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_read_super).
39 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
41 struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb,
42 unsigned int block_group,
43 struct buffer_head ** bh)
45 unsigned long group_desc;
47 struct ext3_group_desc * desc;
48 struct ext3_sb_info *sbi = EXT3_SB(sb);
50 if (block_group >= sbi->s_groups_count) {
51 ext3_error (sb, "ext3_get_group_desc",
52 "block_group >= groups_count - "
53 "block_group = %d, groups_count = %lu",
54 block_group, sbi->s_groups_count);
60 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
61 offset = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
62 if (!sbi->s_group_desc[group_desc]) {
63 ext3_error (sb, "ext3_get_group_desc",
64 "Group descriptor not loaded - "
65 "block_group = %d, group_desc = %lu, desc = %lu",
66 block_group, group_desc, offset);
70 desc = (struct ext3_group_desc *) sbi->s_group_desc[group_desc]->b_data;
72 *bh = sbi->s_group_desc[group_desc];
77 * Read the bitmap for a given block_group, reading into the specified
78 * slot in the superblock's bitmap cache.
80 * Return buffer_head on success or NULL in case of failure.
82 static struct buffer_head *
83 read_block_bitmap(struct super_block *sb, unsigned int block_group)
85 struct ext3_group_desc * desc;
86 struct buffer_head * bh = NULL;
88 desc = ext3_get_group_desc (sb, block_group, NULL);
91 bh = sb_bread(sb, le32_to_cpu(desc->bg_block_bitmap));
93 ext3_error (sb, "read_block_bitmap",
94 "Cannot read block bitmap - "
95 "block_group = %d, block_bitmap = %u",
96 block_group, le32_to_cpu(desc->bg_block_bitmap));
101 * The reservation window structure operations
102 * --------------------------------------------
103 * Operations include:
104 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
106 * We use sorted double linked list for the per-filesystem reservation
107 * window list. (like in vm_region).
109 * Initially, we keep those small operations in the abstract functions,
110 * so later if we need a better searching tree than double linked-list,
111 * we could easily switch to that without changing too much
115 static void __rsv_window_dump(struct rb_root *root, int verbose,
119 struct ext3_reserve_window_node *rsv, *prev;
127 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
129 rsv = list_entry(n, struct ext3_reserve_window_node, rsv_node);
131 printk("reservation window 0x%p "
132 "start: %d, end: %d\n",
133 rsv, rsv->rsv_start, rsv->rsv_end);
134 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
135 printk("Bad reservation %p (start >= end)\n",
139 if (prev && prev->rsv_end >= rsv->rsv_start) {
140 printk("Bad reservation %p (prev->end >= start)\n",
146 printk("Restarting reservation walk in verbose mode\n");
154 printk("Window map complete.\n");
158 #define rsv_window_dump(root, verbose) \
159 __rsv_window_dump((root), (verbose), __FUNCTION__)
161 #define rsv_window_dump(root, verbose) do {} while (0)
165 goal_in_my_reservation(struct ext3_reserve_window *rsv, ext3_grpblk_t grp_goal,
166 unsigned int group, struct super_block * sb)
168 ext3_fsblk_t group_first_block, group_last_block;
170 group_first_block = ext3_group_first_block_no(sb, group);
171 group_last_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1;
173 if ((rsv->_rsv_start > group_last_block) ||
174 (rsv->_rsv_end < group_first_block))
176 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
177 || (grp_goal + group_first_block > rsv->_rsv_end)))
183 * Find the reserved window which includes the goal, or the previous one
184 * if the goal is not in any window.
185 * Returns NULL if there are no windows or if all windows start after the goal.
187 static struct ext3_reserve_window_node *
188 search_reserve_window(struct rb_root *root, ext3_fsblk_t goal)
190 struct rb_node *n = root->rb_node;
191 struct ext3_reserve_window_node *rsv;
197 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
199 if (goal < rsv->rsv_start)
201 else if (goal > rsv->rsv_end)
207 * We've fallen off the end of the tree: the goal wasn't inside
208 * any particular node. OK, the previous node must be to one
209 * side of the interval containing the goal. If it's the RHS,
210 * we need to back up one.
212 if (rsv->rsv_start > goal) {
213 n = rb_prev(&rsv->rsv_node);
214 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
219 void ext3_rsv_window_add(struct super_block *sb,
220 struct ext3_reserve_window_node *rsv)
222 struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
223 struct rb_node *node = &rsv->rsv_node;
224 ext3_fsblk_t start = rsv->rsv_start;
226 struct rb_node ** p = &root->rb_node;
227 struct rb_node * parent = NULL;
228 struct ext3_reserve_window_node *this;
233 this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
235 if (start < this->rsv_start)
237 else if (start > this->rsv_end)
243 rb_link_node(node, parent, p);
244 rb_insert_color(node, root);
247 static void rsv_window_remove(struct super_block *sb,
248 struct ext3_reserve_window_node *rsv)
250 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
251 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
252 rsv->rsv_alloc_hit = 0;
253 rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
256 static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
258 /* a valid reservation end block could not be 0 */
259 return (rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED);
261 void ext3_init_block_alloc_info(struct inode *inode)
263 struct ext3_inode_info *ei = EXT3_I(inode);
264 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
265 struct super_block *sb = inode->i_sb;
267 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
269 struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node;
271 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
272 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
275 * if filesystem is mounted with NORESERVATION, the goal
276 * reservation window size is set to zero to indicate
277 * block reservation is off
279 if (!test_opt(sb, RESERVATION))
280 rsv->rsv_goal_size = 0;
282 rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS;
283 rsv->rsv_alloc_hit = 0;
284 block_i->last_alloc_logical_block = 0;
285 block_i->last_alloc_physical_block = 0;
287 ei->i_block_alloc_info = block_i;
290 void ext3_discard_reservation(struct inode *inode)
292 struct ext3_inode_info *ei = EXT3_I(inode);
293 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
294 struct ext3_reserve_window_node *rsv;
295 spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
300 rsv = &block_i->rsv_window_node;
301 if (!rsv_is_empty(&rsv->rsv_window)) {
303 if (!rsv_is_empty(&rsv->rsv_window))
304 rsv_window_remove(inode->i_sb, rsv);
305 spin_unlock(rsv_lock);
309 /* Free given blocks, update quota and i_blocks field */
310 void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
311 ext3_fsblk_t block, unsigned long count,
312 unsigned long *pdquot_freed_blocks)
314 struct buffer_head *bitmap_bh = NULL;
315 struct buffer_head *gd_bh;
316 unsigned long block_group;
319 unsigned long overflow;
320 struct ext3_group_desc * desc;
321 struct ext3_super_block * es;
322 struct ext3_sb_info *sbi;
324 ext3_grpblk_t group_freed;
326 *pdquot_freed_blocks = 0;
329 if (block < le32_to_cpu(es->s_first_data_block) ||
330 block + count < block ||
331 block + count > le32_to_cpu(es->s_blocks_count)) {
332 ext3_error (sb, "ext3_free_blocks",
333 "Freeing blocks not in datazone - "
334 "block = "E3FSBLK", count = %lu", block, count);
338 ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
342 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
343 EXT3_BLOCKS_PER_GROUP(sb);
344 bit = (block - le32_to_cpu(es->s_first_data_block)) %
345 EXT3_BLOCKS_PER_GROUP(sb);
347 * Check to see if we are freeing blocks across a group
350 if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
351 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
355 bitmap_bh = read_block_bitmap(sb, block_group);
358 desc = ext3_get_group_desc (sb, block_group, &gd_bh);
362 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
363 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
364 in_range (block, le32_to_cpu(desc->bg_inode_table),
365 sbi->s_itb_per_group) ||
366 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
367 sbi->s_itb_per_group))
368 ext3_error (sb, "ext3_free_blocks",
369 "Freeing blocks in system zones - "
370 "Block = "E3FSBLK", count = %lu",
374 * We are about to start releasing blocks in the bitmap,
375 * so we need undo access.
377 /* @@@ check errors */
378 BUFFER_TRACE(bitmap_bh, "getting undo access");
379 err = ext3_journal_get_undo_access(handle, bitmap_bh);
384 * We are about to modify some metadata. Call the journal APIs
385 * to unshare ->b_data if a currently-committing transaction is
388 BUFFER_TRACE(gd_bh, "get_write_access");
389 err = ext3_journal_get_write_access(handle, gd_bh);
393 jbd_lock_bh_state(bitmap_bh);
395 for (i = 0, group_freed = 0; i < count; i++) {
397 * An HJ special. This is expensive...
399 #ifdef CONFIG_JBD_DEBUG
400 jbd_unlock_bh_state(bitmap_bh);
402 struct buffer_head *debug_bh;
403 debug_bh = sb_find_get_block(sb, block + i);
405 BUFFER_TRACE(debug_bh, "Deleted!");
406 if (!bh2jh(bitmap_bh)->b_committed_data)
407 BUFFER_TRACE(debug_bh,
408 "No commited data in bitmap");
409 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
413 jbd_lock_bh_state(bitmap_bh);
415 if (need_resched()) {
416 jbd_unlock_bh_state(bitmap_bh);
418 jbd_lock_bh_state(bitmap_bh);
420 /* @@@ This prevents newly-allocated data from being
421 * freed and then reallocated within the same
424 * Ideally we would want to allow that to happen, but to
425 * do so requires making journal_forget() capable of
426 * revoking the queued write of a data block, which
427 * implies blocking on the journal lock. *forget()
428 * cannot block due to truncate races.
430 * Eventually we can fix this by making journal_forget()
431 * return a status indicating whether or not it was able
432 * to revoke the buffer. On successful revoke, it is
433 * safe not to set the allocation bit in the committed
434 * bitmap, because we know that there is no outstanding
435 * activity on the buffer any more and so it is safe to
438 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
439 J_ASSERT_BH(bitmap_bh,
440 bh2jh(bitmap_bh)->b_committed_data != NULL);
441 ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
442 bh2jh(bitmap_bh)->b_committed_data);
445 * We clear the bit in the bitmap after setting the committed
446 * data bit, because this is the reverse order to that which
447 * the allocator uses.
449 BUFFER_TRACE(bitmap_bh, "clear bit");
450 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
451 bit + i, bitmap_bh->b_data)) {
452 jbd_unlock_bh_state(bitmap_bh);
453 ext3_error(sb, __FUNCTION__,
454 "bit already cleared for block "E3FSBLK,
456 jbd_lock_bh_state(bitmap_bh);
457 BUFFER_TRACE(bitmap_bh, "bit already cleared");
462 jbd_unlock_bh_state(bitmap_bh);
464 spin_lock(sb_bgl_lock(sbi, block_group));
465 desc->bg_free_blocks_count =
466 cpu_to_le16(le16_to_cpu(desc->bg_free_blocks_count) +
468 spin_unlock(sb_bgl_lock(sbi, block_group));
469 percpu_counter_mod(&sbi->s_freeblocks_counter, count);
471 /* We dirtied the bitmap block */
472 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
473 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
475 /* And the group descriptor block */
476 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
477 ret = ext3_journal_dirty_metadata(handle, gd_bh);
479 *pdquot_freed_blocks += group_freed;
481 if (overflow && !err) {
489 ext3_std_error(sb, err);
493 /* Free given blocks, update quota and i_blocks field */
494 void ext3_free_blocks(handle_t *handle, struct inode *inode,
495 ext3_fsblk_t block, unsigned long count)
497 struct super_block * sb;
498 unsigned long dquot_freed_blocks;
502 printk ("ext3_free_blocks: nonexistent device");
505 ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
506 if (dquot_freed_blocks)
507 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
512 * For ext3 allocations, we must not reuse any blocks which are
513 * allocated in the bitmap buffer's "last committed data" copy. This
514 * prevents deletes from freeing up the page for reuse until we have
515 * committed the delete transaction.
517 * If we didn't do this, then deleting something and reallocating it as
518 * data would allow the old block to be overwritten before the
519 * transaction committed (because we force data to disk before commit).
520 * This would lead to corruption if we crashed between overwriting the
521 * data and committing the delete.
523 * @@@ We may want to make this allocation behaviour conditional on
524 * data-writes at some point, and disable it for metadata allocations or
527 static int ext3_test_allocatable(ext3_grpblk_t nr, struct buffer_head *bh)
530 struct journal_head *jh = bh2jh(bh);
532 if (ext3_test_bit(nr, bh->b_data))
535 jbd_lock_bh_state(bh);
536 if (!jh->b_committed_data)
539 ret = !ext3_test_bit(nr, jh->b_committed_data);
540 jbd_unlock_bh_state(bh);
545 bitmap_search_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
546 ext3_grpblk_t maxblocks)
549 struct journal_head *jh = bh2jh(bh);
552 * The bitmap search --- search forward alternately through the actual
553 * bitmap and the last-committed copy until we find a bit free in
556 while (start < maxblocks) {
557 next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
558 if (next >= maxblocks)
560 if (ext3_test_allocatable(next, bh))
562 jbd_lock_bh_state(bh);
563 if (jh->b_committed_data)
564 start = ext3_find_next_zero_bit(jh->b_committed_data,
566 jbd_unlock_bh_state(bh);
572 * Find an allocatable block in a bitmap. We honour both the bitmap and
573 * its last-committed copy (if that exists), and perform the "most
574 * appropriate allocation" algorithm of looking for a free block near
575 * the initial goal; then for a free byte somewhere in the bitmap; then
576 * for any free bit in the bitmap.
579 find_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
580 ext3_grpblk_t maxblocks)
582 ext3_grpblk_t here, next;
587 * The goal was occupied; search forward for a free
588 * block within the next XX blocks.
590 * end_goal is more or less random, but it has to be
591 * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
592 * next 64-bit boundary is simple..
594 ext3_grpblk_t end_goal = (start + 63) & ~63;
595 if (end_goal > maxblocks)
596 end_goal = maxblocks;
597 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
598 if (here < end_goal && ext3_test_allocatable(here, bh))
600 ext3_debug("Bit not found near goal\n");
607 p = ((char *)bh->b_data) + (here >> 3);
608 r = memscan(p, 0, (maxblocks - here + 7) >> 3);
609 next = (r - ((char *)bh->b_data)) << 3;
611 if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
615 * The bitmap search --- search forward alternately through the actual
616 * bitmap and the last-committed copy until we find a bit free in
619 here = bitmap_search_next_usable_block(here, bh, maxblocks);
624 * We think we can allocate this block in this bitmap. Try to set the bit.
625 * If that succeeds then check that nobody has allocated and then freed the
626 * block since we saw that is was not marked in b_committed_data. If it _was_
627 * allocated and freed then clear the bit in the bitmap again and return
631 claim_block(spinlock_t *lock, ext3_grpblk_t block, struct buffer_head *bh)
633 struct journal_head *jh = bh2jh(bh);
636 if (ext3_set_bit_atomic(lock, block, bh->b_data))
638 jbd_lock_bh_state(bh);
639 if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
640 ext3_clear_bit_atomic(lock, block, bh->b_data);
645 jbd_unlock_bh_state(bh);
650 * If we failed to allocate the desired block then we may end up crossing to a
651 * new bitmap. In that case we must release write access to the old one via
652 * ext3_journal_release_buffer(), else we'll run out of credits.
655 ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
656 struct buffer_head *bitmap_bh, ext3_grpblk_t grp_goal,
657 unsigned long *count, struct ext3_reserve_window *my_rsv)
659 ext3_fsblk_t group_first_block;
660 ext3_grpblk_t start, end;
661 unsigned long num = 0;
663 /* we do allocation within the reservation window if we have a window */
665 group_first_block = ext3_group_first_block_no(sb, group);
666 if (my_rsv->_rsv_start >= group_first_block)
667 start = my_rsv->_rsv_start - group_first_block;
669 /* reservation window cross group boundary */
671 end = my_rsv->_rsv_end - group_first_block + 1;
672 if (end > EXT3_BLOCKS_PER_GROUP(sb))
673 /* reservation window crosses group boundary */
674 end = EXT3_BLOCKS_PER_GROUP(sb);
675 if ((start <= grp_goal) && (grp_goal < end))
684 end = EXT3_BLOCKS_PER_GROUP(sb);
687 BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
690 if (grp_goal < 0 || !ext3_test_allocatable(grp_goal, bitmap_bh)) {
691 grp_goal = find_next_usable_block(start, bitmap_bh, end);
697 for (i = 0; i < 7 && grp_goal > start &&
698 ext3_test_allocatable(grp_goal - 1,
706 if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group), grp_goal, bitmap_bh)) {
708 * The block was allocated by another thread, or it was
709 * allocated and then freed by another thread
719 while (num < *count && grp_goal < end
720 && ext3_test_allocatable(grp_goal, bitmap_bh)
721 && claim_block(sb_bgl_lock(EXT3_SB(sb), group), grp_goal, bitmap_bh)) {
726 return grp_goal - num;
733 * find_next_reservable_window():
734 * find a reservable space within the given range.
735 * It does not allocate the reservation window for now:
736 * alloc_new_reservation() will do the work later.
738 * @search_head: the head of the searching list;
739 * This is not necessarily the list head of the whole filesystem
741 * We have both head and start_block to assist the search
742 * for the reservable space. The list starts from head,
743 * but we will shift to the place where start_block is,
744 * then start from there, when looking for a reservable space.
746 * @size: the target new reservation window size
748 * @group_first_block: the first block we consider to start
749 * the real search from
752 * the maximum block number that our goal reservable space
753 * could start from. This is normally the last block in this
754 * group. The search will end when we found the start of next
755 * possible reservable space is out of this boundary.
756 * This could handle the cross boundary reservation window
759 * basically we search from the given range, rather than the whole
760 * reservation double linked list, (start_block, last_block)
761 * to find a free region that is of my size and has not
765 static int find_next_reservable_window(
766 struct ext3_reserve_window_node *search_head,
767 struct ext3_reserve_window_node *my_rsv,
768 struct super_block * sb,
769 ext3_fsblk_t start_block,
770 ext3_fsblk_t last_block)
772 struct rb_node *next;
773 struct ext3_reserve_window_node *rsv, *prev;
775 int size = my_rsv->rsv_goal_size;
777 /* TODO: make the start of the reservation window byte-aligned */
778 /* cur = *start_block & ~7;*/
785 if (cur <= rsv->rsv_end)
786 cur = rsv->rsv_end + 1;
789 * in the case we could not find a reservable space
790 * that is what is expected, during the re-search, we could
791 * remember what's the largest reservable space we could have
792 * and return that one.
794 * For now it will fail if we could not find the reservable
795 * space with expected-size (or more)...
797 if (cur > last_block)
798 return -1; /* fail */
801 next = rb_next(&rsv->rsv_node);
802 rsv = list_entry(next,struct ext3_reserve_window_node,rsv_node);
805 * Reached the last reservation, we can just append to the
811 if (cur + size <= rsv->rsv_start) {
813 * Found a reserveable space big enough. We could
814 * have a reservation across the group boundary here
820 * we come here either :
821 * when we reach the end of the whole list,
822 * and there is empty reservable space after last entry in the list.
823 * append it to the end of the list.
825 * or we found one reservable space in the middle of the list,
826 * return the reservation window that we could append to.
830 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
831 rsv_window_remove(sb, my_rsv);
834 * Let's book the whole avaliable window for now. We will check the
835 * disk bitmap later and then, if there are free blocks then we adjust
836 * the window size if it's larger than requested.
837 * Otherwise, we will remove this node from the tree next time
838 * call find_next_reservable_window.
840 my_rsv->rsv_start = cur;
841 my_rsv->rsv_end = cur + size - 1;
842 my_rsv->rsv_alloc_hit = 0;
845 ext3_rsv_window_add(sb, my_rsv);
851 * alloc_new_reservation()--allocate a new reservation window
853 * To make a new reservation, we search part of the filesystem
854 * reservation list (the list that inside the group). We try to
855 * allocate a new reservation window near the allocation goal,
856 * or the beginning of the group, if there is no goal.
858 * We first find a reservable space after the goal, then from
859 * there, we check the bitmap for the first free block after
860 * it. If there is no free block until the end of group, then the
861 * whole group is full, we failed. Otherwise, check if the free
862 * block is inside the expected reservable space, if so, we
864 * If the first free block is outside the reservable space, then
865 * start from the first free block, we search for next available
868 * on succeed, a new reservation will be found and inserted into the list
869 * It contains at least one free block, and it does not overlap with other
870 * reservation windows.
872 * failed: we failed to find a reservation window in this group
874 * @rsv: the reservation
876 * @grp_goal: The goal (group-relative). It is where the search for a
877 * free reservable space should start from.
878 * if we have a grp_goal(grp_goal >0 ), then start from there,
879 * no grp_goal(grp_goal = -1), we start from the first block
882 * @sb: the super block
883 * @group: the group we are trying to allocate in
884 * @bitmap_bh: the block group block bitmap
887 static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
888 ext3_grpblk_t grp_goal, struct super_block *sb,
889 unsigned int group, struct buffer_head *bitmap_bh)
891 struct ext3_reserve_window_node *search_head;
892 ext3_fsblk_t group_first_block, group_end_block, start_block;
893 ext3_grpblk_t first_free_block;
894 struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
897 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
899 group_first_block = ext3_group_first_block_no(sb, group);
900 group_end_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1;
903 start_block = group_first_block;
905 start_block = grp_goal + group_first_block;
907 size = my_rsv->rsv_goal_size;
909 if (!rsv_is_empty(&my_rsv->rsv_window)) {
911 * if the old reservation is cross group boundary
912 * and if the goal is inside the old reservation window,
913 * we will come here when we just failed to allocate from
914 * the first part of the window. We still have another part
915 * that belongs to the next group. In this case, there is no
916 * point to discard our window and try to allocate a new one
917 * in this group(which will fail). we should
918 * keep the reservation window, just simply move on.
920 * Maybe we could shift the start block of the reservation
921 * window to the first block of next group.
924 if ((my_rsv->rsv_start <= group_end_block) &&
925 (my_rsv->rsv_end > group_end_block) &&
926 (start_block >= my_rsv->rsv_start))
929 if ((my_rsv->rsv_alloc_hit >
930 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
932 * if we previously allocation hit ration is greater than half
933 * we double the size of reservation window next time
934 * otherwise keep the same
937 if (size > EXT3_MAX_RESERVE_BLOCKS)
938 size = EXT3_MAX_RESERVE_BLOCKS;
939 my_rsv->rsv_goal_size= size;
945 * shift the search start to the window near the goal block
947 search_head = search_reserve_window(fs_rsv_root, start_block);
950 * find_next_reservable_window() simply finds a reservable window
951 * inside the given range(start_block, group_end_block).
953 * To make sure the reservation window has a free bit inside it, we
954 * need to check the bitmap after we found a reservable window.
957 ret = find_next_reservable_window(search_head, my_rsv, sb,
958 start_block, group_end_block);
961 if (!rsv_is_empty(&my_rsv->rsv_window))
962 rsv_window_remove(sb, my_rsv);
963 spin_unlock(rsv_lock);
968 * On success, find_next_reservable_window() returns the
969 * reservation window where there is a reservable space after it.
970 * Before we reserve this reservable space, we need
971 * to make sure there is at least a free block inside this region.
973 * searching the first free bit on the block bitmap and copy of
974 * last committed bitmap alternatively, until we found a allocatable
975 * block. Search start from the start block of the reservable space
978 spin_unlock(rsv_lock);
979 first_free_block = bitmap_search_next_usable_block(
980 my_rsv->rsv_start - group_first_block,
981 bitmap_bh, group_end_block - group_first_block + 1);
983 if (first_free_block < 0) {
985 * no free block left on the bitmap, no point
986 * to reserve the space. return failed.
989 if (!rsv_is_empty(&my_rsv->rsv_window))
990 rsv_window_remove(sb, my_rsv);
991 spin_unlock(rsv_lock);
992 return -1; /* failed */
995 start_block = first_free_block + group_first_block;
997 * check if the first free block is within the
998 * free space we just reserved
1000 if (start_block >= my_rsv->rsv_start && start_block < my_rsv->rsv_end)
1001 return 0; /* success */
1003 * if the first free bit we found is out of the reservable space
1004 * continue search for next reservable space,
1005 * start from where the free block is,
1006 * we also shift the list head to where we stopped last time
1008 search_head = my_rsv;
1009 spin_lock(rsv_lock);
1013 static void try_to_extend_reservation(struct ext3_reserve_window_node *my_rsv,
1014 struct super_block *sb, int size)
1016 struct ext3_reserve_window_node *next_rsv;
1017 struct rb_node *next;
1018 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1020 if (!spin_trylock(rsv_lock))
1023 next = rb_next(&my_rsv->rsv_node);
1026 my_rsv->rsv_end += size;
1028 next_rsv = list_entry(next, struct ext3_reserve_window_node, rsv_node);
1030 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1031 my_rsv->rsv_end += size;
1033 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1035 spin_unlock(rsv_lock);
1039 * This is the main function used to allocate a new block and its reservation
1042 * Each time when a new block allocation is need, first try to allocate from
1043 * its own reservation. If it does not have a reservation window, instead of
1044 * looking for a free bit on bitmap first, then look up the reservation list to
1045 * see if it is inside somebody else's reservation window, we try to allocate a
1046 * reservation window for it starting from the goal first. Then do the block
1047 * allocation within the reservation window.
1049 * This will avoid keeping on searching the reservation list again and
1050 * again when somebody is looking for a free block (without
1051 * reservation), and there are lots of free blocks, but they are all
1054 * We use a sorted double linked list for the per-filesystem reservation list.
1055 * The insert, remove and find a free space(non-reserved) operations for the
1056 * sorted double linked list should be fast.
1059 static ext3_grpblk_t
1060 ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1061 unsigned int group, struct buffer_head *bitmap_bh,
1062 ext3_grpblk_t grp_goal,
1063 struct ext3_reserve_window_node * my_rsv,
1064 unsigned long *count, int *errp)
1066 ext3_fsblk_t group_first_block;
1067 ext3_grpblk_t ret = 0;
1069 unsigned long num = *count;
1074 * Make sure we use undo access for the bitmap, because it is critical
1075 * that we do the frozen_data COW on bitmap buffers in all cases even
1076 * if the buffer is in BJ_Forget state in the committing transaction.
1078 BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1079 fatal = ext3_journal_get_undo_access(handle, bitmap_bh);
1086 * we don't deal with reservation when
1087 * filesystem is mounted without reservation
1088 * or the file is not a regular file
1089 * or last attempt to allocate a block with reservation turned on failed
1091 if (my_rsv == NULL ) {
1092 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1093 grp_goal, count, NULL);
1097 * grp_goal is a group relative block number (if there is a goal)
1098 * 0 < grp_goal < EXT3_BLOCKS_PER_GROUP(sb)
1099 * first block is a filesystem wide block number
1100 * first block is the block number of the first block in this group
1102 group_first_block = ext3_group_first_block_no(sb, group);
1105 * Basically we will allocate a new block from inode's reservation
1108 * We need to allocate a new reservation window, if:
1109 * a) inode does not have a reservation window; or
1110 * b) last attempt to allocate a block from existing reservation
1112 * c) we come here with a goal and with a reservation window
1114 * We do not need to allocate a new reservation window if we come here
1115 * at the beginning with a goal and the goal is inside the window, or
1116 * we don't have a goal but already have a reservation window.
1117 * then we could go to allocate from the reservation window directly.
1120 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1121 !goal_in_my_reservation(&my_rsv->rsv_window, grp_goal, group, sb)) {
1122 if (my_rsv->rsv_goal_size < *count)
1123 my_rsv->rsv_goal_size = *count;
1124 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1129 if (!goal_in_my_reservation(&my_rsv->rsv_window, grp_goal, group, sb))
1131 } else if (grp_goal > 0 && (my_rsv->rsv_end-grp_goal+1) < *count)
1132 try_to_extend_reservation(my_rsv, sb,
1133 *count-my_rsv->rsv_end + grp_goal - 1);
1135 if ((my_rsv->rsv_start >= group_first_block + EXT3_BLOCKS_PER_GROUP(sb))
1136 || (my_rsv->rsv_end < group_first_block))
1138 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh, grp_goal,
1139 &num, &my_rsv->rsv_window);
1141 my_rsv->rsv_alloc_hit += num;
1143 break; /* succeed */
1149 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1151 fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1159 BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1160 ext3_journal_release_buffer(handle, bitmap_bh);
1164 static int ext3_has_free_blocks(struct ext3_sb_info *sbi)
1166 ext3_fsblk_t free_blocks, root_blocks;
1168 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1169 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1170 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1171 sbi->s_resuid != current->fsuid &&
1172 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1179 * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1180 * it is profitable to retry the operation, this function will wait
1181 * for the current or commiting transaction to complete, and then
1184 int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1186 if (!ext3_has_free_blocks(EXT3_SB(sb)) || (*retries)++ > 3)
1189 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1191 return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1195 * ext3_new_block uses a goal block to assist allocation. If the goal is
1196 * free, or there is a free block within 32 blocks of the goal, that block
1197 * is allocated. Otherwise a forward search is made for a free block; within
1198 * each block group the search first looks for an entire free byte in the block
1199 * bitmap, and then for any free bit if that fails.
1200 * This function also updates quota and i_blocks field.
1202 ext3_fsblk_t ext3_new_blocks(handle_t *handle, struct inode *inode,
1203 ext3_fsblk_t goal, unsigned long *count, int *errp)
1205 struct buffer_head *bitmap_bh = NULL;
1206 struct buffer_head *gdp_bh;
1209 ext3_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1210 ext3_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1211 ext3_fsblk_t ret_block; /* filesyetem-wide allocated block */
1212 int bgi; /* blockgroup iteration index */
1214 int performed_allocation = 0;
1215 ext3_grpblk_t free_blocks; /* number of free blocks in a group */
1216 struct super_block *sb;
1217 struct ext3_group_desc *gdp;
1218 struct ext3_super_block *es;
1219 struct ext3_sb_info *sbi;
1220 struct ext3_reserve_window_node *my_rsv = NULL;
1221 struct ext3_block_alloc_info *block_i;
1222 unsigned short windowsz = 0;
1224 static int goal_hits, goal_attempts;
1226 unsigned long ngroups;
1227 unsigned long num = *count;
1232 printk("ext3_new_block: nonexistent device");
1237 * Check quota for allocation of this block.
1239 if (DQUOT_ALLOC_BLOCK(inode, num)) {
1245 es = EXT3_SB(sb)->s_es;
1246 ext3_debug("goal=%lu.\n", goal);
1248 * Allocate a block from reservation only when
1249 * filesystem is mounted with reservation(default,-o reservation), and
1250 * it's a regular file, and
1251 * the desired window size is greater than 0 (One could use ioctl
1252 * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1253 * reservation on that particular file)
1255 block_i = EXT3_I(inode)->i_block_alloc_info;
1256 if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1257 my_rsv = &block_i->rsv_window_node;
1259 if (!ext3_has_free_blocks(sbi)) {
1265 * First, test whether the goal block is free.
1267 if (goal < le32_to_cpu(es->s_first_data_block) ||
1268 goal >= le32_to_cpu(es->s_blocks_count))
1269 goal = le32_to_cpu(es->s_first_data_block);
1270 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1271 EXT3_BLOCKS_PER_GROUP(sb);
1272 goal_group = group_no;
1274 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1278 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1280 * if there is not enough free blocks to make a new resevation
1281 * turn off reservation for this allocation
1283 if (my_rsv && (free_blocks < windowsz)
1284 && (rsv_is_empty(&my_rsv->rsv_window)))
1287 if (free_blocks > 0) {
1288 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1289 EXT3_BLOCKS_PER_GROUP(sb));
1290 bitmap_bh = read_block_bitmap(sb, group_no);
1293 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1294 group_no, bitmap_bh, grp_target_blk,
1295 my_rsv, &num, &fatal);
1298 if (grp_alloc_blk >= 0)
1302 ngroups = EXT3_SB(sb)->s_groups_count;
1306 * Now search the rest of the groups. We assume that
1307 * i and gdp correctly point to the last group visited.
1309 for (bgi = 0; bgi < ngroups; bgi++) {
1311 if (group_no >= ngroups)
1313 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1318 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1320 * skip this group if the number of
1321 * free blocks is less than half of the reservation
1324 if (free_blocks <= (windowsz/2))
1328 bitmap_bh = read_block_bitmap(sb, group_no);
1332 * try to allocate block(s) from this group, without a goal(-1).
1334 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1335 group_no, bitmap_bh, -1, my_rsv,
1339 if (grp_alloc_blk >= 0)
1343 * We may end up a bogus ealier ENOSPC error due to
1344 * filesystem is "full" of reservations, but
1345 * there maybe indeed free blocks avaliable on disk
1346 * In this case, we just forget about the reservations
1347 * just do block allocation as without reservations.
1351 group_no = goal_group;
1354 /* No space left on the device */
1360 ext3_debug("using block group %d(%d)\n",
1361 group_no, gdp->bg_free_blocks_count);
1363 BUFFER_TRACE(gdp_bh, "get_write_access");
1364 fatal = ext3_journal_get_write_access(handle, gdp_bh);
1368 ret_block = grp_alloc_blk + ext3_group_first_block_no(sb, group_no);
1370 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1371 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1372 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1373 EXT3_SB(sb)->s_itb_per_group) ||
1374 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1375 EXT3_SB(sb)->s_itb_per_group))
1376 ext3_error(sb, "ext3_new_block",
1377 "Allocating block in system zone - "
1378 "blocks from "E3FSBLK", length %lu",
1381 performed_allocation = 1;
1383 #ifdef CONFIG_JBD_DEBUG
1385 struct buffer_head *debug_bh;
1387 /* Record bitmap buffer state in the newly allocated block */
1388 debug_bh = sb_find_get_block(sb, ret_block);
1390 BUFFER_TRACE(debug_bh, "state when allocated");
1391 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1395 jbd_lock_bh_state(bitmap_bh);
1396 spin_lock(sb_bgl_lock(sbi, group_no));
1397 if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1400 for (i = 0; i < num; i++) {
1401 if (ext3_test_bit(grp_alloc_blk+i,
1402 bh2jh(bitmap_bh)->b_committed_data)) {
1403 printk("%s: block was unexpectedly set in "
1404 "b_committed_data\n", __FUNCTION__);
1408 ext3_debug("found bit %d\n", grp_alloc_blk);
1409 spin_unlock(sb_bgl_lock(sbi, group_no));
1410 jbd_unlock_bh_state(bitmap_bh);
1413 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1414 ext3_error(sb, "ext3_new_block",
1415 "block("E3FSBLK") >= blocks count(%d) - "
1416 "block_group = %d, es == %p ", ret_block,
1417 le32_to_cpu(es->s_blocks_count), group_no, es);
1422 * It is up to the caller to add the new buffer to a journal
1423 * list of some description. We don't know in advance whether
1424 * the caller wants to use it as metadata or data.
1426 ext3_debug("allocating block %lu. Goal hits %d of %d.\n",
1427 ret_block, goal_hits, goal_attempts);
1429 spin_lock(sb_bgl_lock(sbi, group_no));
1430 gdp->bg_free_blocks_count =
1431 cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count) - num);
1432 spin_unlock(sb_bgl_lock(sbi, group_no));
1433 percpu_counter_mod(&sbi->s_freeblocks_counter, -num);
1435 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1436 err = ext3_journal_dirty_metadata(handle, gdp_bh);
1446 DQUOT_FREE_BLOCK(inode, *count-num);
1455 ext3_std_error(sb, fatal);
1458 * Undo the block allocation
1460 if (!performed_allocation)
1461 DQUOT_FREE_BLOCK(inode, *count);
1466 ext3_fsblk_t ext3_new_block(handle_t *handle, struct inode *inode,
1467 ext3_fsblk_t goal, int *errp)
1469 unsigned long count = 1;
1471 return ext3_new_blocks(handle, inode, goal, &count, errp);
1474 ext3_fsblk_t ext3_count_free_blocks(struct super_block *sb)
1476 ext3_fsblk_t desc_count;
1477 struct ext3_group_desc *gdp;
1479 unsigned long ngroups = EXT3_SB(sb)->s_groups_count;
1481 struct ext3_super_block *es;
1482 ext3_fsblk_t bitmap_count;
1484 struct buffer_head *bitmap_bh = NULL;
1486 es = EXT3_SB(sb)->s_es;
1492 for (i = 0; i < ngroups; i++) {
1493 gdp = ext3_get_group_desc(sb, i, NULL);
1496 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1498 bitmap_bh = read_block_bitmap(sb, i);
1499 if (bitmap_bh == NULL)
1502 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1503 printk("group %d: stored = %d, counted = %lu\n",
1504 i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1508 printk("ext3_count_free_blocks: stored = "E3FSBLK
1509 ", computed = "E3FSBLK", "E3FSBLK"\n",
1510 le32_to_cpu(es->s_free_blocks_count),
1511 desc_count, bitmap_count);
1512 return bitmap_count;
1516 for (i = 0; i < ngroups; i++) {
1517 gdp = ext3_get_group_desc(sb, i, NULL);
1520 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1528 block_in_use(ext3_fsblk_t block, struct super_block *sb, unsigned char *map)
1530 return ext3_test_bit ((block -
1531 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block)) %
1532 EXT3_BLOCKS_PER_GROUP(sb), map);
1535 static inline int test_root(int a, int b)
1544 static int ext3_group_sparse(int group)
1550 return (test_root(group, 7) || test_root(group, 5) ||
1551 test_root(group, 3));
1555 * ext3_bg_has_super - number of blocks used by the superblock in group
1556 * @sb: superblock for filesystem
1557 * @group: group number to check
1559 * Return the number of blocks used by the superblock (primary or backup)
1560 * in this group. Currently this will be only 0 or 1.
1562 int ext3_bg_has_super(struct super_block *sb, int group)
1564 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1565 EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1566 !ext3_group_sparse(group))
1571 static unsigned long ext3_bg_num_gdb_meta(struct super_block *sb, int group)
1573 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1574 unsigned long first = metagroup * EXT3_DESC_PER_BLOCK(sb);
1575 unsigned long last = first + EXT3_DESC_PER_BLOCK(sb) - 1;
1577 if (group == first || group == first + 1 || group == last)
1582 static unsigned long ext3_bg_num_gdb_nometa(struct super_block *sb, int group)
1584 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1585 EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1586 !ext3_group_sparse(group))
1588 return EXT3_SB(sb)->s_gdb_count;
1592 * ext3_bg_num_gdb - number of blocks used by the group table in group
1593 * @sb: superblock for filesystem
1594 * @group: group number to check
1596 * Return the number of blocks used by the group descriptor table
1597 * (primary or backup) in this group. In the future there may be a
1598 * different number of descriptor blocks in each group.
1600 unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1602 unsigned long first_meta_bg =
1603 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_meta_bg);
1604 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1606 if (!EXT3_HAS_INCOMPAT_FEATURE(sb,EXT3_FEATURE_INCOMPAT_META_BG) ||
1607 metagroup < first_meta_bg)
1608 return ext3_bg_num_gdb_nometa(sb,group);
1610 return ext3_bg_num_gdb_meta(sb,group);