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>
16 #include <linux/capability.h>
18 #include <linux/jbd.h>
19 #include <linux/ext3_fs.h>
20 #include <linux/ext3_jbd.h>
21 #include <linux/quotaops.h>
22 #include <linux/buffer_head.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 * desc;
49 struct ext3_sb_info *sbi = EXT3_SB(sb);
51 if (block_group >= sbi->s_groups_count) {
52 ext3_error (sb, "ext3_get_group_desc",
53 "block_group >= groups_count - "
54 "block_group = %d, groups_count = %lu",
55 block_group, sbi->s_groups_count);
61 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
62 offset = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
63 if (!sbi->s_group_desc[group_desc]) {
64 ext3_error (sb, "ext3_get_group_desc",
65 "Group descriptor not loaded - "
66 "block_group = %d, group_desc = %lu, desc = %lu",
67 block_group, group_desc, offset);
71 desc = (struct ext3_group_desc *) sbi->s_group_desc[group_desc]->b_data;
73 *bh = sbi->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, ext3_grpblk_t grp_goal,
167 unsigned int group, struct super_block * sb)
169 ext3_fsblk_t group_first_block, group_last_block;
171 group_first_block = ext3_group_first_block_no(sb, group);
172 group_last_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1;
174 if ((rsv->_rsv_start > group_last_block) ||
175 (rsv->_rsv_end < group_first_block))
177 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
178 || (grp_goal + group_first_block > rsv->_rsv_end)))
184 * Find the reserved window which includes the goal, or the previous one
185 * if the goal is not in any window.
186 * Returns NULL if there are no windows or if all windows start after the goal.
188 static struct ext3_reserve_window_node *
189 search_reserve_window(struct rb_root *root, ext3_fsblk_t goal)
191 struct rb_node *n = root->rb_node;
192 struct ext3_reserve_window_node *rsv;
198 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
200 if (goal < rsv->rsv_start)
202 else if (goal > rsv->rsv_end)
208 * We've fallen off the end of the tree: the goal wasn't inside
209 * any particular node. OK, the previous node must be to one
210 * side of the interval containing the goal. If it's the RHS,
211 * we need to back up one.
213 if (rsv->rsv_start > goal) {
214 n = rb_prev(&rsv->rsv_node);
215 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
220 void ext3_rsv_window_add(struct super_block *sb,
221 struct ext3_reserve_window_node *rsv)
223 struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
224 struct rb_node *node = &rsv->rsv_node;
225 ext3_fsblk_t start = rsv->rsv_start;
227 struct rb_node ** p = &root->rb_node;
228 struct rb_node * parent = NULL;
229 struct ext3_reserve_window_node *this;
234 this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
236 if (start < this->rsv_start)
238 else if (start > this->rsv_end)
244 rb_link_node(node, parent, p);
245 rb_insert_color(node, root);
248 static void rsv_window_remove(struct super_block *sb,
249 struct ext3_reserve_window_node *rsv)
251 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
252 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
253 rsv->rsv_alloc_hit = 0;
254 rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
257 static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
259 /* a valid reservation end block could not be 0 */
260 return (rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED);
262 void ext3_init_block_alloc_info(struct inode *inode)
264 struct ext3_inode_info *ei = EXT3_I(inode);
265 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
266 struct super_block *sb = inode->i_sb;
268 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
270 struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node;
272 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
273 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
276 * if filesystem is mounted with NORESERVATION, the goal
277 * reservation window size is set to zero to indicate
278 * block reservation is off
280 if (!test_opt(sb, RESERVATION))
281 rsv->rsv_goal_size = 0;
283 rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS;
284 rsv->rsv_alloc_hit = 0;
285 block_i->last_alloc_logical_block = 0;
286 block_i->last_alloc_physical_block = 0;
288 ei->i_block_alloc_info = block_i;
291 void ext3_discard_reservation(struct inode *inode)
293 struct ext3_inode_info *ei = EXT3_I(inode);
294 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
295 struct ext3_reserve_window_node *rsv;
296 spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
301 rsv = &block_i->rsv_window_node;
302 if (!rsv_is_empty(&rsv->rsv_window)) {
304 if (!rsv_is_empty(&rsv->rsv_window))
305 rsv_window_remove(inode->i_sb, rsv);
306 spin_unlock(rsv_lock);
310 /* Free given blocks, update quota and i_blocks field */
311 void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
312 ext3_fsblk_t block, unsigned long count,
313 unsigned long *pdquot_freed_blocks)
315 struct buffer_head *bitmap_bh = NULL;
316 struct buffer_head *gd_bh;
317 unsigned long block_group;
320 unsigned long overflow;
321 struct ext3_group_desc * desc;
322 struct ext3_super_block * es;
323 struct ext3_sb_info *sbi;
325 ext3_grpblk_t group_freed;
327 *pdquot_freed_blocks = 0;
330 if (block < le32_to_cpu(es->s_first_data_block) ||
331 block + count < block ||
332 block + count > le32_to_cpu(es->s_blocks_count)) {
333 ext3_error (sb, "ext3_free_blocks",
334 "Freeing blocks not in datazone - "
335 "block = "E3FSBLK", count = %lu", block, count);
339 ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
343 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
344 EXT3_BLOCKS_PER_GROUP(sb);
345 bit = (block - le32_to_cpu(es->s_first_data_block)) %
346 EXT3_BLOCKS_PER_GROUP(sb);
348 * Check to see if we are freeing blocks across a group
351 if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
352 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
356 bitmap_bh = read_block_bitmap(sb, block_group);
359 desc = ext3_get_group_desc (sb, block_group, &gd_bh);
363 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
364 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
365 in_range (block, le32_to_cpu(desc->bg_inode_table),
366 sbi->s_itb_per_group) ||
367 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
368 sbi->s_itb_per_group))
369 ext3_error (sb, "ext3_free_blocks",
370 "Freeing blocks in system zones - "
371 "Block = "E3FSBLK", count = %lu",
375 * We are about to start releasing blocks in the bitmap,
376 * so we need undo access.
378 /* @@@ check errors */
379 BUFFER_TRACE(bitmap_bh, "getting undo access");
380 err = ext3_journal_get_undo_access(handle, bitmap_bh);
385 * We are about to modify some metadata. Call the journal APIs
386 * to unshare ->b_data if a currently-committing transaction is
389 BUFFER_TRACE(gd_bh, "get_write_access");
390 err = ext3_journal_get_write_access(handle, gd_bh);
394 jbd_lock_bh_state(bitmap_bh);
396 for (i = 0, group_freed = 0; i < count; i++) {
398 * An HJ special. This is expensive...
400 #ifdef CONFIG_JBD_DEBUG
401 jbd_unlock_bh_state(bitmap_bh);
403 struct buffer_head *debug_bh;
404 debug_bh = sb_find_get_block(sb, block + i);
406 BUFFER_TRACE(debug_bh, "Deleted!");
407 if (!bh2jh(bitmap_bh)->b_committed_data)
408 BUFFER_TRACE(debug_bh,
409 "No commited data in bitmap");
410 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
414 jbd_lock_bh_state(bitmap_bh);
416 if (need_resched()) {
417 jbd_unlock_bh_state(bitmap_bh);
419 jbd_lock_bh_state(bitmap_bh);
421 /* @@@ This prevents newly-allocated data from being
422 * freed and then reallocated within the same
425 * Ideally we would want to allow that to happen, but to
426 * do so requires making journal_forget() capable of
427 * revoking the queued write of a data block, which
428 * implies blocking on the journal lock. *forget()
429 * cannot block due to truncate races.
431 * Eventually we can fix this by making journal_forget()
432 * return a status indicating whether or not it was able
433 * to revoke the buffer. On successful revoke, it is
434 * safe not to set the allocation bit in the committed
435 * bitmap, because we know that there is no outstanding
436 * activity on the buffer any more and so it is safe to
439 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
440 J_ASSERT_BH(bitmap_bh,
441 bh2jh(bitmap_bh)->b_committed_data != NULL);
442 ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
443 bh2jh(bitmap_bh)->b_committed_data);
446 * We clear the bit in the bitmap after setting the committed
447 * data bit, because this is the reverse order to that which
448 * the allocator uses.
450 BUFFER_TRACE(bitmap_bh, "clear bit");
451 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
452 bit + i, bitmap_bh->b_data)) {
453 jbd_unlock_bh_state(bitmap_bh);
454 ext3_error(sb, __FUNCTION__,
455 "bit already cleared for block "E3FSBLK,
457 jbd_lock_bh_state(bitmap_bh);
458 BUFFER_TRACE(bitmap_bh, "bit already cleared");
463 jbd_unlock_bh_state(bitmap_bh);
465 spin_lock(sb_bgl_lock(sbi, block_group));
466 desc->bg_free_blocks_count =
467 cpu_to_le16(le16_to_cpu(desc->bg_free_blocks_count) +
469 spin_unlock(sb_bgl_lock(sbi, block_group));
470 percpu_counter_mod(&sbi->s_freeblocks_counter, count);
472 /* We dirtied the bitmap block */
473 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
474 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
476 /* And the group descriptor block */
477 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
478 ret = ext3_journal_dirty_metadata(handle, gd_bh);
480 *pdquot_freed_blocks += group_freed;
482 if (overflow && !err) {
490 ext3_std_error(sb, err);
494 /* Free given blocks, update quota and i_blocks field */
495 void ext3_free_blocks(handle_t *handle, struct inode *inode,
496 ext3_fsblk_t block, unsigned long count)
498 struct super_block * sb;
499 unsigned long dquot_freed_blocks;
503 printk ("ext3_free_blocks: nonexistent device");
506 ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
507 if (dquot_freed_blocks)
508 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
513 * For ext3 allocations, we must not reuse any blocks which are
514 * allocated in the bitmap buffer's "last committed data" copy. This
515 * prevents deletes from freeing up the page for reuse until we have
516 * committed the delete transaction.
518 * If we didn't do this, then deleting something and reallocating it as
519 * data would allow the old block to be overwritten before the
520 * transaction committed (because we force data to disk before commit).
521 * This would lead to corruption if we crashed between overwriting the
522 * data and committing the delete.
524 * @@@ We may want to make this allocation behaviour conditional on
525 * data-writes at some point, and disable it for metadata allocations or
528 static int ext3_test_allocatable(ext3_grpblk_t nr, struct buffer_head *bh)
531 struct journal_head *jh = bh2jh(bh);
533 if (ext3_test_bit(nr, bh->b_data))
536 jbd_lock_bh_state(bh);
537 if (!jh->b_committed_data)
540 ret = !ext3_test_bit(nr, jh->b_committed_data);
541 jbd_unlock_bh_state(bh);
546 bitmap_search_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
547 ext3_grpblk_t maxblocks)
550 struct journal_head *jh = bh2jh(bh);
553 * The bitmap search --- search forward alternately through the actual
554 * bitmap and the last-committed copy until we find a bit free in
557 while (start < maxblocks) {
558 next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
559 if (next >= maxblocks)
561 if (ext3_test_allocatable(next, bh))
563 jbd_lock_bh_state(bh);
564 if (jh->b_committed_data)
565 start = ext3_find_next_zero_bit(jh->b_committed_data,
567 jbd_unlock_bh_state(bh);
573 * Find an allocatable block in a bitmap. We honour both the bitmap and
574 * its last-committed copy (if that exists), and perform the "most
575 * appropriate allocation" algorithm of looking for a free block near
576 * the initial goal; then for a free byte somewhere in the bitmap; then
577 * for any free bit in the bitmap.
580 find_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
581 ext3_grpblk_t maxblocks)
583 ext3_grpblk_t here, next;
588 * The goal was occupied; search forward for a free
589 * block within the next XX blocks.
591 * end_goal is more or less random, but it has to be
592 * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
593 * next 64-bit boundary is simple..
595 ext3_grpblk_t end_goal = (start + 63) & ~63;
596 if (end_goal > maxblocks)
597 end_goal = maxblocks;
598 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
599 if (here < end_goal && ext3_test_allocatable(here, bh))
601 ext3_debug("Bit not found near goal\n");
608 p = ((char *)bh->b_data) + (here >> 3);
609 r = memscan(p, 0, (maxblocks - here + 7) >> 3);
610 next = (r - ((char *)bh->b_data)) << 3;
612 if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
616 * The bitmap search --- search forward alternately through the actual
617 * bitmap and the last-committed copy until we find a bit free in
620 here = bitmap_search_next_usable_block(here, bh, maxblocks);
625 * We think we can allocate this block in this bitmap. Try to set the bit.
626 * If that succeeds then check that nobody has allocated and then freed the
627 * block since we saw that is was not marked in b_committed_data. If it _was_
628 * allocated and freed then clear the bit in the bitmap again and return
632 claim_block(spinlock_t *lock, ext3_grpblk_t block, struct buffer_head *bh)
634 struct journal_head *jh = bh2jh(bh);
637 if (ext3_set_bit_atomic(lock, block, bh->b_data))
639 jbd_lock_bh_state(bh);
640 if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
641 ext3_clear_bit_atomic(lock, block, bh->b_data);
646 jbd_unlock_bh_state(bh);
651 * If we failed to allocate the desired block then we may end up crossing to a
652 * new bitmap. In that case we must release write access to the old one via
653 * ext3_journal_release_buffer(), else we'll run out of credits.
656 ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
657 struct buffer_head *bitmap_bh, ext3_grpblk_t grp_goal,
658 unsigned long *count, struct ext3_reserve_window *my_rsv)
660 ext3_fsblk_t group_first_block;
661 ext3_grpblk_t start, end;
662 unsigned long num = 0;
664 /* we do allocation within the reservation window if we have a window */
666 group_first_block = ext3_group_first_block_no(sb, group);
667 if (my_rsv->_rsv_start >= group_first_block)
668 start = my_rsv->_rsv_start - group_first_block;
670 /* reservation window cross group boundary */
672 end = my_rsv->_rsv_end - group_first_block + 1;
673 if (end > EXT3_BLOCKS_PER_GROUP(sb))
674 /* reservation window crosses group boundary */
675 end = EXT3_BLOCKS_PER_GROUP(sb);
676 if ((start <= grp_goal) && (grp_goal < end))
685 end = EXT3_BLOCKS_PER_GROUP(sb);
688 BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
691 if (grp_goal < 0 || !ext3_test_allocatable(grp_goal, bitmap_bh)) {
692 grp_goal = find_next_usable_block(start, bitmap_bh, end);
698 for (i = 0; i < 7 && grp_goal > start &&
699 ext3_test_allocatable(grp_goal - 1,
707 if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group), grp_goal, bitmap_bh)) {
709 * The block was allocated by another thread, or it was
710 * allocated and then freed by another thread
720 while (num < *count && grp_goal < end
721 && ext3_test_allocatable(grp_goal, bitmap_bh)
722 && claim_block(sb_bgl_lock(EXT3_SB(sb), group), grp_goal, bitmap_bh)) {
727 return grp_goal - num;
734 * find_next_reservable_window():
735 * find a reservable space within the given range.
736 * It does not allocate the reservation window for now:
737 * alloc_new_reservation() will do the work later.
739 * @search_head: the head of the searching list;
740 * This is not necessarily the list head of the whole filesystem
742 * We have both head and start_block to assist the search
743 * for the reservable space. The list starts from head,
744 * but we will shift to the place where start_block is,
745 * then start from there, when looking for a reservable space.
747 * @size: the target new reservation window size
749 * @group_first_block: the first block we consider to start
750 * the real search from
753 * the maximum block number that our goal reservable space
754 * could start from. This is normally the last block in this
755 * group. The search will end when we found the start of next
756 * possible reservable space is out of this boundary.
757 * This could handle the cross boundary reservation window
760 * basically we search from the given range, rather than the whole
761 * reservation double linked list, (start_block, last_block)
762 * to find a free region that is of my size and has not
766 static int find_next_reservable_window(
767 struct ext3_reserve_window_node *search_head,
768 struct ext3_reserve_window_node *my_rsv,
769 struct super_block * sb,
770 ext3_fsblk_t start_block,
771 ext3_fsblk_t last_block)
773 struct rb_node *next;
774 struct ext3_reserve_window_node *rsv, *prev;
776 int size = my_rsv->rsv_goal_size;
778 /* TODO: make the start of the reservation window byte-aligned */
779 /* cur = *start_block & ~7;*/
786 if (cur <= rsv->rsv_end)
787 cur = rsv->rsv_end + 1;
790 * in the case we could not find a reservable space
791 * that is what is expected, during the re-search, we could
792 * remember what's the largest reservable space we could have
793 * and return that one.
795 * For now it will fail if we could not find the reservable
796 * space with expected-size (or more)...
798 if (cur > last_block)
799 return -1; /* fail */
802 next = rb_next(&rsv->rsv_node);
803 rsv = list_entry(next,struct ext3_reserve_window_node,rsv_node);
806 * Reached the last reservation, we can just append to the
812 if (cur + size <= rsv->rsv_start) {
814 * Found a reserveable space big enough. We could
815 * have a reservation across the group boundary here
821 * we come here either :
822 * when we reach the end of the whole list,
823 * and there is empty reservable space after last entry in the list.
824 * append it to the end of the list.
826 * or we found one reservable space in the middle of the list,
827 * return the reservation window that we could append to.
831 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
832 rsv_window_remove(sb, my_rsv);
835 * Let's book the whole avaliable window for now. We will check the
836 * disk bitmap later and then, if there are free blocks then we adjust
837 * the window size if it's larger than requested.
838 * Otherwise, we will remove this node from the tree next time
839 * call find_next_reservable_window.
841 my_rsv->rsv_start = cur;
842 my_rsv->rsv_end = cur + size - 1;
843 my_rsv->rsv_alloc_hit = 0;
846 ext3_rsv_window_add(sb, my_rsv);
852 * alloc_new_reservation()--allocate a new reservation window
854 * To make a new reservation, we search part of the filesystem
855 * reservation list (the list that inside the group). We try to
856 * allocate a new reservation window near the allocation goal,
857 * or the beginning of the group, if there is no goal.
859 * We first find a reservable space after the goal, then from
860 * there, we check the bitmap for the first free block after
861 * it. If there is no free block until the end of group, then the
862 * whole group is full, we failed. Otherwise, check if the free
863 * block is inside the expected reservable space, if so, we
865 * If the first free block is outside the reservable space, then
866 * start from the first free block, we search for next available
869 * on succeed, a new reservation will be found and inserted into the list
870 * It contains at least one free block, and it does not overlap with other
871 * reservation windows.
873 * failed: we failed to find a reservation window in this group
875 * @rsv: the reservation
877 * @grp_goal: The goal (group-relative). It is where the search for a
878 * free reservable space should start from.
879 * if we have a grp_goal(grp_goal >0 ), then start from there,
880 * no grp_goal(grp_goal = -1), we start from the first block
883 * @sb: the super block
884 * @group: the group we are trying to allocate in
885 * @bitmap_bh: the block group block bitmap
888 static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
889 ext3_grpblk_t grp_goal, struct super_block *sb,
890 unsigned int group, struct buffer_head *bitmap_bh)
892 struct ext3_reserve_window_node *search_head;
893 ext3_fsblk_t group_first_block, group_end_block, start_block;
894 ext3_grpblk_t first_free_block;
895 struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
898 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
900 group_first_block = ext3_group_first_block_no(sb, group);
901 group_end_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1;
904 start_block = group_first_block;
906 start_block = grp_goal + group_first_block;
908 size = my_rsv->rsv_goal_size;
910 if (!rsv_is_empty(&my_rsv->rsv_window)) {
912 * if the old reservation is cross group boundary
913 * and if the goal is inside the old reservation window,
914 * we will come here when we just failed to allocate from
915 * the first part of the window. We still have another part
916 * that belongs to the next group. In this case, there is no
917 * point to discard our window and try to allocate a new one
918 * in this group(which will fail). we should
919 * keep the reservation window, just simply move on.
921 * Maybe we could shift the start block of the reservation
922 * window to the first block of next group.
925 if ((my_rsv->rsv_start <= group_end_block) &&
926 (my_rsv->rsv_end > group_end_block) &&
927 (start_block >= my_rsv->rsv_start))
930 if ((my_rsv->rsv_alloc_hit >
931 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
933 * if we previously allocation hit ration is greater than half
934 * we double the size of reservation window next time
935 * otherwise keep the same
938 if (size > EXT3_MAX_RESERVE_BLOCKS)
939 size = EXT3_MAX_RESERVE_BLOCKS;
940 my_rsv->rsv_goal_size= size;
946 * shift the search start to the window near the goal block
948 search_head = search_reserve_window(fs_rsv_root, start_block);
951 * find_next_reservable_window() simply finds a reservable window
952 * inside the given range(start_block, group_end_block).
954 * To make sure the reservation window has a free bit inside it, we
955 * need to check the bitmap after we found a reservable window.
958 ret = find_next_reservable_window(search_head, my_rsv, sb,
959 start_block, group_end_block);
962 if (!rsv_is_empty(&my_rsv->rsv_window))
963 rsv_window_remove(sb, my_rsv);
964 spin_unlock(rsv_lock);
969 * On success, find_next_reservable_window() returns the
970 * reservation window where there is a reservable space after it.
971 * Before we reserve this reservable space, we need
972 * to make sure there is at least a free block inside this region.
974 * searching the first free bit on the block bitmap and copy of
975 * last committed bitmap alternatively, until we found a allocatable
976 * block. Search start from the start block of the reservable space
979 spin_unlock(rsv_lock);
980 first_free_block = bitmap_search_next_usable_block(
981 my_rsv->rsv_start - group_first_block,
982 bitmap_bh, group_end_block - group_first_block + 1);
984 if (first_free_block < 0) {
986 * no free block left on the bitmap, no point
987 * to reserve the space. return failed.
990 if (!rsv_is_empty(&my_rsv->rsv_window))
991 rsv_window_remove(sb, my_rsv);
992 spin_unlock(rsv_lock);
993 return -1; /* failed */
996 start_block = first_free_block + group_first_block;
998 * check if the first free block is within the
999 * free space we just reserved
1001 if (start_block >= my_rsv->rsv_start && start_block < my_rsv->rsv_end)
1002 return 0; /* success */
1004 * if the first free bit we found is out of the reservable space
1005 * continue search for next reservable space,
1006 * start from where the free block is,
1007 * we also shift the list head to where we stopped last time
1009 search_head = my_rsv;
1010 spin_lock(rsv_lock);
1014 static void try_to_extend_reservation(struct ext3_reserve_window_node *my_rsv,
1015 struct super_block *sb, int size)
1017 struct ext3_reserve_window_node *next_rsv;
1018 struct rb_node *next;
1019 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1021 if (!spin_trylock(rsv_lock))
1024 next = rb_next(&my_rsv->rsv_node);
1027 my_rsv->rsv_end += size;
1029 next_rsv = list_entry(next, struct ext3_reserve_window_node, rsv_node);
1031 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1032 my_rsv->rsv_end += size;
1034 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1036 spin_unlock(rsv_lock);
1040 * This is the main function used to allocate a new block and its reservation
1043 * Each time when a new block allocation is need, first try to allocate from
1044 * its own reservation. If it does not have a reservation window, instead of
1045 * looking for a free bit on bitmap first, then look up the reservation list to
1046 * see if it is inside somebody else's reservation window, we try to allocate a
1047 * reservation window for it starting from the goal first. Then do the block
1048 * allocation within the reservation window.
1050 * This will avoid keeping on searching the reservation list again and
1051 * again when somebody is looking for a free block (without
1052 * reservation), and there are lots of free blocks, but they are all
1055 * We use a sorted double linked list for the per-filesystem reservation list.
1056 * The insert, remove and find a free space(non-reserved) operations for the
1057 * sorted double linked list should be fast.
1060 static ext3_grpblk_t
1061 ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1062 unsigned int group, struct buffer_head *bitmap_bh,
1063 ext3_grpblk_t grp_goal,
1064 struct ext3_reserve_window_node * my_rsv,
1065 unsigned long *count, int *errp)
1067 ext3_fsblk_t group_first_block;
1068 ext3_grpblk_t ret = 0;
1070 unsigned long num = *count;
1075 * Make sure we use undo access for the bitmap, because it is critical
1076 * that we do the frozen_data COW on bitmap buffers in all cases even
1077 * if the buffer is in BJ_Forget state in the committing transaction.
1079 BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1080 fatal = ext3_journal_get_undo_access(handle, bitmap_bh);
1087 * we don't deal with reservation when
1088 * filesystem is mounted without reservation
1089 * or the file is not a regular file
1090 * or last attempt to allocate a block with reservation turned on failed
1092 if (my_rsv == NULL ) {
1093 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1094 grp_goal, count, NULL);
1098 * grp_goal is a group relative block number (if there is a goal)
1099 * 0 < grp_goal < EXT3_BLOCKS_PER_GROUP(sb)
1100 * first block is a filesystem wide block number
1101 * first block is the block number of the first block in this group
1103 group_first_block = ext3_group_first_block_no(sb, group);
1106 * Basically we will allocate a new block from inode's reservation
1109 * We need to allocate a new reservation window, if:
1110 * a) inode does not have a reservation window; or
1111 * b) last attempt to allocate a block from existing reservation
1113 * c) we come here with a goal and with a reservation window
1115 * We do not need to allocate a new reservation window if we come here
1116 * at the beginning with a goal and the goal is inside the window, or
1117 * we don't have a goal but already have a reservation window.
1118 * then we could go to allocate from the reservation window directly.
1121 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1122 !goal_in_my_reservation(&my_rsv->rsv_window, grp_goal, group, sb)) {
1123 if (my_rsv->rsv_goal_size < *count)
1124 my_rsv->rsv_goal_size = *count;
1125 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1130 if (!goal_in_my_reservation(&my_rsv->rsv_window, grp_goal, group, sb))
1132 } else if (grp_goal > 0 && (my_rsv->rsv_end-grp_goal+1) < *count)
1133 try_to_extend_reservation(my_rsv, sb,
1134 *count-my_rsv->rsv_end + grp_goal - 1);
1136 if ((my_rsv->rsv_start >= group_first_block + EXT3_BLOCKS_PER_GROUP(sb))
1137 || (my_rsv->rsv_end < group_first_block))
1139 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh, grp_goal,
1140 &num, &my_rsv->rsv_window);
1142 my_rsv->rsv_alloc_hit += num;
1144 break; /* succeed */
1150 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1152 fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1160 BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1161 ext3_journal_release_buffer(handle, bitmap_bh);
1165 static int ext3_has_free_blocks(struct ext3_sb_info *sbi)
1167 ext3_fsblk_t free_blocks, root_blocks;
1169 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1170 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1171 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1172 sbi->s_resuid != current->fsuid &&
1173 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1180 * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1181 * it is profitable to retry the operation, this function will wait
1182 * for the current or commiting transaction to complete, and then
1185 int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1187 if (!ext3_has_free_blocks(EXT3_SB(sb)) || (*retries)++ > 3)
1190 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1192 return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1196 * ext3_new_block uses a goal block to assist allocation. If the goal is
1197 * free, or there is a free block within 32 blocks of the goal, that block
1198 * is allocated. Otherwise a forward search is made for a free block; within
1199 * each block group the search first looks for an entire free byte in the block
1200 * bitmap, and then for any free bit if that fails.
1201 * This function also updates quota and i_blocks field.
1203 ext3_fsblk_t ext3_new_blocks(handle_t *handle, struct inode *inode,
1204 ext3_fsblk_t goal, unsigned long *count, int *errp)
1206 struct buffer_head *bitmap_bh = NULL;
1207 struct buffer_head *gdp_bh;
1210 ext3_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1211 ext3_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1212 ext3_fsblk_t ret_block; /* filesyetem-wide allocated block */
1213 int bgi; /* blockgroup iteration index */
1215 int performed_allocation = 0;
1216 ext3_grpblk_t free_blocks; /* number of free blocks in a group */
1217 struct super_block *sb;
1218 struct ext3_group_desc *gdp;
1219 struct ext3_super_block *es;
1220 struct ext3_sb_info *sbi;
1221 struct ext3_reserve_window_node *my_rsv = NULL;
1222 struct ext3_block_alloc_info *block_i;
1223 unsigned short windowsz = 0;
1225 static int goal_hits, goal_attempts;
1227 unsigned long ngroups;
1228 unsigned long num = *count;
1233 printk("ext3_new_block: nonexistent device");
1238 * Check quota for allocation of this block.
1240 if (DQUOT_ALLOC_BLOCK(inode, num)) {
1246 es = EXT3_SB(sb)->s_es;
1247 ext3_debug("goal=%lu.\n", goal);
1249 * Allocate a block from reservation only when
1250 * filesystem is mounted with reservation(default,-o reservation), and
1251 * it's a regular file, and
1252 * the desired window size is greater than 0 (One could use ioctl
1253 * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1254 * reservation on that particular file)
1256 block_i = EXT3_I(inode)->i_block_alloc_info;
1257 if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1258 my_rsv = &block_i->rsv_window_node;
1260 if (!ext3_has_free_blocks(sbi)) {
1266 * First, test whether the goal block is free.
1268 if (goal < le32_to_cpu(es->s_first_data_block) ||
1269 goal >= le32_to_cpu(es->s_blocks_count))
1270 goal = le32_to_cpu(es->s_first_data_block);
1271 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1272 EXT3_BLOCKS_PER_GROUP(sb);
1273 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1277 goal_group = group_no;
1279 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1281 * if there is not enough free blocks to make a new resevation
1282 * turn off reservation for this allocation
1284 if (my_rsv && (free_blocks < windowsz)
1285 && (rsv_is_empty(&my_rsv->rsv_window)))
1288 if (free_blocks > 0) {
1289 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1290 EXT3_BLOCKS_PER_GROUP(sb));
1291 bitmap_bh = read_block_bitmap(sb, group_no);
1294 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1295 group_no, bitmap_bh, grp_target_blk,
1296 my_rsv, &num, &fatal);
1299 if (grp_alloc_blk >= 0)
1303 ngroups = EXT3_SB(sb)->s_groups_count;
1307 * Now search the rest of the groups. We assume that
1308 * i and gdp correctly point to the last group visited.
1310 for (bgi = 0; bgi < ngroups; bgi++) {
1312 if (group_no >= ngroups)
1314 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1319 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1321 * skip this group if the number of
1322 * free blocks is less than half of the reservation
1325 if (free_blocks <= (windowsz/2))
1329 bitmap_bh = read_block_bitmap(sb, group_no);
1333 * try to allocate block(s) from this group, without a goal(-1).
1335 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1336 group_no, bitmap_bh, -1, my_rsv,
1340 if (grp_alloc_blk >= 0)
1344 * We may end up a bogus ealier ENOSPC error due to
1345 * filesystem is "full" of reservations, but
1346 * there maybe indeed free blocks avaliable on disk
1347 * In this case, we just forget about the reservations
1348 * just do block allocation as without reservations.
1352 group_no = goal_group;
1355 /* No space left on the device */
1361 ext3_debug("using block group %d(%d)\n",
1362 group_no, gdp->bg_free_blocks_count);
1364 BUFFER_TRACE(gdp_bh, "get_write_access");
1365 fatal = ext3_journal_get_write_access(handle, gdp_bh);
1369 ret_block = grp_alloc_blk + ext3_group_first_block_no(sb, group_no);
1371 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1372 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1373 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1374 EXT3_SB(sb)->s_itb_per_group) ||
1375 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1376 EXT3_SB(sb)->s_itb_per_group))
1377 ext3_error(sb, "ext3_new_block",
1378 "Allocating block in system zone - "
1379 "blocks from "E3FSBLK", length %lu",
1382 performed_allocation = 1;
1384 #ifdef CONFIG_JBD_DEBUG
1386 struct buffer_head *debug_bh;
1388 /* Record bitmap buffer state in the newly allocated block */
1389 debug_bh = sb_find_get_block(sb, ret_block);
1391 BUFFER_TRACE(debug_bh, "state when allocated");
1392 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1396 jbd_lock_bh_state(bitmap_bh);
1397 spin_lock(sb_bgl_lock(sbi, group_no));
1398 if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1401 for (i = 0; i < num; i++) {
1402 if (ext3_test_bit(grp_alloc_blk+i,
1403 bh2jh(bitmap_bh)->b_committed_data)) {
1404 printk("%s: block was unexpectedly set in "
1405 "b_committed_data\n", __FUNCTION__);
1409 ext3_debug("found bit %d\n", grp_alloc_blk);
1410 spin_unlock(sb_bgl_lock(sbi, group_no));
1411 jbd_unlock_bh_state(bitmap_bh);
1414 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1415 ext3_error(sb, "ext3_new_block",
1416 "block("E3FSBLK") >= blocks count(%d) - "
1417 "block_group = %d, es == %p ", ret_block,
1418 le32_to_cpu(es->s_blocks_count), group_no, es);
1423 * It is up to the caller to add the new buffer to a journal
1424 * list of some description. We don't know in advance whether
1425 * the caller wants to use it as metadata or data.
1427 ext3_debug("allocating block %lu. Goal hits %d of %d.\n",
1428 ret_block, goal_hits, goal_attempts);
1430 spin_lock(sb_bgl_lock(sbi, group_no));
1431 gdp->bg_free_blocks_count =
1432 cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count) - num);
1433 spin_unlock(sb_bgl_lock(sbi, group_no));
1434 percpu_counter_mod(&sbi->s_freeblocks_counter, -num);
1436 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1437 err = ext3_journal_dirty_metadata(handle, gdp_bh);
1447 DQUOT_FREE_BLOCK(inode, *count-num);
1456 ext3_std_error(sb, fatal);
1459 * Undo the block allocation
1461 if (!performed_allocation)
1462 DQUOT_FREE_BLOCK(inode, *count);
1467 ext3_fsblk_t ext3_new_block(handle_t *handle, struct inode *inode,
1468 ext3_fsblk_t goal, int *errp)
1470 unsigned long count = 1;
1472 return ext3_new_blocks(handle, inode, goal, &count, errp);
1475 ext3_fsblk_t ext3_count_free_blocks(struct super_block *sb)
1477 ext3_fsblk_t desc_count;
1478 struct ext3_group_desc *gdp;
1480 unsigned long ngroups = EXT3_SB(sb)->s_groups_count;
1482 struct ext3_super_block *es;
1483 ext3_fsblk_t bitmap_count;
1485 struct buffer_head *bitmap_bh = NULL;
1487 es = EXT3_SB(sb)->s_es;
1493 for (i = 0; i < ngroups; i++) {
1494 gdp = ext3_get_group_desc(sb, i, NULL);
1497 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1499 bitmap_bh = read_block_bitmap(sb, i);
1500 if (bitmap_bh == NULL)
1503 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1504 printk("group %d: stored = %d, counted = %lu\n",
1505 i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1509 printk("ext3_count_free_blocks: stored = "E3FSBLK
1510 ", computed = "E3FSBLK", "E3FSBLK"\n",
1511 le32_to_cpu(es->s_free_blocks_count),
1512 desc_count, bitmap_count);
1513 return bitmap_count;
1517 for (i = 0; i < ngroups; i++) {
1518 gdp = ext3_get_group_desc(sb, i, NULL);
1521 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1529 block_in_use(ext3_fsblk_t block, struct super_block *sb, unsigned char *map)
1531 return ext3_test_bit ((block -
1532 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block)) %
1533 EXT3_BLOCKS_PER_GROUP(sb), map);
1536 static inline int test_root(int a, int b)
1545 static int ext3_group_sparse(int group)
1551 return (test_root(group, 7) || test_root(group, 5) ||
1552 test_root(group, 3));
1556 * ext3_bg_has_super - number of blocks used by the superblock in group
1557 * @sb: superblock for filesystem
1558 * @group: group number to check
1560 * Return the number of blocks used by the superblock (primary or backup)
1561 * in this group. Currently this will be only 0 or 1.
1563 int ext3_bg_has_super(struct super_block *sb, int group)
1565 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1566 EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1567 !ext3_group_sparse(group))
1572 static unsigned long ext3_bg_num_gdb_meta(struct super_block *sb, int group)
1574 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1575 unsigned long first = metagroup * EXT3_DESC_PER_BLOCK(sb);
1576 unsigned long last = first + EXT3_DESC_PER_BLOCK(sb) - 1;
1578 if (group == first || group == first + 1 || group == last)
1583 static unsigned long ext3_bg_num_gdb_nometa(struct super_block *sb, int group)
1585 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1586 EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1587 !ext3_group_sparse(group))
1589 return EXT3_SB(sb)->s_gdb_count;
1593 * ext3_bg_num_gdb - number of blocks used by the group table in group
1594 * @sb: superblock for filesystem
1595 * @group: group number to check
1597 * Return the number of blocks used by the group descriptor table
1598 * (primary or backup) in this group. In the future there may be a
1599 * different number of descriptor blocks in each group.
1601 unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1603 unsigned long first_meta_bg =
1604 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_meta_bg);
1605 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1607 if (!EXT3_HAS_INCOMPAT_FEATURE(sb,EXT3_FEATURE_INCOMPAT_META_BG) ||
1608 metagroup < first_meta_bg)
1609 return ext3_bg_num_gdb_nometa(sb,group);
1611 return ext3_bg_num_gdb_meta(sb,group);