb_bgl_lock(sbi, block_group));
633 le16_add_cpu(&desc->bg_free_blocks_count, group_freed);
634 spin_unlock(sb_bgl_lock(sbi, block_group));
635 percpu_counter_add(&sbi->s_freeblocks_counter, count);
637 /* We dirtied the bitmap block */
638 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
639 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
641 /* And the group descriptor block */
642 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
643 ret = ext3_journal_dirty_metadata(handle, gd_bh);
645 *pdquot_freed_blocks += group_freed;
647 if (overflow && !err) {
655 ext3_std_error(sb, err);
660 * ext3_free_blocks() -- Free given blocks and update quota
661 * @handle: handle for this transaction
663 * @block: start physical block to free
664 * @count: number of blocks to count
666 void ext3_free_blocks(handle_t *handle, struct inode *inode,
667 ext3_fsblk_t block, unsigned long count)
669 struct super_block * sb;
670 unsigned long dquot_freed_blocks;
674 printk ("ext3_free_blocks: nonexistent device");
677 ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
678 if (dquot_freed_blocks)
679 vfs_dq_free_block(inode, dquot_freed_blocks);
684 * ext3_test_allocatable()
685 * @nr: given allocation block group
686 * @bh: bufferhead contains the bitmap of the given block group
688 * For ext3 allocations, we must not reuse any blocks which are
689 * allocated in the bitmap buffer's "last committed data" copy. This
690 * prevents deletes from freeing up the page for reuse until we have
691 * committed the delete transaction.
693 * If we didn't do this, then deleting something and reallocating it as
694 * data would allow the old block to be overwritten before the
695 * transaction committed (because we force data to disk before commit).
696 * This would lead to corruption if we crashed between overwriting the
697 * data and committing the delete.
699 * @@@ We may want to make this allocation behaviour conditional on
700 * data-writes at some point, and disable it for metadata allocations or
703 static int ext3_test_allocatable(ext3_grpblk_t nr, struct buffer_head *bh)
706 struct journal_head *jh = bh2jh(bh);
708 if (ext3_test_bit(nr, bh->b_data))
711 jbd_lock_bh_state(bh);
712 if (!jh->b_committed_data)
715 ret = !ext3_test_bit(nr, jh->b_committed_data);
716 jbd_unlock_bh_state(bh);
721 * bitmap_search_next_usable_block()
722 * @start: the starting block (group relative) of the search
723 * @bh: bufferhead contains the block group bitmap
724 * @maxblocks: the ending block (group relative) of the reservation
726 * The bitmap search --- search forward alternately through the actual
727 * bitmap on disk and the last-committed copy in journal, until we find a
728 * bit free in both bitmaps.
731 bitmap_search_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
732 ext3_grpblk_t maxblocks)
735 struct journal_head *jh = bh2jh(bh);
737 while (start < maxblocks) {
738 next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
739 if (next >= maxblocks)
741 if (ext3_test_allocatable(next, bh))
743 jbd_lock_bh_state(bh);
744 if (jh->b_committed_data)
745 start = ext3_find_next_zero_bit(jh->b_committed_data,
747 jbd_unlock_bh_state(bh);
753 * find_next_usable_block()
754 * @start: the starting block (group relative) to find next
755 * allocatable block in bitmap.
756 * @bh: bufferhead contains the block group bitmap
757 * @maxblocks: the ending block (group relative) for the search
759 * Find an allocatable block in a bitmap. We honor both the bitmap and
760 * its last-committed copy (if that exists), and perform the "most
761 * appropriate allocation" algorithm of looking for a free block near
762 * the initial goal; then for a free byte somewhere in the bitmap; then
763 * for any free bit in the bitmap.
766 find_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
767 ext3_grpblk_t maxblocks)
769 ext3_grpblk_t here, next;
774 * The goal was occupied; search forward for a free
775 * block within the next XX blocks.
777 * end_goal is more or less random, but it has to be
778 * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
779 * next 64-bit boundary is simple..
781 ext3_grpblk_t end_goal = (start + 63) & ~63;
782 if (end_goal > maxblocks)
783 end_goal = maxblocks;
784 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
785 if (here < end_goal && ext3_test_allocatable(here, bh))
787 ext3_debug("Bit not found near goal\n");
794 p = ((char *)bh->b_data) + (here >> 3);
795 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
796 next = (r - ((char *)bh->b_data)) << 3;
798 if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
802 * The bitmap search --- search forward alternately through the actual
803 * bitmap and the last-committed copy until we find a bit free in
806 here = bitmap_search_next_usable_block(here, bh, maxblocks);
812 * @block: the free block (group relative) to allocate
813 * @bh: the bufferhead containts the block group bitmap
815 * We think we can allocate this block in this bitmap. Try to set the bit.
816 * If that succeeds then check that nobody has allocated and then freed the
817 * block since we saw that is was not marked in b_committed_data. If it _was_
818 * allocated and freed then clear the bit in the bitmap again and return
822 claim_block(spinlock_t *lock, ext3_grpblk_t block, struct buffer_head *bh)
824 struct journal_head *jh = bh2jh(bh);
827 if (ext3_set_bit_atomic(lock, block, bh->b_data))
829 jbd_lock_bh_state(bh);
830 if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
831 ext3_clear_bit_atomic(lock, block, bh->b_data);
836 jbd_unlock_bh_state(bh);
841 * ext3_try_to_allocate()
843 * @handle: handle to this transaction
844 * @group: given allocation block group
845 * @bitmap_bh: bufferhead holds the block bitmap
846 * @grp_goal: given target block within the group
847 * @count: target number of blocks to allocate
848 * @my_rsv: reservation window
850 * Attempt to allocate blocks within a give range. Set the range of allocation
851 * first, then find the first free bit(s) from the bitmap (within the range),
852 * and at last, allocate the blocks by claiming the found free bit as allocated.
854 * To set the range of this allocation:
855 * if there is a reservation window, only try to allocate block(s) from the
856 * file's own reservation window;
857 * Otherwise, the allocation range starts from the give goal block, ends at
858 * the block group's last block.
860 * If we failed to allocate the desired block then we may end up crossing to a
861 * new bitmap. In that case we must release write access to the old one via
862 * ext3_journal_release_buffer(), else we'll run out of credits.
865 ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
866 struct buffer_head *bitmap_bh, ext3_grpblk_t grp_goal,
867 unsigned long *count, struct ext3_reserve_window *my_rsv)
869 ext3_fsblk_t group_first_block;
870 ext3_grpblk_t start, end;
871 unsigned long num = 0;
873 /* we do allocation within the reservation window if we have a window */
875 group_first_block = ext3_group_first_block_no(sb, group);
876 if (my_rsv->_rsv_start >= group_first_block)
877 start = my_rsv->_rsv_start - group_first_block;
879 /* reservation window cross group boundary */
881 end = my_rsv->_rsv_end - group_first_block + 1;
882 if (end > EXT3_BLOCKS_PER_GROUP(sb))
883 /* reservation window crosses group boundary */
884 end = EXT3_BLOCKS_PER_GROUP(sb);
885 if ((start <= grp_goal) && (grp_goal < end))
894 end = EXT3_BLOCKS_PER_GROUP(sb);
897 BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
900 if (grp_goal < 0 || !ext3_test_allocatable(grp_goal, bitmap_bh)) {
901 grp_goal = find_next_usable_block(start, bitmap_bh, end);
907 for (i = 0; i < 7 && grp_goal > start &&
908 ext3_test_allocatable(grp_goal - 1,
916 if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group),
917 grp_goal, bitmap_bh)) {
919 * The block was allocated by another thread, or it was
920 * allocated and then freed by another thread
930 while (num < *count && grp_goal < end
931 && ext3_test_allocatable(grp_goal, bitmap_bh)
932 && claim_block(sb_bgl_lock(EXT3_SB(sb), group),
933 grp_goal, bitmap_bh)) {
938 return grp_goal - num;
945 * find_next_reservable_window():
946 * find a reservable space within the given range.
947 * It does not allocate the reservation window for now:
948 * alloc_new_reservation() will do the work later.
950 * @search_head: the head of the searching list;
951 * This is not necessarily the list head of the whole filesystem
953 * We have both head and start_block to assist the search
954 * for the reservable space. The list starts from head,
955 * but we will shift to the place where start_block is,
956 * then start from there, when looking for a reservable space.
958 * @size: the target new reservation window size
960 * @group_first_block: the first block we consider to start
961 * the real search from
964 * the maximum block number that our goal reservable space
965 * could start from. This is normally the last block in this
966 * group. The search will end when we found the start of next
967 * possible reservable space is out of this boundary.
968 * This could handle the cross boundary reservation window
971 * basically we search from the given range, rather than the whole
972 * reservation double linked list, (start_block, last_block)
973 * to find a free region that is of my size and has not
977 static int find_next_reservable_window(
978 struct ext3_reserve_window_node *search_head,
979 struct ext3_reserve_window_node *my_rsv,
980 struct super_block * sb,
981 ext3_fsblk_t start_block,
982 ext3_fsblk_t last_block)
984 struct rb_node *next;
985 struct ext3_reserve_window_node *rsv, *prev;
987 int size = my_rsv->rsv_goal_size;
989 /* TODO: make the start of the reservation window byte-aligned */
990 /* cur = *start_block & ~7;*/
997 if (cur <= rsv->rsv_end)
998 cur = rsv->rsv_end + 1;
1001 * in the case we could not find a reservable space
1002 * that is what is expected, during the re-search, we could
1003 * remember what's the largest reservable space we could have
1004 * and return that one.
1006 * For now it will fail if we could not find the reservable
1007 * space with expected-size (or more)...
1009 if (cur > last_block)
1010 return -1; /* fail */
1013 next = rb_next(&rsv->rsv_node);
1014 rsv = rb_entry(next,struct ext3_reserve_window_node,rsv_node);
1017 * Reached the last reservation, we can just append to the
1023 if (cur + size <= rsv->rsv_start) {
1025 * Found a reserveable space big enough. We could
1026 * have a reservation across the group boundary here
1032 * we come here either :
1033 * when we reach the end of the whole list,
1034 * and there is empty reservable space after last entry in the list.
1035 * append it to the end of the list.
1037 * or we found one reservable space in the middle of the list,
1038 * return the reservation window that we could append to.
1042 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
1043 rsv_window_remove(sb, my_rsv);
1046 * Let's book the whole avaliable window for now. We will check the
1047 * disk bitmap later and then, if there are free blocks then we adjust
1048 * the window size if it's larger than requested.
1049 * Otherwise, we will remove this node from the tree next time
1050 * call find_next_reservable_window.
1052 my_rsv->rsv_start = cur;
1053 my_rsv->rsv_end = cur + size - 1;
1054 my_rsv->rsv_alloc_hit = 0;
1057 ext3_rsv_window_add(sb, my_rsv);
1063 * alloc_new_reservation()--allocate a new reservation window
1065 * To make a new reservation, we search part of the filesystem
1066 * reservation list (the list that inside the group). We try to
1067 * allocate a new reservation window near the allocation goal,
1068 * or the beginning of the group, if there is no goal.
1070 * We first find a reservable space after the goal, then from
1071 * there, we check the bitmap for the first free block after
1072 * it. If there is no free block until the end of group, then the
1073 * whole group is full, we failed. Otherwise, check if the free
1074 * block is inside the expected reservable space, if so, we
1076 * If the first free block is outside the reservable space, then
1077 * start from the first free block, we search for next available
1080 * on succeed, a new reservation will be found and inserted into the list
1081 * It contains at least one free block, and it does not overlap with other
1082 * reservation windows.
1084 * failed: we failed to find a reservation window in this group
1086 * @rsv: the reservation
1088 * @grp_goal: The goal (group-relative). It is where the search for a
1089 * free reservable space should start from.
1090 * if we have a grp_goal(grp_goal >0 ), then start from there,
1091 * no grp_goal(grp_goal = -1), we start from the first block
1094 * @sb: the super block
1095 * @group: the group we are trying to allocate in
1096 * @bitmap_bh: the block group block bitmap
1099 static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
1100 ext3_grpblk_t grp_goal, struct super_block *sb,
1101 unsigned int group, struct buffer_head *bitmap_bh)
1103 struct ext3_reserve_window_node *search_head;
1104 ext3_fsblk_t group_first_block, group_end_block, start_block;
1105 ext3_grpblk_t first_free_block;
1106 struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
1109 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1111 group_first_block = ext3_group_first_block_no(sb, group);
1112 group_end_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1115 start_block = group_first_block;
1117 start_block = grp_goal + group_first_block;
1119 size = my_rsv->rsv_goal_size;
1121 if (!rsv_is_empty(&my_rsv->rsv_window)) {
1123 * if the old reservation is cross group boundary
1124 * and if the goal is inside the old reservation window,
1125 * we will come here when we just failed to allocate from
1126 * the first part of the window. We still have another part
1127 * that belongs to the next group. In this case, there is no
1128 * point to discard our window and try to allocate a new one
1129 * in this group(which will fail). we should
1130 * keep the reservation window, just simply move on.
1132 * Maybe we could shift the start block of the reservation
1133 * window to the first block of next group.
1136 if ((my_rsv->rsv_start <= group_end_block) &&
1137 (my_rsv->rsv_end > group_end_block) &&
1138 (start_block >= my_rsv->rsv_start))
1141 if ((my_rsv->rsv_alloc_hit >
1142 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
1144 * if the previously allocation hit ratio is
1145 * greater than 1/2, then we double the size of
1146 * the reservation window the next time,
1147 * otherwise we keep the same size window
1150 if (size > EXT3_MAX_RESERVE_BLOCKS)
1151 size = EXT3_MAX_RESERVE_BLOCKS;
1152 my_rsv->rsv_goal_size= size;
1156 spin_lock(rsv_lock);
1158 * shift the search start to the window near the goal block
1160 search_head = search_reserve_window(fs_rsv_root, start_block);
1163 * find_next_reservable_window() simply finds a reservable window
1164 * inside the given range(start_block, group_end_block).
1166 * To make sure the reservation window has a free bit inside it, we
1167 * need to check the bitmap after we found a reservable window.
1170 ret = find_next_reservable_window(search_head, my_rsv, sb,
1171 start_block, group_end_block);
1174 if (!rsv_is_empty(&my_rsv->rsv_window))
1175 rsv_window_remove(sb, my_rsv);
1176 spin_unlock(rsv_lock);
1181 * On success, find_next_reservable_window() returns the
1182 * reservation window where there is a reservable space after it.
1183 * Before we reserve this reservable space, we need
1184 * to make sure there is at least a free block inside this region.
1186 * searching the first free bit on the block bitmap and copy of
1187 * last committed bitmap alternatively, until we found a allocatable
1188 * block. Search start from the start block of the reservable space
1191 spin_unlock(rsv_lock);
1192 first_free_block = bitmap_search_next_usable_block(
1193 my_rsv->rsv_start - group_first_block,
1194 bitmap_bh, group_end_block - group_first_block + 1);
1196 if (first_free_block < 0) {
1198 * no free block left on the bitmap, no point
1199 * to reserve the space. return failed.
1201 spin_lock(rsv_lock);
1202 if (!rsv_is_empty(&my_rsv->rsv_window))
1203 rsv_window_remove(sb, my_rsv);
1204 spin_unlock(rsv_lock);
1205 return -1; /* failed */
1208 start_block = first_free_block + group_first_block;
1210 * check if the first free block is within the
1211 * free space we just reserved
1213 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1214 return 0; /* success */
1216 * if the first free bit we found is out of the reservable space
1217 * continue search for next reservable space,
1218 * start from where the free block is,
1219 * we also shift the list head to where we stopped last time
1221 search_head = my_rsv;
1222 spin_lock(rsv_lock);
1227 * try_to_extend_reservation()
1228 * @my_rsv: given reservation window
1230 * @size: the delta to extend
1232 * Attempt to expand the reservation window large enough to have
1233 * required number of free blocks
1235 * Since ext3_try_to_allocate() will always allocate blocks within
1236 * the reservation window range, if the window size is too small,
1237 * multiple blocks allocation has to stop at the end of the reservation
1238 * window. To make this more efficient, given the total number of
1239 * blocks needed and the current size of the window, we try to
1240 * expand the reservation window size if necessary on a best-effort
1241 * basis before ext3_new_blocks() tries to allocate blocks,
1243 static void try_to_extend_reservation(struct ext3_reserve_window_node *my_rsv,
1244 struct super_block *sb, int size)
1246 struct ext3_reserve_window_node *next_rsv;
1247 struct rb_node *next;
1248 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1250 if (!spin_trylock(rsv_lock))
1253 next = rb_next(&my_rsv->rsv_node);
1256 my_rsv->rsv_end += size;
1258 next_rsv = rb_entry(next, struct ext3_reserve_window_node, rsv_node);
1260 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1261 my_rsv->rsv_end += size;
1263 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1265 spin_unlock(rsv_lock);
1269 * ext3_try_to_allocate_with_rsv()
1271 * @handle: handle to this transaction
1272 * @group: given allocation block group
1273 * @bitmap_bh: bufferhead holds the block bitmap
1274 * @grp_goal: given target block within the group
1275 * @count: target number of blocks to allocate
1276 * @my_rsv: reservation window
1277 * @errp: pointer to store the error code
1279 * This is the main function used to allocate a new block and its reservation
1282 * Each time when a new block allocation is need, first try to allocate from
1283 * its own reservation. If it does not have a reservation window, instead of
1284 * looking for a free bit on bitmap first, then look up the reservation list to
1285 * see if it is inside somebody else's reservation window, we try to allocate a
1286 * reservation window for it starting from the goal first. Then do the block
1287 * allocation within the reservation window.
1289 * This will avoid keeping on searching the reservation list again and
1290 * again when somebody is looking for a free block (without
1291 * reservation), and there are lots of free blocks, but they are all
1294 * We use a red-black tree for the per-filesystem reservation list.
1297 static ext3_grpblk_t
1298 ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1299 unsigned int group, struct buffer_head *bitmap_bh,
1300 ext3_grpblk_t grp_goal,
1301 struct ext3_reserve_window_node * my_rsv,
1302 unsigned long *count, int *errp)
1304 ext3_fsblk_t group_first_block, group_last_block;
1305 ext3_grpblk_t ret = 0;
1307 unsigned long num = *count;
1312 * Make sure we use undo access for the bitmap, because it is critical
1313 * that we do the frozen_data COW on bitmap buffers in all cases even
1314 * if the buffer is in BJ_Forget state in the committing transaction.
1316 BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1317 fatal = ext3_journal_get_undo_access(handle, bitmap_bh);
1324 * we don't deal with reservation when
1325 * filesystem is mounted without reservation
1326 * or the file is not a regular file
1327 * or last attempt to allocate a block with reservation turned on failed
1329 if (my_rsv == NULL ) {
1330 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1331 grp_goal, count, NULL);
1335 * grp_goal is a group relative block number (if there is a goal)
1336 * 0 <= grp_goal < EXT3_BLOCKS_PER_GROUP(sb)
1337 * first block is a filesystem wide block number
1338 * first block is the block number of the first block in this group
1340 group_first_block = ext3_group_first_block_no(sb, group);
1341 group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1344 * Basically we will allocate a new block from inode's reservation
1347 * We need to allocate a new reservation window, if:
1348 * a) inode does not have a reservation window; or
1349 * b) last attempt to allocate a block from existing reservation
1351 * c) we come here with a goal and with a reservation window
1353 * We do not need to allocate a new reservation window if we come here
1354 * at the beginning with a goal and the goal is inside the window, or
1355 * we don't have a goal but already have a reservation window.
1356 * then we could go to allocate from the reservation window directly.
1359 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1360 !goal_in_my_reservation(&my_rsv->rsv_window,
1361 grp_goal, group, sb)) {
1362 if (my_rsv->rsv_goal_size < *count)
1363 my_rsv->rsv_goal_size = *count;
1364 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1369 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1370 grp_goal, group, sb))
1372 } else if (grp_goal >= 0) {
1373 int curr = my_rsv->rsv_end -
1374 (grp_goal + group_first_block) + 1;
1377 try_to_extend_reservation(my_rsv, sb,
1381 if ((my_rsv->rsv_start > group_last_block) ||
1382 (my_rsv->rsv_end < group_first_block)) {
1383 rsv_window_dump(&EXT3_SB(sb)->s_rsv_window_root, 1);
1386 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1387 grp_goal, &num, &my_rsv->rsv_window);
1389 my_rsv->rsv_alloc_hit += num;
1391 break; /* succeed */
1397 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1399 fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1407 BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1408 ext3_journal_release_buffer(handle, bitmap_bh);
1413 * ext3_has_free_blocks()
1414 * @sbi: in-core super block structure.
1416 * Check if filesystem has at least 1 free block available for allocation.
1418 static int ext3_has_free_blocks(struct ext3_sb_info *sbi)
1420 ext3_fsblk_t free_blocks, root_blocks;
1422 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1423 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1424 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1425 sbi->s_resuid != current_fsuid() &&
1426 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1433 * ext3_should_retry_alloc()
1435 * @retries number of attemps has been made
1437 * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1438 * it is profitable to retry the operation, this function will wait
1439 * for the current or commiting transaction to complete, and then
1442 * if the total number of retries exceed three times, return FALSE.
1444 int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1446 if (!ext3_has_free_blocks(EXT3_SB(sb)) || (*retries)++ > 3)
1449 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1451 return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1455 * ext3_new_blocks() -- core block(s) allocation function
1456 * @handle: handle to this transaction
1457 * @inode: file inode
1458 * @goal: given target block(filesystem wide)
1459 * @count: target number of blocks to allocate
1462 * ext3_new_blocks uses a goal block to assist allocation. It tries to
1463 * allocate block(s) from the block group contains the goal block first. If that
1464 * fails, it will try to allocate block(s) from other block groups without
1465 * any specific goal block.
1468 ext3_fsblk_t ext3_new_blocks(handle_t *handle, struct inode *inode,
1469 ext3_fsblk_t goal, unsigned long *count, int *errp)
1471 struct buffer_head *bitmap_bh = NULL;
1472 struct buffer_head *gdp_bh;
1475 ext3_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1476 ext3_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1477 ext3_fsblk_t ret_block; /* filesyetem-wide allocated block */
1478 int bgi; /* blockgroup iteration index */
1480 int performed_allocation = 0;
1481 ext3_grpblk_t free_blocks; /* number of free blocks in a group */
1482 struct super_block *sb;
1483 struct ext3_group_desc *gdp;
1484 struct ext3_super_block *es;
1485 struct ext3_sb_info *sbi;
1486 struct ext3_reserve_window_node *my_rsv = NULL;
1487 struct ext3_block_alloc_info *block_i;
1488 unsigned short windowsz = 0;
1490 static int goal_hits, goal_attempts;
1492 unsigned long ngroups;
1493 unsigned long num = *count;
1498 printk("ext3_new_block: nonexistent device");
1503 * Check quota for allocation of this block.
1505 if (vfs_dq_alloc_block(inode, num)) {
1511 es = EXT3_SB(sb)->s_es;
1512 ext3_debug("goal=%lu.\n", goal);
1514 * Allocate a block from reservation only when
1515 * filesystem is mounted with reservation(default,-o reservation), and
1516 * it's a regular file, and
1517 * the desired window size is greater than 0 (One could use ioctl
1518 * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1519 * reservation on that particular file)
1521 block_i = EXT3_I(inode)->i_block_alloc_info;
1522 if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1523 my_rsv = &block_i->rsv_window_node;
1525 if (!ext3_has_free_blocks(sbi)) {
1531 * First, test whether the goal block is free.
1533 if (goal < le32_to_cpu(es->s_first_data_block) ||
1534 goal >= le32_to_cpu(es->s_blocks_count))
1535 goal = le32_to_cpu(es->s_first_data_block);
1536 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1537 EXT3_BLOCKS_PER_GROUP(sb);
1538 goal_group = group_no;
1540 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1544 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1546 * if there is not enough free blocks to make a new resevation
1547 * turn off reservation for this allocation
1549 if (my_rsv && (free_blocks < windowsz)
1550 && (free_blocks > 0)
1551 && (rsv_is_empty(&my_rsv->rsv_window)))
1554 if (free_blocks > 0) {
1555 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1556 EXT3_BLOCKS_PER_GROUP(sb));
1557 bitmap_bh = read_block_bitmap(sb, group_no);
1560 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1561 group_no, bitmap_bh, grp_target_blk,
1562 my_rsv, &num, &fatal);
1565 if (grp_alloc_blk >= 0)
1569 ngroups = EXT3_SB(sb)->s_groups_count;
1573 * Now search the rest of the groups. We assume that
1574 * group_no and gdp correctly point to the last group visited.
1576 for (bgi = 0; bgi < ngroups; bgi++) {
1578 if (group_no >= ngroups)
1580 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1583 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1585 * skip this group if the number of
1586 * free blocks is less than half of the reservation
1589 if (my_rsv && (free_blocks <= (windowsz/2)))
1593 bitmap_bh = read_block_bitmap(sb, group_no);
1597 * try to allocate block(s) from this group, without a goal(-1).
1599 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1600 group_no, bitmap_bh, -1, my_rsv,
1604 if (grp_alloc_blk >= 0)
1608 * We may end up a bogus ealier ENOSPC error due to
1609 * filesystem is "full" of reservations, but
1610 * there maybe indeed free blocks avaliable on disk
1611 * In this case, we just forget about the reservations
1612 * just do block allocation as without reservations.
1617 group_no = goal_group;
1620 /* No space left on the device */
1626 ext3_debug("using block group %d(%d)\n",
1627 group_no, gdp->bg_free_blocks_count);
1629 BUFFER_TRACE(gdp_bh, "get_write_access");
1630 fatal = ext3_journal_get_write_access(handle, gdp_bh);
1634 ret_block = grp_alloc_blk + ext3_group_first_block_no(sb, group_no);
1636 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1637 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1638 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1639 EXT3_SB(sb)->s_itb_per_group) ||
1640 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1641 EXT3_SB(sb)->s_itb_per_group)) {
1642 ext3_error(sb, "ext3_new_block",
1643 "Allocating block in system zone - "
1644 "blocks from "E3FSBLK", length %lu",
1647 * claim_block() marked the blocks we allocated as in use. So we
1648 * may want to selectively mark some of the blocks as free.
1653 performed_allocation = 1;
1655 #ifdef CONFIG_JBD_DEBUG
1657 struct buffer_head *debug_bh;
1659 /* Record bitmap buffer state in the newly allocated block */
1660 debug_bh = sb_find_get_block(sb, ret_block);
1662 BUFFER_TRACE(debug_bh, "state when allocated");
1663 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1667 jbd_lock_bh_state(bitmap_bh);
1668 spin_lock(sb_bgl_lock(sbi, group_no));
1669 if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1672 for (i = 0; i < num; i++) {
1673 if (ext3_test_bit(grp_alloc_blk+i,
1674 bh2jh(bitmap_bh)->b_committed_data)) {
1675 printk("%s: block was unexpectedly set in "
1676 "b_committed_data\n", __func__);
1680 ext3_debug("found bit %d\n", grp_alloc_blk);
1681 spin_unlock(sb_bgl_lock(sbi, group_no));
1682 jbd_unlock_bh_state(bitmap_bh);
1685 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1686 ext3_error(sb, "ext3_new_block",
1687 "block("E3FSBLK") >= blocks count(%d) - "
1688 "block_group = %d, es == %p ", ret_block,
1689 le32_to_cpu(es->s_blocks_count), group_no, es);
1694 * It is up to the caller to add the new buffer to a journal
1695 * list of some description. We don't know in advance whether
1696 * the caller wants to use it as metadata or data.
1698 ext3_debug("allocating block %lu. Goal hits %d of %d.\n",
1699 ret_block, goal_hits, goal_attempts);
1701 spin_lock(sb_bgl_lock(sbi, group_no));
1702 le16_add_cpu(&gdp->bg_free_blocks_count, -num);
1703 spin_unlock(sb_bgl_lock(sbi, group_no));
1704 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1706 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1707 err = ext3_journal_dirty_metadata(handle, gdp_bh);
1716 vfs_dq_free_block(inode, *count-num);
1725 ext3_std_error(sb, fatal);
1728 * Undo the block allocation
1730 if (!performed_allocation)
1731 vfs_dq_free_block(inode, *count);
1736 ext3_fsblk_t ext3_new_block(handle_t *handle, struct inode *inode,
1737 ext3_fsblk_t goal, int *errp)
1739 unsigned long count = 1;
1741 return ext3_new_blocks(handle, inode, goal, &count, errp);
1745 * ext3_count_free_blocks() -- count filesystem free blocks
1748 * Adds up the number of free blocks from each block group.
1750 ext3_fsblk_t ext3_count_free_blocks(struct super_block *sb)
1752 ext3_fsblk_t desc_count;
1753 struct ext3_group_desc *gdp;
1755 unsigned long ngroups = EXT3_SB(sb)->s_groups_count;
1757 struct ext3_super_block *es;
1758 ext3_fsblk_t bitmap_count;
1760 struct buffer_head *bitmap_bh = NULL;
1762 es = EXT3_SB(sb)->s_es;
1768 for (i = 0; i < ngroups; i++) {
1769 gdp = ext3_get_group_desc(sb, i, NULL);
1772 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1774 bitmap_bh = read_block_bitmap(sb, i);
1775 if (bitmap_bh == NULL)
1778 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1779 printk("group %d: stored = %d, counted = %lu\n",
1780 i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1784 printk("ext3_count_free_blocks: stored = "E3FSBLK
1785 ", computed = "E3FSBLK", "E3FSBLK"\n",
1786 le32_to_cpu(es->s_free_blocks_count),
1787 desc_count, bitmap_count);
1788 return bitmap_count;
1792 for (i = 0; i < ngroups; i++) {
1793 gdp = ext3_get_group_desc(sb, i, NULL);
1796 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1803 static inline int test_root(int a, int b)
1812 static int ext3_group_sparse(int group)
1818 return (test_root(group, 7) || test_root(group, 5) ||
1819 test_root(group, 3));
1823 * ext3_bg_has_super - number of blocks used by the superblock in group
1824 * @sb: superblock for filesystem
1825 * @group: group number to check
1827 * Return the number of blocks used by the superblock (primary or backup)
1828 * in this group. Currently this will be only 0 or 1.
1830 int ext3_bg_has_super(struct super_block *sb, int group)
1832 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1833 EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1834 !ext3_group_sparse(group))
1839 static unsigned long ext3_bg_num_gdb_meta(struct super_block *sb, int group)
1841 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1842 unsigned long first = metagroup * EXT3_DESC_PER_BLOCK(sb);
1843 unsigned long last = first + EXT3_DESC_PER_BLOCK(sb) - 1;
1845 if (group == first || group == first + 1 || group == last)
1850 static unsigned long ext3_bg_num_gdb_nometa(struct super_block *sb, int group)
1852 return ext3_bg_has_super(sb, group) ? EXT3_SB(sb)->s_gdb_count : 0;
1856 * ext3_bg_num_gdb - number of blocks used by the group table in group
1857 * @sb: superblock for filesystem
1858 * @group: group number to check
1860 * Return the number of blocks used by the group descriptor table
1861 * (primary or backup) in this group. In the future there may be a
1862 * different number of descriptor blocks in each group.
1864 unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1866 unsigned long first_meta_bg =
1867 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_meta_bg);
1868 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1870 if (!EXT3_HAS_INCOMPAT_FEATURE(sb,EXT3_FEATURE_INCOMPAT_META_BG) ||
1871 metagroup < first_meta_bg)
1872 return ext3_bg_num_gdb_nometa(sb,group);
1874 return ext3_bg_num_gdb_meta(sb,group);