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/config.h>
15 #include <linux/time.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>
22 #include <linux/vs_dlimit.h>
25 * balloc.c contains the blocks allocation and deallocation routines
29 * The free blocks are managed by bitmaps. A file system contains several
30 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
31 * block for inodes, N blocks for the inode table and data blocks.
33 * The file system contains group descriptors which are located after the
34 * super block. Each descriptor contains the number of the bitmap block and
35 * the free blocks count in the block. The descriptors are loaded in memory
36 * when a file system is mounted (see ext3_read_super).
40 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
42 struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb,
43 unsigned int block_group,
44 struct buffer_head ** bh)
46 unsigned long group_desc;
48 struct ext3_group_desc * gdp;
50 if (block_group >= EXT3_SB(sb)->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, EXT3_SB(sb)->s_groups_count);
60 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
61 desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
62 if (!EXT3_SB(sb)->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, desc);
70 gdp = (struct ext3_group_desc *)
71 EXT3_SB(sb)->s_group_desc[group_desc]->b_data;
73 *bh = EXT3_SB(sb)->s_group_desc[group_desc];
78 * Read the bitmap for a given block_group, reading into the specified
79 * slot in the superblock's bitmap cache.
81 * Return buffer_head on success or NULL in case of failure.
83 static struct buffer_head *
84 read_block_bitmap(struct super_block *sb, unsigned int block_group)
86 struct ext3_group_desc * desc;
87 struct buffer_head * bh = NULL;
89 desc = ext3_get_group_desc (sb, block_group, NULL);
92 bh = sb_bread(sb, le32_to_cpu(desc->bg_block_bitmap));
94 ext3_error (sb, "read_block_bitmap",
95 "Cannot read block bitmap - "
96 "block_group = %d, block_bitmap = %u",
97 block_group, le32_to_cpu(desc->bg_block_bitmap));
102 * The reservation window structure operations
103 * --------------------------------------------
104 * Operations include:
105 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
107 * We use sorted double linked list for the per-filesystem reservation
108 * window list. (like in vm_region).
110 * Initially, we keep those small operations in the abstract functions,
111 * so later if we need a better searching tree than double linked-list,
112 * we could easily switch to that without changing too much
116 static void __rsv_window_dump(struct rb_root *root, int verbose,
120 struct ext3_reserve_window_node *rsv, *prev;
128 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
130 rsv = list_entry(n, struct ext3_reserve_window_node, rsv_node);
132 printk("reservation window 0x%p "
133 "start: %d, end: %d\n",
134 rsv, rsv->rsv_start, rsv->rsv_end);
135 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
136 printk("Bad reservation %p (start >= end)\n",
140 if (prev && prev->rsv_end >= rsv->rsv_start) {
141 printk("Bad reservation %p (prev->end >= start)\n",
147 printk("Restarting reservation walk in verbose mode\n");
155 printk("Window map complete.\n");
159 #define rsv_window_dump(root, verbose) \
160 __rsv_window_dump((root), (verbose), __FUNCTION__)
162 #define rsv_window_dump(root, verbose) do {} while (0)
166 goal_in_my_reservation(struct ext3_reserve_window *rsv, int goal,
167 unsigned int group, struct super_block * sb)
169 unsigned long group_first_block, group_last_block;
171 group_first_block = le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) +
172 group * EXT3_BLOCKS_PER_GROUP(sb);
173 group_last_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1;
175 if ((rsv->_rsv_start > group_last_block) ||
176 (rsv->_rsv_end < group_first_block))
178 if ((goal >= 0) && ((goal + group_first_block < rsv->_rsv_start)
179 || (goal + group_first_block > rsv->_rsv_end)))
185 * Find the reserved window which includes the goal, or the previous one
186 * if the goal is not in any window.
187 * Returns NULL if there are no windows or if all windows start after the goal.
189 static struct ext3_reserve_window_node *
190 search_reserve_window(struct rb_root *root, unsigned long goal)
192 struct rb_node *n = root->rb_node;
193 struct ext3_reserve_window_node *rsv;
199 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
201 if (goal < rsv->rsv_start)
203 else if (goal > rsv->rsv_end)
209 * We've fallen off the end of the tree: the goal wasn't inside
210 * any particular node. OK, the previous node must be to one
211 * side of the interval containing the goal. If it's the RHS,
212 * we need to back up one.
214 if (rsv->rsv_start > goal) {
215 n = rb_prev(&rsv->rsv_node);
216 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
221 void ext3_rsv_window_add(struct super_block *sb,
222 struct ext3_reserve_window_node *rsv)
224 struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
225 struct rb_node *node = &rsv->rsv_node;
226 unsigned int start = rsv->rsv_start;
228 struct rb_node ** p = &root->rb_node;
229 struct rb_node * parent = NULL;
230 struct ext3_reserve_window_node *this;
235 this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
237 if (start < this->rsv_start)
239 else if (start > this->rsv_end)
245 rb_link_node(node, parent, p);
246 rb_insert_color(node, root);
249 static void rsv_window_remove(struct super_block *sb,
250 struct ext3_reserve_window_node *rsv)
252 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
253 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
254 atomic_set(&rsv->rsv_alloc_hit, 0);
255 rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
258 static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
260 /* a valid reservation end block could not be 0 */
261 return (rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED);
264 void ext3_discard_reservation(struct inode *inode)
266 struct ext3_inode_info *ei = EXT3_I(inode);
267 struct ext3_reserve_window_node *rsv = &ei->i_rsv_window;
268 spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
270 if (!rsv_is_empty(&rsv->rsv_window)) {
272 rsv_window_remove(inode->i_sb, rsv);
273 spin_unlock(rsv_lock);
277 /* Free given blocks, update quota and i_blocks field */
278 void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
279 unsigned long block, unsigned long count,
280 int *pdquot_freed_blocks)
282 struct buffer_head *bitmap_bh = NULL;
283 struct buffer_head *gd_bh;
284 unsigned long block_group;
287 unsigned long overflow;
288 struct ext3_group_desc * gdp;
289 struct ext3_super_block * es;
290 struct ext3_sb_info *sbi;
293 *pdquot_freed_blocks = 0;
295 es = EXT3_SB(sb)->s_es;
296 if (block < le32_to_cpu(es->s_first_data_block) ||
297 block + count < block ||
298 block + count > le32_to_cpu(es->s_blocks_count)) {
299 ext3_error (sb, "ext3_free_blocks",
300 "Freeing blocks not in datazone - "
301 "block = %lu, count = %lu", block, count);
305 ext3_debug ("freeing block %lu\n", block);
309 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
310 EXT3_BLOCKS_PER_GROUP(sb);
311 bit = (block - le32_to_cpu(es->s_first_data_block)) %
312 EXT3_BLOCKS_PER_GROUP(sb);
314 * Check to see if we are freeing blocks across a group
317 if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
318 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
322 bitmap_bh = read_block_bitmap(sb, block_group);
325 gdp = ext3_get_group_desc (sb, block_group, &gd_bh);
329 if (in_range (le32_to_cpu(gdp->bg_block_bitmap), block, count) ||
330 in_range (le32_to_cpu(gdp->bg_inode_bitmap), block, count) ||
331 in_range (block, le32_to_cpu(gdp->bg_inode_table),
332 EXT3_SB(sb)->s_itb_per_group) ||
333 in_range (block + count - 1, le32_to_cpu(gdp->bg_inode_table),
334 EXT3_SB(sb)->s_itb_per_group))
335 ext3_error (sb, "ext3_free_blocks",
336 "Freeing blocks in system zones - "
337 "Block = %lu, count = %lu",
341 * We are about to start releasing blocks in the bitmap,
342 * so we need undo access.
344 /* @@@ check errors */
345 BUFFER_TRACE(bitmap_bh, "getting undo access");
346 err = ext3_journal_get_undo_access(handle, bitmap_bh, NULL);
351 * We are about to modify some metadata. Call the journal APIs
352 * to unshare ->b_data if a currently-committing transaction is
355 BUFFER_TRACE(gd_bh, "get_write_access");
356 err = ext3_journal_get_write_access(handle, gd_bh);
360 jbd_lock_bh_state(bitmap_bh);
362 for (i = 0; i < count; i++) {
364 * An HJ special. This is expensive...
366 #ifdef CONFIG_JBD_DEBUG
367 jbd_unlock_bh_state(bitmap_bh);
369 struct buffer_head *debug_bh;
370 debug_bh = sb_find_get_block(sb, block + i);
372 BUFFER_TRACE(debug_bh, "Deleted!");
373 if (!bh2jh(bitmap_bh)->b_committed_data)
374 BUFFER_TRACE(debug_bh,
375 "No commited data in bitmap");
376 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
380 jbd_lock_bh_state(bitmap_bh);
382 if (need_resched()) {
383 jbd_unlock_bh_state(bitmap_bh);
385 jbd_lock_bh_state(bitmap_bh);
387 /* @@@ This prevents newly-allocated data from being
388 * freed and then reallocated within the same
391 * Ideally we would want to allow that to happen, but to
392 * do so requires making journal_forget() capable of
393 * revoking the queued write of a data block, which
394 * implies blocking on the journal lock. *forget()
395 * cannot block due to truncate races.
397 * Eventually we can fix this by making journal_forget()
398 * return a status indicating whether or not it was able
399 * to revoke the buffer. On successful revoke, it is
400 * safe not to set the allocation bit in the committed
401 * bitmap, because we know that there is no outstanding
402 * activity on the buffer any more and so it is safe to
405 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
406 J_ASSERT_BH(bitmap_bh,
407 bh2jh(bitmap_bh)->b_committed_data != NULL);
408 ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
409 bh2jh(bitmap_bh)->b_committed_data);
412 * We clear the bit in the bitmap after setting the committed
413 * data bit, because this is the reverse order to that which
414 * the allocator uses.
416 BUFFER_TRACE(bitmap_bh, "clear bit");
417 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
418 bit + i, bitmap_bh->b_data)) {
419 jbd_unlock_bh_state(bitmap_bh);
420 ext3_error(sb, __FUNCTION__,
421 "bit already cleared for block %lu", block + i);
422 jbd_lock_bh_state(bitmap_bh);
423 BUFFER_TRACE(bitmap_bh, "bit already cleared");
425 (*pdquot_freed_blocks)++;
428 jbd_unlock_bh_state(bitmap_bh);
430 spin_lock(sb_bgl_lock(sbi, block_group));
431 gdp->bg_free_blocks_count =
432 cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count) +
433 *pdquot_freed_blocks);
434 spin_unlock(sb_bgl_lock(sbi, block_group));
435 percpu_counter_mod(&sbi->s_freeblocks_counter, count);
437 /* We dirtied the bitmap block */
438 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
439 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
441 /* And the group descriptor block */
442 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
443 ret = ext3_journal_dirty_metadata(handle, gd_bh);
446 if (overflow && !err) {
454 ext3_std_error(sb, err);
458 /* Free given blocks, update quota and i_blocks field */
459 void ext3_free_blocks(handle_t *handle, struct inode *inode,
460 unsigned long block, unsigned long count)
462 struct super_block * sb;
463 int dquot_freed_blocks;
467 printk ("ext3_free_blocks: nonexistent device");
470 ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
471 if (dquot_freed_blocks) {
472 DLIMIT_FREE_BLOCK(sb, inode->i_xid, dquot_freed_blocks);
473 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
479 * For ext3 allocations, we must not reuse any blocks which are
480 * allocated in the bitmap buffer's "last committed data" copy. This
481 * prevents deletes from freeing up the page for reuse until we have
482 * committed the delete transaction.
484 * If we didn't do this, then deleting something and reallocating it as
485 * data would allow the old block to be overwritten before the
486 * transaction committed (because we force data to disk before commit).
487 * This would lead to corruption if we crashed between overwriting the
488 * data and committing the delete.
490 * @@@ We may want to make this allocation behaviour conditional on
491 * data-writes at some point, and disable it for metadata allocations or
494 static int ext3_test_allocatable(int nr, struct buffer_head *bh)
497 struct journal_head *jh = bh2jh(bh);
499 if (ext3_test_bit(nr, bh->b_data))
502 jbd_lock_bh_state(bh);
503 if (!jh->b_committed_data)
506 ret = !ext3_test_bit(nr, jh->b_committed_data);
507 jbd_unlock_bh_state(bh);
512 bitmap_search_next_usable_block(int start, struct buffer_head *bh,
516 struct journal_head *jh = bh2jh(bh);
519 * The bitmap search --- search forward alternately through the actual
520 * bitmap and the last-committed copy until we find a bit free in
523 while (start < maxblocks) {
524 next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
525 if (next >= maxblocks)
527 if (ext3_test_allocatable(next, bh))
529 jbd_lock_bh_state(bh);
530 if (jh->b_committed_data)
531 start = ext3_find_next_zero_bit(jh->b_committed_data,
533 jbd_unlock_bh_state(bh);
539 * Find an allocatable block in a bitmap. We honour both the bitmap and
540 * its last-committed copy (if that exists), and perform the "most
541 * appropriate allocation" algorithm of looking for a free block near
542 * the initial goal; then for a free byte somewhere in the bitmap; then
543 * for any free bit in the bitmap.
546 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
553 * The goal was occupied; search forward for a free
554 * block within the next XX blocks.
556 * end_goal is more or less random, but it has to be
557 * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
558 * next 64-bit boundary is simple..
560 int end_goal = (start + 63) & ~63;
561 if (end_goal > maxblocks)
562 end_goal = maxblocks;
563 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
564 if (here < end_goal && ext3_test_allocatable(here, bh))
566 ext3_debug("Bit not found near goal\n");
573 p = ((char *)bh->b_data) + (here >> 3);
574 r = memscan(p, 0, (maxblocks - here + 7) >> 3);
575 next = (r - ((char *)bh->b_data)) << 3;
577 if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
581 * The bitmap search --- search forward alternately through the actual
582 * bitmap and the last-committed copy until we find a bit free in
585 here = bitmap_search_next_usable_block(here, bh, maxblocks);
590 * We think we can allocate this block in this bitmap. Try to set the bit.
591 * If that succeeds then check that nobody has allocated and then freed the
592 * block since we saw that is was not marked in b_committed_data. If it _was_
593 * allocated and freed then clear the bit in the bitmap again and return
597 claim_block(spinlock_t *lock, int block, struct buffer_head *bh)
599 struct journal_head *jh = bh2jh(bh);
602 if (ext3_set_bit_atomic(lock, block, bh->b_data))
604 jbd_lock_bh_state(bh);
605 if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
606 ext3_clear_bit_atomic(lock, block, bh->b_data);
611 jbd_unlock_bh_state(bh);
616 * If we failed to allocate the desired block then we may end up crossing to a
617 * new bitmap. In that case we must release write access to the old one via
618 * ext3_journal_release_buffer(), else we'll run out of credits.
621 ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
622 struct buffer_head *bitmap_bh, int goal, struct ext3_reserve_window *my_rsv)
624 int group_first_block, start, end;
626 /* we do allocation within the reservation window if we have a window */
629 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) +
630 group * EXT3_BLOCKS_PER_GROUP(sb);
631 if (my_rsv->_rsv_start >= group_first_block)
632 start = my_rsv->_rsv_start - group_first_block;
634 /* reservation window cross group boundary */
636 end = my_rsv->_rsv_end - group_first_block + 1;
637 if (end > EXT3_BLOCKS_PER_GROUP(sb))
638 /* reservation window crosses group boundary */
639 end = EXT3_BLOCKS_PER_GROUP(sb);
640 if ((start <= goal) && (goal < end))
649 end = EXT3_BLOCKS_PER_GROUP(sb);
652 BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
655 if (goal < 0 || !ext3_test_allocatable(goal, bitmap_bh)) {
656 goal = find_next_usable_block(start, bitmap_bh, end);
662 for (i = 0; i < 7 && goal > start &&
663 ext3_test_allocatable(goal - 1,
671 if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group), goal, bitmap_bh)) {
673 * The block was allocated by another thread, or it was
674 * allocated and then freed by another thread
688 * find_next_reservable_window():
689 * find a reservable space within the given range.
690 * It does not allocate the reservation window for now:
691 * alloc_new_reservation() will do the work later.
693 * @search_head: the head of the searching list;
694 * This is not necessarily the list head of the whole filesystem
696 * We have both head and start_block to assist the search
697 * for the reservable space. The list starts from head,
698 * but we will shift to the place where start_block is,
699 * then start from there, when looking for a reservable space.
701 * @size: the target new reservation window size
703 * @group_first_block: the first block we consider to start
704 * the real search from
707 * the maximum block number that our goal reservable space
708 * could start from. This is normally the last block in this
709 * group. The search will end when we found the start of next
710 * possible reservable space is out of this boundary.
711 * This could handle the cross boundary reservation window
714 * basically we search from the given range, rather than the whole
715 * reservation double linked list, (start_block, last_block)
716 * to find a free region that is of my size and has not
719 * on succeed, it returns the reservation window to be appended to.
720 * failed, return NULL.
722 static struct ext3_reserve_window_node *find_next_reservable_window(
723 struct ext3_reserve_window_node *search_head,
724 unsigned long size, int *start_block,
727 struct rb_node *next;
728 struct ext3_reserve_window_node *rsv, *prev;
731 /* TODO: make the start of the reservation window byte-aligned */
732 /* cur = *start_block & ~7;*/
739 if (cur <= rsv->rsv_end)
740 cur = rsv->rsv_end + 1;
743 * in the case we could not find a reservable space
744 * that is what is expected, during the re-search, we could
745 * remember what's the largest reservable space we could have
746 * and return that one.
748 * For now it will fail if we could not find the reservable
749 * space with expected-size (or more)...
751 if (cur > last_block)
752 return NULL; /* fail */
755 next = rb_next(&rsv->rsv_node);
756 rsv = list_entry(next, struct ext3_reserve_window_node, rsv_node);
759 * Reached the last reservation, we can just append to the
765 if (cur + size <= rsv->rsv_start) {
767 * Found a reserveable space big enough. We could
768 * have a reservation across the group boundary here
774 * we come here either :
775 * when we reach the end of the whole list,
776 * and there is empty reservable space after last entry in the list.
777 * append it to the end of the list.
779 * or we found one reservable space in the middle of the list,
780 * return the reservation window that we could append to.
788 * alloc_new_reservation()--allocate a new reservation window
790 * To make a new reservation, we search part of the filesystem
791 * reservation list (the list that inside the group). We try to
792 * allocate a new reservation window near the allocation goal,
793 * or the beginning of the group, if there is no goal.
795 * We first find a reservable space after the goal, then from
796 * there, we check the bitmap for the first free block after
797 * it. If there is no free block until the end of group, then the
798 * whole group is full, we failed. Otherwise, check if the free
799 * block is inside the expected reservable space, if so, we
801 * If the first free block is outside the reservable space, then
802 * start from the first free block, we search for next available
805 * on succeed, a new reservation will be found and inserted into the list
806 * It contains at least one free block, and it does not overlap with other
807 * reservation windows.
809 * failed: we failed to find a reservation window in this group
811 * @rsv: the reservation
813 * @goal: The goal (group-relative). It is where the search for a
814 * free reservable space should start from.
815 * if we have a goal(goal >0 ), then start from there,
816 * no goal(goal = -1), we start from the first block
819 * @sb: the super block
820 * @group: the group we are trying to allocate in
821 * @bitmap_bh: the block group block bitmap
823 static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
824 int goal, struct super_block *sb,
825 unsigned int group, struct buffer_head *bitmap_bh)
827 struct ext3_reserve_window_node *search_head;
828 int group_first_block, group_end_block, start_block;
829 int first_free_block;
830 int reservable_space_start;
831 struct ext3_reserve_window_node *prev_rsv;
832 struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
835 group_first_block = le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) +
836 group * EXT3_BLOCKS_PER_GROUP(sb);
837 group_end_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1;
840 start_block = group_first_block;
842 start_block = goal + group_first_block;
844 size = atomic_read(&my_rsv->rsv_goal_size);
845 if (!rsv_is_empty(&my_rsv->rsv_window)) {
847 * if the old reservation is cross group boundary
848 * and if the goal is inside the old reservation window,
849 * we will come here when we just failed to allocate from
850 * the first part of the window. We still have another part
851 * that belongs to the next group. In this case, there is no
852 * point to discard our window and try to allocate a new one
853 * in this group(which will fail). we should
854 * keep the reservation window, just simply move on.
856 * Maybe we could shift the start block of the reservation
857 * window to the first block of next group.
860 if ((my_rsv->rsv_start <= group_end_block) &&
861 (my_rsv->rsv_end > group_end_block) &&
862 (start_block >= my_rsv->rsv_start))
865 if ((atomic_read(&my_rsv->rsv_alloc_hit) >
866 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
868 * if we previously allocation hit ration is greater than half
869 * we double the size of reservation window next time
870 * otherwise keep the same
873 if (size > EXT3_MAX_RESERVE_BLOCKS)
874 size = EXT3_MAX_RESERVE_BLOCKS;
875 atomic_set(&my_rsv->rsv_goal_size, size);
879 * shift the search start to the window near the goal block
881 search_head = search_reserve_window(fs_rsv_root, start_block);
884 * find_next_reservable_window() simply finds a reservable window
885 * inside the given range(start_block, group_end_block).
887 * To make sure the reservation window has a free bit inside it, we
888 * need to check the bitmap after we found a reservable window.
891 prev_rsv = find_next_reservable_window(search_head, size,
892 &start_block, group_end_block);
893 if (prev_rsv == NULL)
895 reservable_space_start = start_block;
897 * On success, find_next_reservable_window() returns the
898 * reservation window where there is a reservable space after it.
899 * Before we reserve this reservable space, we need
900 * to make sure there is at least a free block inside this region.
902 * searching the first free bit on the block bitmap and copy of
903 * last committed bitmap alternatively, until we found a allocatable
904 * block. Search start from the start block of the reservable space
907 first_free_block = bitmap_search_next_usable_block(
908 reservable_space_start - group_first_block,
909 bitmap_bh, group_end_block - group_first_block + 1);
911 if (first_free_block < 0) {
913 * no free block left on the bitmap, no point
914 * to reserve the space. return failed.
918 start_block = first_free_block + group_first_block;
920 * check if the first free block is within the
921 * free space we just found
923 if ((start_block >= reservable_space_start) &&
924 (start_block < reservable_space_start + size))
925 goto found_rsv_window;
927 * if the first free bit we found is out of the reservable space
928 * this means there is no free block on the reservable space
929 * we should continue search for next reservable space,
930 * start from where the free block is,
931 * we also shift the list head to where we stopped last time
933 search_head = prev_rsv;
938 * great! the reservable space contains some free blocks.
939 * if the search returns that we should add the new
940 * window just next to where the old window, we don't
941 * need to remove the old window first then add it to the
942 * same place, just update the new start and new end.
944 if (my_rsv != prev_rsv) {
945 if (!rsv_is_empty(&my_rsv->rsv_window))
946 rsv_window_remove(sb, my_rsv);
948 my_rsv->rsv_start = reservable_space_start;
949 my_rsv->rsv_end = my_rsv->rsv_start + size - 1;
950 atomic_set(&my_rsv->rsv_alloc_hit, 0);
951 if (my_rsv != prev_rsv) {
952 ext3_rsv_window_add(sb, my_rsv);
954 return 0; /* succeed */
957 * failed to find a new reservation window in the current
958 * group, remove the current(stale) reservation window
961 if (!rsv_is_empty(&my_rsv->rsv_window))
962 rsv_window_remove(sb, my_rsv);
963 return -1; /* failed */
967 * This is the main function used to allocate a new block and its reservation
970 * Each time when a new block allocation is need, first try to allocate from
971 * its own reservation. If it does not have a reservation window, instead of
972 * looking for a free bit on bitmap first, then look up the reservation list to
973 * see if it is inside somebody else's reservation window, we try to allocate a
974 * reservation window for it starting from the goal first. Then do the block
975 * allocation within the reservation window.
977 * This will avoid keeping on searching the reservation list again and
978 * again when someboday is looking for a free block (without
979 * reservation), and there are lots of free blocks, but they are all
982 * We use a sorted double linked list for the per-filesystem reservation list.
983 * The insert, remove and find a free space(non-reserved) operations for the
984 * sorted double linked list should be fast.
988 ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
989 unsigned int group, struct buffer_head *bitmap_bh,
990 int goal, struct ext3_reserve_window_node * my_rsv,
993 spinlock_t *rsv_lock;
994 unsigned long group_first_block;
1002 * Make sure we use undo access for the bitmap, because it is critical
1003 * that we do the frozen_data COW on bitmap buffers in all cases even
1004 * if the buffer is in BJ_Forget state in the committing transaction.
1006 BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1007 fatal = ext3_journal_get_undo_access(handle, bitmap_bh, &credits);
1014 * we don't deal with reservation when
1015 * filesystem is mounted without reservation
1016 * or the file is not a regular file
1017 * or last attempt to allocate a block with reservation turned on failed
1019 if (my_rsv == NULL ) {
1020 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh, goal, NULL);
1023 rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1025 * goal is a group relative block number (if there is a goal)
1026 * 0 < goal < EXT3_BLOCKS_PER_GROUP(sb)
1027 * first block is a filesystem wide block number
1028 * first block is the block number of the first block in this group
1030 group_first_block = le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) +
1031 group * EXT3_BLOCKS_PER_GROUP(sb);
1034 * Basically we will allocate a new block from inode's reservation
1037 * We need to allocate a new reservation window, if:
1038 * a) inode does not have a reservation window; or
1039 * b) last attempt to allocate a block from existing reservation
1041 * c) we come here with a goal and with a reservation window
1043 * We do not need to allocate a new reservation window if we come here
1044 * at the beginning with a goal and the goal is inside the window, or
1045 * we don't have a goal but already have a reservation window.
1046 * then we could go to allocate from the reservation window directly.
1049 struct ext3_reserve_window rsv_copy;
1053 seq = read_seqbegin(&my_rsv->rsv_seqlock);
1054 rsv_copy._rsv_start = my_rsv->rsv_start;
1055 rsv_copy._rsv_end = my_rsv->rsv_end;
1056 } while (read_seqretry(&my_rsv->rsv_seqlock, seq));
1058 if (rsv_is_empty(&rsv_copy) || (ret < 0) ||
1059 !goal_in_my_reservation(&rsv_copy, goal, group, sb)) {
1060 spin_lock(rsv_lock);
1061 write_seqlock(&my_rsv->rsv_seqlock);
1062 ret = alloc_new_reservation(my_rsv, goal, sb,
1064 rsv_copy._rsv_start = my_rsv->rsv_start;
1065 rsv_copy._rsv_end = my_rsv->rsv_end;
1066 write_sequnlock(&my_rsv->rsv_seqlock);
1067 spin_unlock(rsv_lock);
1071 if (!goal_in_my_reservation(&rsv_copy, goal, group, sb))
1074 if ((rsv_copy._rsv_start >= group_first_block + EXT3_BLOCKS_PER_GROUP(sb))
1075 || (rsv_copy._rsv_end < group_first_block))
1077 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh, goal,
1080 if (!read_seqretry(&my_rsv->rsv_seqlock, seq))
1081 atomic_inc(&my_rsv->rsv_alloc_hit);
1082 break; /* succeed */
1087 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1089 fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1097 BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1098 ext3_journal_release_buffer(handle, bitmap_bh, credits);
1102 static int ext3_has_free_blocks(struct super_block *sb)
1104 struct ext3_sb_info *sbi = EXT3_SB(sb);
1105 int free_blocks, root_blocks, cond;
1107 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1108 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1110 vxdprintk(VXD_CBIT(dlim, 3),
1111 "ext3_has_free_blocks(%p): free=%u, root=%u",
1112 sb, free_blocks, root_blocks);
1114 DLIMIT_ADJUST_BLOCK(sb, vx_current_xid(), &free_blocks, &root_blocks);
1116 cond = (free_blocks < root_blocks + 1 &&
1117 !capable(CAP_SYS_RESOURCE) &&
1118 sbi->s_resuid != current->fsuid &&
1119 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid)));
1121 vxdprintk(VXD_CBIT(dlim, 3),
1122 "ext3_has_free_blocks(%p): %u<%u+1, %c, %u!=%u r=%d",
1123 sb, free_blocks, root_blocks,
1124 !capable(CAP_SYS_RESOURCE)?'1':'0',
1125 sbi->s_resuid, current->fsuid, cond?0:1);
1127 return (cond ? 0 : 1);
1131 * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1132 * it is profitable to retry the operation, this function will wait
1133 * for the current or commiting transaction to complete, and then
1136 int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1138 if (!ext3_has_free_blocks(sb) || (*retries)++ > 3)
1141 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1143 return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1147 * ext3_new_block uses a goal block to assist allocation. If the goal is
1148 * free, or there is a free block within 32 blocks of the goal, that block
1149 * is allocated. Otherwise a forward search is made for a free block; within
1150 * each block group the search first looks for an entire free byte in the block
1151 * bitmap, and then for any free bit if that fails.
1152 * This function also updates quota and i_blocks field.
1154 int ext3_new_block(handle_t *handle, struct inode *inode,
1155 unsigned long goal, int *errp)
1157 struct buffer_head *bitmap_bh = NULL;
1158 struct buffer_head *gdp_bh;
1162 int bgi; /* blockgroup iteration index */
1165 int performed_allocation = 0;
1167 struct super_block *sb;
1168 struct ext3_group_desc *gdp;
1169 struct ext3_super_block *es;
1170 struct ext3_sb_info *sbi;
1171 struct ext3_reserve_window_node *my_rsv = NULL;
1172 struct ext3_reserve_window_node *rsv = &EXT3_I(inode)->i_rsv_window;
1173 unsigned short windowsz = 0;
1175 static int goal_hits, goal_attempts;
1177 unsigned long ngroups;
1182 printk("ext3_new_block: nonexistent device");
1187 * Check quota for allocation of this block.
1189 if (DQUOT_ALLOC_BLOCK(inode, 1)) {
1193 if (DLIMIT_ALLOC_BLOCK(sb, inode->i_xid, 1))
1197 es = EXT3_SB(sb)->s_es;
1198 ext3_debug("goal=%lu.\n", goal);
1200 * Allocate a block from reservation only when
1201 * filesystem is mounted with reservation(default,-o reservation), and
1202 * it's a regular file, and
1203 * the desired window size is greater than 0 (One could use ioctl
1204 * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1205 * reservation on that particular file)
1207 windowsz = atomic_read(&rsv->rsv_goal_size);
1208 if (test_opt(sb, RESERVATION) &&
1209 S_ISREG(inode->i_mode) && (windowsz > 0))
1211 if (!ext3_has_free_blocks(sb)) {
1217 * First, test whether the goal block is free.
1219 if (goal < le32_to_cpu(es->s_first_data_block) ||
1220 goal >= le32_to_cpu(es->s_blocks_count))
1221 goal = le32_to_cpu(es->s_first_data_block);
1222 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1223 EXT3_BLOCKS_PER_GROUP(sb);
1224 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1228 goal_group = group_no;
1230 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1231 if (free_blocks > 0) {
1232 ret_block = ((goal - le32_to_cpu(es->s_first_data_block)) %
1233 EXT3_BLOCKS_PER_GROUP(sb));
1234 bitmap_bh = read_block_bitmap(sb, group_no);
1237 ret_block = ext3_try_to_allocate_with_rsv(sb, handle, group_no,
1238 bitmap_bh, ret_block, my_rsv, &fatal);
1245 ngroups = EXT3_SB(sb)->s_groups_count;
1249 * Now search the rest of the groups. We assume that
1250 * i and gdp correctly point to the last group visited.
1252 for (bgi = 0; bgi < ngroups; bgi++) {
1254 if (group_no >= ngroups)
1256 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1261 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1263 * skip this group if the number of
1264 * free blocks is less than half of the reservation
1267 if (free_blocks <= (windowsz/2))
1271 bitmap_bh = read_block_bitmap(sb, group_no);
1274 ret_block = ext3_try_to_allocate_with_rsv(sb, handle, group_no,
1275 bitmap_bh, -1, my_rsv, &fatal);
1282 * We may end up a bogus ealier ENOSPC error due to
1283 * filesystem is "full" of reservations, but
1284 * there maybe indeed free blocks avaliable on disk
1285 * In this case, we just forget about the reservations
1286 * just do block allocation as without reservations.
1290 group_no = goal_group;
1293 /* No space left on the device */
1299 ext3_debug("using block group %d(%d)\n",
1300 group_no, gdp->bg_free_blocks_count);
1302 BUFFER_TRACE(gdp_bh, "get_write_access");
1303 fatal = ext3_journal_get_write_access(handle, gdp_bh);
1307 target_block = ret_block + group_no * EXT3_BLOCKS_PER_GROUP(sb)
1308 + le32_to_cpu(es->s_first_data_block);
1310 if (target_block == le32_to_cpu(gdp->bg_block_bitmap) ||
1311 target_block == le32_to_cpu(gdp->bg_inode_bitmap) ||
1312 in_range(target_block, le32_to_cpu(gdp->bg_inode_table),
1313 EXT3_SB(sb)->s_itb_per_group))
1314 ext3_error(sb, "ext3_new_block",
1315 "Allocating block in system zone - "
1316 "block = %u", target_block);
1318 performed_allocation = 1;
1320 #ifdef CONFIG_JBD_DEBUG
1322 struct buffer_head *debug_bh;
1324 /* Record bitmap buffer state in the newly allocated block */
1325 debug_bh = sb_find_get_block(sb, target_block);
1327 BUFFER_TRACE(debug_bh, "state when allocated");
1328 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1332 jbd_lock_bh_state(bitmap_bh);
1333 spin_lock(sb_bgl_lock(sbi, group_no));
1334 if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1335 if (ext3_test_bit(ret_block,
1336 bh2jh(bitmap_bh)->b_committed_data)) {
1337 printk("%s: block was unexpectedly set in "
1338 "b_committed_data\n", __FUNCTION__);
1341 ext3_debug("found bit %d\n", ret_block);
1342 spin_unlock(sb_bgl_lock(sbi, group_no));
1343 jbd_unlock_bh_state(bitmap_bh);
1346 /* ret_block was blockgroup-relative. Now it becomes fs-relative */
1347 ret_block = target_block;
1349 if (ret_block >= le32_to_cpu(es->s_blocks_count)) {
1350 ext3_error(sb, "ext3_new_block",
1351 "block(%d) >= blocks count(%d) - "
1352 "block_group = %d, es == %p ", ret_block,
1353 le32_to_cpu(es->s_blocks_count), group_no, es);
1358 * It is up to the caller to add the new buffer to a journal
1359 * list of some description. We don't know in advance whether
1360 * the caller wants to use it as metadata or data.
1362 ext3_debug("allocating block %d. Goal hits %d of %d.\n",
1363 ret_block, goal_hits, goal_attempts);
1365 spin_lock(sb_bgl_lock(sbi, group_no));
1366 gdp->bg_free_blocks_count =
1367 cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count) - 1);
1368 spin_unlock(sb_bgl_lock(sbi, group_no));
1369 percpu_counter_mod(&sbi->s_freeblocks_counter, -1);
1371 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1372 err = ext3_journal_dirty_metadata(handle, gdp_bh);
1387 if (!performed_allocation)
1388 DLIMIT_FREE_BLOCK(sb, inode->i_xid, 1);
1392 ext3_std_error(sb, fatal);
1395 * Undo the block allocation
1397 if (!performed_allocation)
1398 DQUOT_FREE_BLOCK(inode, 1);
1403 unsigned long ext3_count_free_blocks(struct super_block *sb)
1405 unsigned long desc_count;
1406 struct ext3_group_desc *gdp;
1408 unsigned long ngroups;
1410 struct ext3_super_block *es;
1411 unsigned long bitmap_count, x;
1412 struct buffer_head *bitmap_bh = NULL;
1415 es = EXT3_SB(sb)->s_es;
1419 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
1420 gdp = ext3_get_group_desc(sb, i, NULL);
1423 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1425 bitmap_bh = read_block_bitmap(sb, i);
1426 if (bitmap_bh == NULL)
1429 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1430 printk("group %d: stored = %d, counted = %lu\n",
1431 i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1435 printk("ext3_count_free_blocks: stored = %u, computed = %lu, %lu\n",
1436 le32_to_cpu(es->s_free_blocks_count), desc_count, bitmap_count);
1438 return bitmap_count;
1441 ngroups = EXT3_SB(sb)->s_groups_count;
1443 for (i = 0; i < ngroups; i++) {
1444 gdp = ext3_get_group_desc(sb, i, NULL);
1447 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1454 static inline int block_in_use(unsigned long block,
1455 struct super_block * sb,
1456 unsigned char * map)
1458 return ext3_test_bit ((block -
1459 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block)) %
1460 EXT3_BLOCKS_PER_GROUP(sb), map);
1463 static inline int test_root(int a, int b)
1472 static int ext3_group_sparse(int group)
1476 return (test_root(group, 3) || test_root(group, 5) ||
1477 test_root(group, 7));
1481 * ext3_bg_has_super - number of blocks used by the superblock in group
1482 * @sb: superblock for filesystem
1483 * @group: group number to check
1485 * Return the number of blocks used by the superblock (primary or backup)
1486 * in this group. Currently this will be only 0 or 1.
1488 int ext3_bg_has_super(struct super_block *sb, int group)
1490 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1491 !ext3_group_sparse(group))
1497 * ext3_bg_num_gdb - number of blocks used by the group table in group
1498 * @sb: superblock for filesystem
1499 * @group: group number to check
1501 * Return the number of blocks used by the group descriptor table
1502 * (primary or backup) in this group. In the future there may be a
1503 * different number of descriptor blocks in each group.
1505 unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1507 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1508 !ext3_group_sparse(group))
1510 return EXT3_SB(sb)->s_gdb_count;
1513 #ifdef CONFIG_EXT3_CHECK
1514 /* Called at mount-time, super-block is locked */
1515 void ext3_check_blocks_bitmap (struct super_block * sb)
1517 struct ext3_super_block *es;
1518 unsigned long desc_count, bitmap_count, x, j;
1519 unsigned long desc_blocks;
1520 struct buffer_head *bitmap_bh = NULL;
1521 struct ext3_group_desc *gdp;
1524 es = EXT3_SB(sb)->s_es;
1528 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
1529 gdp = ext3_get_group_desc (sb, i, NULL);
1532 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1534 bitmap_bh = read_block_bitmap(sb, i);
1535 if (bitmap_bh == NULL)
1538 if (ext3_bg_has_super(sb, i) &&
1539 !ext3_test_bit(0, bitmap_bh->b_data))
1540 ext3_error(sb, __FUNCTION__,
1541 "Superblock in group %d is marked free", i);
1543 desc_blocks = ext3_bg_num_gdb(sb, i);
1544 for (j = 0; j < desc_blocks; j++)
1545 if (!ext3_test_bit(j + 1, bitmap_bh->b_data))
1546 ext3_error(sb, __FUNCTION__,
1547 "Descriptor block #%ld in group "
1548 "%d is marked free", j, i);
1550 if (!block_in_use (le32_to_cpu(gdp->bg_block_bitmap),
1551 sb, bitmap_bh->b_data))
1552 ext3_error (sb, "ext3_check_blocks_bitmap",
1553 "Block bitmap for group %d is marked free",
1556 if (!block_in_use (le32_to_cpu(gdp->bg_inode_bitmap),
1557 sb, bitmap_bh->b_data))
1558 ext3_error (sb, "ext3_check_blocks_bitmap",
1559 "Inode bitmap for group %d is marked free",
1562 for (j = 0; j < EXT3_SB(sb)->s_itb_per_group; j++)
1563 if (!block_in_use (le32_to_cpu(gdp->bg_inode_table) + j,
1564 sb, bitmap_bh->b_data))
1565 ext3_error (sb, "ext3_check_blocks_bitmap",
1566 "Block #%d of the inode table in "
1567 "group %d is marked free", j, i);
1569 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1570 if (le16_to_cpu(gdp->bg_free_blocks_count) != x)
1571 ext3_error (sb, "ext3_check_blocks_bitmap",
1572 "Wrong free blocks count for group %d, "
1573 "stored = %d, counted = %lu", i,
1574 le16_to_cpu(gdp->bg_free_blocks_count), x);
1578 if (le32_to_cpu(es->s_free_blocks_count) != bitmap_count)
1579 ext3_error (sb, "ext3_check_blocks_bitmap",
1580 "Wrong free blocks count in super block, "
1581 "stored = %lu, counted = %lu",
1582 (unsigned long)le32_to_cpu(es->s_free_blocks_count),