2 * linux/fs/ext4/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/jbd2.h>
18 #include <linux/ext4_fs.h>
19 #include <linux/ext4_jbd2.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 ext4_read_super).
39 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
42 * ext4_get_group_desc() -- load group descriptor from disk
44 * @block_group: given block group
45 * @bh: pointer to the buffer head to store the block
48 struct ext4_group_desc
* ext4_get_group_desc(struct super_block
* sb
,
49 unsigned int block_group
,
50 struct buffer_head
** bh
)
52 unsigned long group_desc
;
54 struct ext4_group_desc
* desc
;
55 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
57 if (block_group
>= sbi
->s_groups_count
) {
58 ext4_error (sb
, "ext4_get_group_desc",
59 "block_group >= groups_count - "
60 "block_group = %d, groups_count = %lu",
61 block_group
, sbi
->s_groups_count
);
67 group_desc
= block_group
>> EXT4_DESC_PER_BLOCK_BITS(sb
);
68 offset
= block_group
& (EXT4_DESC_PER_BLOCK(sb
) - 1);
69 if (!sbi
->s_group_desc
[group_desc
]) {
70 ext4_error (sb
, "ext4_get_group_desc",
71 "Group descriptor not loaded - "
72 "block_group = %d, group_desc = %lu, desc = %lu",
73 block_group
, group_desc
, offset
);
77 desc
= (struct ext4_group_desc
*) sbi
->s_group_desc
[group_desc
]->b_data
;
79 *bh
= sbi
->s_group_desc
[group_desc
];
86 * @block_group: given block group
88 * Read the bitmap for a given block_group, reading into the specified
89 * slot in the superblock's bitmap cache.
91 * Return buffer_head on success or NULL in case of failure.
93 static struct buffer_head
*
94 read_block_bitmap(struct super_block
*sb
, unsigned int block_group
)
96 struct ext4_group_desc
* desc
;
97 struct buffer_head
* bh
= NULL
;
99 desc
= ext4_get_group_desc (sb
, block_group
, NULL
);
102 bh
= sb_bread(sb
, ext4_block_bitmap(desc
));
104 ext4_error (sb
, "read_block_bitmap",
105 "Cannot read block bitmap - "
106 "block_group = %d, block_bitmap = %llu",
108 ext4_block_bitmap(desc
));
113 * The reservation window structure operations
114 * --------------------------------------------
115 * Operations include:
116 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
118 * We use a red-black tree to represent per-filesystem reservation
124 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
125 * @rb_root: root of per-filesystem reservation rb tree
126 * @verbose: verbose mode
127 * @fn: function which wishes to dump the reservation map
129 * If verbose is turned on, it will print the whole block reservation
130 * windows(start, end). Otherwise, it will only print out the "bad" windows,
131 * those windows that overlap with their immediate neighbors.
134 static void __rsv_window_dump(struct rb_root
*root
, int verbose
,
138 struct ext4_reserve_window_node
*rsv
, *prev
;
146 printk("Block Allocation Reservation Windows Map (%s):\n", fn
);
148 rsv
= list_entry(n
, struct ext4_reserve_window_node
, rsv_node
);
150 printk("reservation window 0x%p "
151 "start: %llu, end: %llu\n",
152 rsv
, rsv
->rsv_start
, rsv
->rsv_end
);
153 if (rsv
->rsv_start
&& rsv
->rsv_start
>= rsv
->rsv_end
) {
154 printk("Bad reservation %p (start >= end)\n",
158 if (prev
&& prev
->rsv_end
>= rsv
->rsv_start
) {
159 printk("Bad reservation %p (prev->end >= start)\n",
165 printk("Restarting reservation walk in verbose mode\n");
173 printk("Window map complete.\n");
177 #define rsv_window_dump(root, verbose) \
178 __rsv_window_dump((root), (verbose), __FUNCTION__)
180 #define rsv_window_dump(root, verbose) do {} while (0)
184 * goal_in_my_reservation()
185 * @rsv: inode's reservation window
186 * @grp_goal: given goal block relative to the allocation block group
187 * @group: the current allocation block group
188 * @sb: filesystem super block
190 * Test if the given goal block (group relative) is within the file's
191 * own block reservation window range.
193 * If the reservation window is outside the goal allocation group, return 0;
194 * grp_goal (given goal block) could be -1, which means no specific
195 * goal block. In this case, always return 1.
196 * If the goal block is within the reservation window, return 1;
197 * otherwise, return 0;
200 goal_in_my_reservation(struct ext4_reserve_window
*rsv
, ext4_grpblk_t grp_goal
,
201 unsigned int group
, struct super_block
* sb
)
203 ext4_fsblk_t group_first_block
, group_last_block
;
205 group_first_block
= ext4_group_first_block_no(sb
, group
);
206 group_last_block
= group_first_block
+ (EXT4_BLOCKS_PER_GROUP(sb
) - 1);
208 if ((rsv
->_rsv_start
> group_last_block
) ||
209 (rsv
->_rsv_end
< group_first_block
))
211 if ((grp_goal
>= 0) && ((grp_goal
+ group_first_block
< rsv
->_rsv_start
)
212 || (grp_goal
+ group_first_block
> rsv
->_rsv_end
)))
218 * search_reserve_window()
219 * @rb_root: root of reservation tree
220 * @goal: target allocation block
222 * Find the reserved window which includes the goal, or the previous one
223 * if the goal is not in any window.
224 * Returns NULL if there are no windows or if all windows start after the goal.
226 static struct ext4_reserve_window_node
*
227 search_reserve_window(struct rb_root
*root
, ext4_fsblk_t goal
)
229 struct rb_node
*n
= root
->rb_node
;
230 struct ext4_reserve_window_node
*rsv
;
236 rsv
= rb_entry(n
, struct ext4_reserve_window_node
, rsv_node
);
238 if (goal
< rsv
->rsv_start
)
240 else if (goal
> rsv
->rsv_end
)
246 * We've fallen off the end of the tree: the goal wasn't inside
247 * any particular node. OK, the previous node must be to one
248 * side of the interval containing the goal. If it's the RHS,
249 * we need to back up one.
251 if (rsv
->rsv_start
> goal
) {
252 n
= rb_prev(&rsv
->rsv_node
);
253 rsv
= rb_entry(n
, struct ext4_reserve_window_node
, rsv_node
);
259 * ext4_rsv_window_add() -- Insert a window to the block reservation rb tree.
261 * @rsv: reservation window to add
263 * Must be called with rsv_lock hold.
265 void ext4_rsv_window_add(struct super_block
*sb
,
266 struct ext4_reserve_window_node
*rsv
)
268 struct rb_root
*root
= &EXT4_SB(sb
)->s_rsv_window_root
;
269 struct rb_node
*node
= &rsv
->rsv_node
;
270 ext4_fsblk_t start
= rsv
->rsv_start
;
272 struct rb_node
** p
= &root
->rb_node
;
273 struct rb_node
* parent
= NULL
;
274 struct ext4_reserve_window_node
*this;
279 this = rb_entry(parent
, struct ext4_reserve_window_node
, rsv_node
);
281 if (start
< this->rsv_start
)
283 else if (start
> this->rsv_end
)
286 rsv_window_dump(root
, 1);
291 rb_link_node(node
, parent
, p
);
292 rb_insert_color(node
, root
);
296 * ext4_rsv_window_remove() -- unlink a window from the reservation rb tree
298 * @rsv: reservation window to remove
300 * Mark the block reservation window as not allocated, and unlink it
301 * from the filesystem reservation window rb tree. Must be called with
304 static void rsv_window_remove(struct super_block
*sb
,
305 struct ext4_reserve_window_node
*rsv
)
307 rsv
->rsv_start
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
308 rsv
->rsv_end
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
309 rsv
->rsv_alloc_hit
= 0;
310 rb_erase(&rsv
->rsv_node
, &EXT4_SB(sb
)->s_rsv_window_root
);
314 * rsv_is_empty() -- Check if the reservation window is allocated.
315 * @rsv: given reservation window to check
317 * returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED.
319 static inline int rsv_is_empty(struct ext4_reserve_window
*rsv
)
321 /* a valid reservation end block could not be 0 */
322 return rsv
->_rsv_end
== EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
326 * ext4_init_block_alloc_info()
327 * @inode: file inode structure
329 * Allocate and initialize the reservation window structure, and
330 * link the window to the ext4 inode structure at last
332 * The reservation window structure is only dynamically allocated
333 * and linked to ext4 inode the first time the open file
334 * needs a new block. So, before every ext4_new_block(s) call, for
335 * regular files, we should check whether the reservation window
336 * structure exists or not. In the latter case, this function is called.
337 * Fail to do so will result in block reservation being turned off for that
340 * This function is called from ext4_get_blocks_handle(), also called
341 * when setting the reservation window size through ioctl before the file
342 * is open for write (needs block allocation).
344 * Needs truncate_mutex protection prior to call this function.
346 void ext4_init_block_alloc_info(struct inode
*inode
)
348 struct ext4_inode_info
*ei
= EXT4_I(inode
);
349 struct ext4_block_alloc_info
*block_i
= ei
->i_block_alloc_info
;
350 struct super_block
*sb
= inode
->i_sb
;
352 block_i
= kmalloc(sizeof(*block_i
), GFP_NOFS
);
354 struct ext4_reserve_window_node
*rsv
= &block_i
->rsv_window_node
;
356 rsv
->rsv_start
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
357 rsv
->rsv_end
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
360 * if filesystem is mounted with NORESERVATION, the goal
361 * reservation window size is set to zero to indicate
362 * block reservation is off
364 if (!test_opt(sb
, RESERVATION
))
365 rsv
->rsv_goal_size
= 0;
367 rsv
->rsv_goal_size
= EXT4_DEFAULT_RESERVE_BLOCKS
;
368 rsv
->rsv_alloc_hit
= 0;
369 block_i
->last_alloc_logical_block
= 0;
370 block_i
->last_alloc_physical_block
= 0;
372 ei
->i_block_alloc_info
= block_i
;
376 * ext4_discard_reservation()
379 * Discard(free) block reservation window on last file close, or truncate
382 * It is being called in three cases:
383 * ext4_release_file(): last writer close the file
384 * ext4_clear_inode(): last iput(), when nobody link to this file.
385 * ext4_truncate(): when the block indirect map is about to change.
388 void ext4_discard_reservation(struct inode
*inode
)
390 struct ext4_inode_info
*ei
= EXT4_I(inode
);
391 struct ext4_block_alloc_info
*block_i
= ei
->i_block_alloc_info
;
392 struct ext4_reserve_window_node
*rsv
;
393 spinlock_t
*rsv_lock
= &EXT4_SB(inode
->i_sb
)->s_rsv_window_lock
;
398 rsv
= &block_i
->rsv_window_node
;
399 if (!rsv_is_empty(&rsv
->rsv_window
)) {
401 if (!rsv_is_empty(&rsv
->rsv_window
))
402 rsv_window_remove(inode
->i_sb
, rsv
);
403 spin_unlock(rsv_lock
);
408 * ext4_free_blocks_sb() -- Free given blocks and update quota
409 * @handle: handle to this transaction
411 * @block: start physcial block to free
412 * @count: number of blocks to free
413 * @pdquot_freed_blocks: pointer to quota
415 void ext4_free_blocks_sb(handle_t
*handle
, struct super_block
*sb
,
416 ext4_fsblk_t block
, unsigned long count
,
417 unsigned long *pdquot_freed_blocks
)
419 struct buffer_head
*bitmap_bh
= NULL
;
420 struct buffer_head
*gd_bh
;
421 unsigned long block_group
;
424 unsigned long overflow
;
425 struct ext4_group_desc
* desc
;
426 struct ext4_super_block
* es
;
427 struct ext4_sb_info
*sbi
;
429 ext4_grpblk_t group_freed
;
431 *pdquot_freed_blocks
= 0;
434 if (block
< le32_to_cpu(es
->s_first_data_block
) ||
435 block
+ count
< block
||
436 block
+ count
> ext4_blocks_count(es
)) {
437 ext4_error (sb
, "ext4_free_blocks",
438 "Freeing blocks not in datazone - "
439 "block = %llu, count = %lu", block
, count
);
443 ext4_debug ("freeing block(s) %llu-%llu\n", block
, block
+ count
- 1);
447 ext4_get_group_no_and_offset(sb
, block
, &block_group
, &bit
);
449 * Check to see if we are freeing blocks across a group
452 if (bit
+ count
> EXT4_BLOCKS_PER_GROUP(sb
)) {
453 overflow
= bit
+ count
- EXT4_BLOCKS_PER_GROUP(sb
);
457 bitmap_bh
= read_block_bitmap(sb
, block_group
);
460 desc
= ext4_get_group_desc (sb
, block_group
, &gd_bh
);
464 if (in_range(ext4_block_bitmap(desc
), block
, count
) ||
465 in_range(ext4_inode_bitmap(desc
), block
, count
) ||
466 in_range(block
, ext4_inode_table(desc
), sbi
->s_itb_per_group
) ||
467 in_range(block
+ count
- 1, ext4_inode_table(desc
),
468 sbi
->s_itb_per_group
))
469 ext4_error (sb
, "ext4_free_blocks",
470 "Freeing blocks in system zones - "
471 "Block = %llu, count = %lu",
475 * We are about to start releasing blocks in the bitmap,
476 * so we need undo access.
478 /* @@@ check errors */
479 BUFFER_TRACE(bitmap_bh
, "getting undo access");
480 err
= ext4_journal_get_undo_access(handle
, bitmap_bh
);
485 * We are about to modify some metadata. Call the journal APIs
486 * to unshare ->b_data if a currently-committing transaction is
489 BUFFER_TRACE(gd_bh
, "get_write_access");
490 err
= ext4_journal_get_write_access(handle
, gd_bh
);
494 jbd_lock_bh_state(bitmap_bh
);
496 for (i
= 0, group_freed
= 0; i
< count
; i
++) {
498 * An HJ special. This is expensive...
500 #ifdef CONFIG_JBD_DEBUG
501 jbd_unlock_bh_state(bitmap_bh
);
503 struct buffer_head
*debug_bh
;
504 debug_bh
= sb_find_get_block(sb
, block
+ i
);
506 BUFFER_TRACE(debug_bh
, "Deleted!");
507 if (!bh2jh(bitmap_bh
)->b_committed_data
)
508 BUFFER_TRACE(debug_bh
,
509 "No commited data in bitmap");
510 BUFFER_TRACE2(debug_bh
, bitmap_bh
, "bitmap");
514 jbd_lock_bh_state(bitmap_bh
);
516 if (need_resched()) {
517 jbd_unlock_bh_state(bitmap_bh
);
519 jbd_lock_bh_state(bitmap_bh
);
521 /* @@@ This prevents newly-allocated data from being
522 * freed and then reallocated within the same
525 * Ideally we would want to allow that to happen, but to
526 * do so requires making jbd2_journal_forget() capable of
527 * revoking the queued write of a data block, which
528 * implies blocking on the journal lock. *forget()
529 * cannot block due to truncate races.
531 * Eventually we can fix this by making jbd2_journal_forget()
532 * return a status indicating whether or not it was able
533 * to revoke the buffer. On successful revoke, it is
534 * safe not to set the allocation bit in the committed
535 * bitmap, because we know that there is no outstanding
536 * activity on the buffer any more and so it is safe to
539 BUFFER_TRACE(bitmap_bh
, "set in b_committed_data");
540 J_ASSERT_BH(bitmap_bh
,
541 bh2jh(bitmap_bh
)->b_committed_data
!= NULL
);
542 ext4_set_bit_atomic(sb_bgl_lock(sbi
, block_group
), bit
+ i
,
543 bh2jh(bitmap_bh
)->b_committed_data
);
546 * We clear the bit in the bitmap after setting the committed
547 * data bit, because this is the reverse order to that which
548 * the allocator uses.
550 BUFFER_TRACE(bitmap_bh
, "clear bit");
551 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi
, block_group
),
552 bit
+ i
, bitmap_bh
->b_data
)) {
553 jbd_unlock_bh_state(bitmap_bh
);
554 ext4_error(sb
, __FUNCTION__
,
555 "bit already cleared for block %llu",
556 (ext4_fsblk_t
)(block
+ i
));
557 jbd_lock_bh_state(bitmap_bh
);
558 BUFFER_TRACE(bitmap_bh
, "bit already cleared");
563 jbd_unlock_bh_state(bitmap_bh
);
565 spin_lock(sb_bgl_lock(sbi
, block_group
));
566 desc
->bg_free_blocks_count
=
567 cpu_to_le16(le16_to_cpu(desc
->bg_free_blocks_count
) +
569 spin_unlock(sb_bgl_lock(sbi
, block_group
));
570 percpu_counter_mod(&sbi
->s_freeblocks_counter
, count
);
572 /* We dirtied the bitmap block */
573 BUFFER_TRACE(bitmap_bh
, "dirtied bitmap block");
574 err
= ext4_journal_dirty_metadata(handle
, bitmap_bh
);
576 /* And the group descriptor block */
577 BUFFER_TRACE(gd_bh
, "dirtied group descriptor block");
578 ret
= ext4_journal_dirty_metadata(handle
, gd_bh
);
580 *pdquot_freed_blocks
+= group_freed
;
582 if (overflow
&& !err
) {
590 ext4_std_error(sb
, err
);
595 * ext4_free_blocks() -- Free given blocks and update quota
596 * @handle: handle for this transaction
598 * @block: start physical block to free
599 * @count: number of blocks to count
601 void ext4_free_blocks(handle_t
*handle
, struct inode
*inode
,
602 ext4_fsblk_t block
, unsigned long count
)
604 struct super_block
* sb
;
605 unsigned long dquot_freed_blocks
;
609 printk ("ext4_free_blocks: nonexistent device");
612 ext4_free_blocks_sb(handle
, sb
, block
, count
, &dquot_freed_blocks
);
613 if (dquot_freed_blocks
)
614 DQUOT_FREE_BLOCK(inode
, dquot_freed_blocks
);
619 * ext4_test_allocatable()
620 * @nr: given allocation block group
621 * @bh: bufferhead contains the bitmap of the given block group
623 * For ext4 allocations, we must not reuse any blocks which are
624 * allocated in the bitmap buffer's "last committed data" copy. This
625 * prevents deletes from freeing up the page for reuse until we have
626 * committed the delete transaction.
628 * If we didn't do this, then deleting something and reallocating it as
629 * data would allow the old block to be overwritten before the
630 * transaction committed (because we force data to disk before commit).
631 * This would lead to corruption if we crashed between overwriting the
632 * data and committing the delete.
634 * @@@ We may want to make this allocation behaviour conditional on
635 * data-writes at some point, and disable it for metadata allocations or
638 static int ext4_test_allocatable(ext4_grpblk_t nr
, struct buffer_head
*bh
)
641 struct journal_head
*jh
= bh2jh(bh
);
643 if (ext4_test_bit(nr
, bh
->b_data
))
646 jbd_lock_bh_state(bh
);
647 if (!jh
->b_committed_data
)
650 ret
= !ext4_test_bit(nr
, jh
->b_committed_data
);
651 jbd_unlock_bh_state(bh
);
656 * bitmap_search_next_usable_block()
657 * @start: the starting block (group relative) of the search
658 * @bh: bufferhead contains the block group bitmap
659 * @maxblocks: the ending block (group relative) of the reservation
661 * The bitmap search --- search forward alternately through the actual
662 * bitmap on disk and the last-committed copy in journal, until we find a
663 * bit free in both bitmaps.
666 bitmap_search_next_usable_block(ext4_grpblk_t start
, struct buffer_head
*bh
,
667 ext4_grpblk_t maxblocks
)
670 struct journal_head
*jh
= bh2jh(bh
);
672 while (start
< maxblocks
) {
673 next
= ext4_find_next_zero_bit(bh
->b_data
, maxblocks
, start
);
674 if (next
>= maxblocks
)
676 if (ext4_test_allocatable(next
, bh
))
678 jbd_lock_bh_state(bh
);
679 if (jh
->b_committed_data
)
680 start
= ext4_find_next_zero_bit(jh
->b_committed_data
,
682 jbd_unlock_bh_state(bh
);
688 * find_next_usable_block()
689 * @start: the starting block (group relative) to find next
690 * allocatable block in bitmap.
691 * @bh: bufferhead contains the block group bitmap
692 * @maxblocks: the ending block (group relative) for the search
694 * Find an allocatable block in a bitmap. We honor both the bitmap and
695 * its last-committed copy (if that exists), and perform the "most
696 * appropriate allocation" algorithm of looking for a free block near
697 * the initial goal; then for a free byte somewhere in the bitmap; then
698 * for any free bit in the bitmap.
701 find_next_usable_block(ext4_grpblk_t start
, struct buffer_head
*bh
,
702 ext4_grpblk_t maxblocks
)
704 ext4_grpblk_t here
, next
;
709 * The goal was occupied; search forward for a free
710 * block within the next XX blocks.
712 * end_goal is more or less random, but it has to be
713 * less than EXT4_BLOCKS_PER_GROUP. Aligning up to the
714 * next 64-bit boundary is simple..
716 ext4_grpblk_t end_goal
= (start
+ 63) & ~63;
717 if (end_goal
> maxblocks
)
718 end_goal
= maxblocks
;
719 here
= ext4_find_next_zero_bit(bh
->b_data
, end_goal
, start
);
720 if (here
< end_goal
&& ext4_test_allocatable(here
, bh
))
722 ext4_debug("Bit not found near goal\n");
729 p
= ((char *)bh
->b_data
) + (here
>> 3);
730 r
= memscan(p
, 0, (maxblocks
- here
+ 7) >> 3);
731 next
= (r
- ((char *)bh
->b_data
)) << 3;
733 if (next
< maxblocks
&& next
>= start
&& ext4_test_allocatable(next
, bh
))
737 * The bitmap search --- search forward alternately through the actual
738 * bitmap and the last-committed copy until we find a bit free in
741 here
= bitmap_search_next_usable_block(here
, bh
, maxblocks
);
747 * @block: the free block (group relative) to allocate
748 * @bh: the bufferhead containts the block group bitmap
750 * We think we can allocate this block in this bitmap. Try to set the bit.
751 * If that succeeds then check that nobody has allocated and then freed the
752 * block since we saw that is was not marked in b_committed_data. If it _was_
753 * allocated and freed then clear the bit in the bitmap again and return
757 claim_block(spinlock_t
*lock
, ext4_grpblk_t block
, struct buffer_head
*bh
)
759 struct journal_head
*jh
= bh2jh(bh
);
762 if (ext4_set_bit_atomic(lock
, block
, bh
->b_data
))
764 jbd_lock_bh_state(bh
);
765 if (jh
->b_committed_data
&& ext4_test_bit(block
,jh
->b_committed_data
)) {
766 ext4_clear_bit_atomic(lock
, block
, bh
->b_data
);
771 jbd_unlock_bh_state(bh
);
776 * ext4_try_to_allocate()
778 * @handle: handle to this transaction
779 * @group: given allocation block group
780 * @bitmap_bh: bufferhead holds the block bitmap
781 * @grp_goal: given target block within the group
782 * @count: target number of blocks to allocate
783 * @my_rsv: reservation window
785 * Attempt to allocate blocks within a give range. Set the range of allocation
786 * first, then find the first free bit(s) from the bitmap (within the range),
787 * and at last, allocate the blocks by claiming the found free bit as allocated.
789 * To set the range of this allocation:
790 * if there is a reservation window, only try to allocate block(s) from the
791 * file's own reservation window;
792 * Otherwise, the allocation range starts from the give goal block, ends at
793 * the block group's last block.
795 * If we failed to allocate the desired block then we may end up crossing to a
796 * new bitmap. In that case we must release write access to the old one via
797 * ext4_journal_release_buffer(), else we'll run out of credits.
800 ext4_try_to_allocate(struct super_block
*sb
, handle_t
*handle
, int group
,
801 struct buffer_head
*bitmap_bh
, ext4_grpblk_t grp_goal
,
802 unsigned long *count
, struct ext4_reserve_window
*my_rsv
)
804 ext4_fsblk_t group_first_block
;
805 ext4_grpblk_t start
, end
;
806 unsigned long num
= 0;
808 /* we do allocation within the reservation window if we have a window */
810 group_first_block
= ext4_group_first_block_no(sb
, group
);
811 if (my_rsv
->_rsv_start
>= group_first_block
)
812 start
= my_rsv
->_rsv_start
- group_first_block
;
814 /* reservation window cross group boundary */
816 end
= my_rsv
->_rsv_end
- group_first_block
+ 1;
817 if (end
> EXT4_BLOCKS_PER_GROUP(sb
))
818 /* reservation window crosses group boundary */
819 end
= EXT4_BLOCKS_PER_GROUP(sb
);
820 if ((start
<= grp_goal
) && (grp_goal
< end
))
829 end
= EXT4_BLOCKS_PER_GROUP(sb
);
832 BUG_ON(start
> EXT4_BLOCKS_PER_GROUP(sb
));
835 if (grp_goal
< 0 || !ext4_test_allocatable(grp_goal
, bitmap_bh
)) {
836 grp_goal
= find_next_usable_block(start
, bitmap_bh
, end
);
842 for (i
= 0; i
< 7 && grp_goal
> start
&&
843 ext4_test_allocatable(grp_goal
- 1,
851 if (!claim_block(sb_bgl_lock(EXT4_SB(sb
), group
),
852 grp_goal
, bitmap_bh
)) {
854 * The block was allocated by another thread, or it was
855 * allocated and then freed by another thread
865 while (num
< *count
&& grp_goal
< end
866 && ext4_test_allocatable(grp_goal
, bitmap_bh
)
867 && claim_block(sb_bgl_lock(EXT4_SB(sb
), group
),
868 grp_goal
, bitmap_bh
)) {
873 return grp_goal
- num
;
880 * find_next_reservable_window():
881 * find a reservable space within the given range.
882 * It does not allocate the reservation window for now:
883 * alloc_new_reservation() will do the work later.
885 * @search_head: the head of the searching list;
886 * This is not necessarily the list head of the whole filesystem
888 * We have both head and start_block to assist the search
889 * for the reservable space. The list starts from head,
890 * but we will shift to the place where start_block is,
891 * then start from there, when looking for a reservable space.
893 * @size: the target new reservation window size
895 * @group_first_block: the first block we consider to start
896 * the real search from
899 * the maximum block number that our goal reservable space
900 * could start from. This is normally the last block in this
901 * group. The search will end when we found the start of next
902 * possible reservable space is out of this boundary.
903 * This could handle the cross boundary reservation window
906 * basically we search from the given range, rather than the whole
907 * reservation double linked list, (start_block, last_block)
908 * to find a free region that is of my size and has not
912 static int find_next_reservable_window(
913 struct ext4_reserve_window_node
*search_head
,
914 struct ext4_reserve_window_node
*my_rsv
,
915 struct super_block
* sb
,
916 ext4_fsblk_t start_block
,
917 ext4_fsblk_t last_block
)
919 struct rb_node
*next
;
920 struct ext4_reserve_window_node
*rsv
, *prev
;
922 int size
= my_rsv
->rsv_goal_size
;
924 /* TODO: make the start of the reservation window byte-aligned */
925 /* cur = *start_block & ~7;*/
932 if (cur
<= rsv
->rsv_end
)
933 cur
= rsv
->rsv_end
+ 1;
936 * in the case we could not find a reservable space
937 * that is what is expected, during the re-search, we could
938 * remember what's the largest reservable space we could have
939 * and return that one.
941 * For now it will fail if we could not find the reservable
942 * space with expected-size (or more)...
944 if (cur
> last_block
)
945 return -1; /* fail */
948 next
= rb_next(&rsv
->rsv_node
);
949 rsv
= list_entry(next
,struct ext4_reserve_window_node
,rsv_node
);
952 * Reached the last reservation, we can just append to the
958 if (cur
+ size
<= rsv
->rsv_start
) {
960 * Found a reserveable space big enough. We could
961 * have a reservation across the group boundary here
967 * we come here either :
968 * when we reach the end of the whole list,
969 * and there is empty reservable space after last entry in the list.
970 * append it to the end of the list.
972 * or we found one reservable space in the middle of the list,
973 * return the reservation window that we could append to.
977 if ((prev
!= my_rsv
) && (!rsv_is_empty(&my_rsv
->rsv_window
)))
978 rsv_window_remove(sb
, my_rsv
);
981 * Let's book the whole avaliable window for now. We will check the
982 * disk bitmap later and then, if there are free blocks then we adjust
983 * the window size if it's larger than requested.
984 * Otherwise, we will remove this node from the tree next time
985 * call find_next_reservable_window.
987 my_rsv
->rsv_start
= cur
;
988 my_rsv
->rsv_end
= cur
+ size
- 1;
989 my_rsv
->rsv_alloc_hit
= 0;
992 ext4_rsv_window_add(sb
, my_rsv
);
998 * alloc_new_reservation()--allocate a new reservation window
1000 * To make a new reservation, we search part of the filesystem
1001 * reservation list (the list that inside the group). We try to
1002 * allocate a new reservation window near the allocation goal,
1003 * or the beginning of the group, if there is no goal.
1005 * We first find a reservable space after the goal, then from
1006 * there, we check the bitmap for the first free block after
1007 * it. If there is no free block until the end of group, then the
1008 * whole group is full, we failed. Otherwise, check if the free
1009 * block is inside the expected reservable space, if so, we
1011 * If the first free block is outside the reservable space, then
1012 * start from the first free block, we search for next available
1015 * on succeed, a new reservation will be found and inserted into the list
1016 * It contains at least one free block, and it does not overlap with other
1017 * reservation windows.
1019 * failed: we failed to find a reservation window in this group
1021 * @rsv: the reservation
1023 * @grp_goal: The goal (group-relative). It is where the search for a
1024 * free reservable space should start from.
1025 * if we have a grp_goal(grp_goal >0 ), then start from there,
1026 * no grp_goal(grp_goal = -1), we start from the first block
1029 * @sb: the super block
1030 * @group: the group we are trying to allocate in
1031 * @bitmap_bh: the block group block bitmap
1034 static int alloc_new_reservation(struct ext4_reserve_window_node
*my_rsv
,
1035 ext4_grpblk_t grp_goal
, struct super_block
*sb
,
1036 unsigned int group
, struct buffer_head
*bitmap_bh
)
1038 struct ext4_reserve_window_node
*search_head
;
1039 ext4_fsblk_t group_first_block
, group_end_block
, start_block
;
1040 ext4_grpblk_t first_free_block
;
1041 struct rb_root
*fs_rsv_root
= &EXT4_SB(sb
)->s_rsv_window_root
;
1044 spinlock_t
*rsv_lock
= &EXT4_SB(sb
)->s_rsv_window_lock
;
1046 group_first_block
= ext4_group_first_block_no(sb
, group
);
1047 group_end_block
= group_first_block
+ (EXT4_BLOCKS_PER_GROUP(sb
) - 1);
1050 start_block
= group_first_block
;
1052 start_block
= grp_goal
+ group_first_block
;
1054 size
= my_rsv
->rsv_goal_size
;
1056 if (!rsv_is_empty(&my_rsv
->rsv_window
)) {
1058 * if the old reservation is cross group boundary
1059 * and if the goal is inside the old reservation window,
1060 * we will come here when we just failed to allocate from
1061 * the first part of the window. We still have another part
1062 * that belongs to the next group. In this case, there is no
1063 * point to discard our window and try to allocate a new one
1064 * in this group(which will fail). we should
1065 * keep the reservation window, just simply move on.
1067 * Maybe we could shift the start block of the reservation
1068 * window to the first block of next group.
1071 if ((my_rsv
->rsv_start
<= group_end_block
) &&
1072 (my_rsv
->rsv_end
> group_end_block
) &&
1073 (start_block
>= my_rsv
->rsv_start
))
1076 if ((my_rsv
->rsv_alloc_hit
>
1077 (my_rsv
->rsv_end
- my_rsv
->rsv_start
+ 1) / 2)) {
1079 * if the previously allocation hit ratio is
1080 * greater than 1/2, then we double the size of
1081 * the reservation window the next time,
1082 * otherwise we keep the same size window
1085 if (size
> EXT4_MAX_RESERVE_BLOCKS
)
1086 size
= EXT4_MAX_RESERVE_BLOCKS
;
1087 my_rsv
->rsv_goal_size
= size
;
1091 spin_lock(rsv_lock
);
1093 * shift the search start to the window near the goal block
1095 search_head
= search_reserve_window(fs_rsv_root
, start_block
);
1098 * find_next_reservable_window() simply finds a reservable window
1099 * inside the given range(start_block, group_end_block).
1101 * To make sure the reservation window has a free bit inside it, we
1102 * need to check the bitmap after we found a reservable window.
1105 ret
= find_next_reservable_window(search_head
, my_rsv
, sb
,
1106 start_block
, group_end_block
);
1109 if (!rsv_is_empty(&my_rsv
->rsv_window
))
1110 rsv_window_remove(sb
, my_rsv
);
1111 spin_unlock(rsv_lock
);
1116 * On success, find_next_reservable_window() returns the
1117 * reservation window where there is a reservable space after it.
1118 * Before we reserve this reservable space, we need
1119 * to make sure there is at least a free block inside this region.
1121 * searching the first free bit on the block bitmap and copy of
1122 * last committed bitmap alternatively, until we found a allocatable
1123 * block. Search start from the start block of the reservable space
1126 spin_unlock(rsv_lock
);
1127 first_free_block
= bitmap_search_next_usable_block(
1128 my_rsv
->rsv_start
- group_first_block
,
1129 bitmap_bh
, group_end_block
- group_first_block
+ 1);
1131 if (first_free_block
< 0) {
1133 * no free block left on the bitmap, no point
1134 * to reserve the space. return failed.
1136 spin_lock(rsv_lock
);
1137 if (!rsv_is_empty(&my_rsv
->rsv_window
))
1138 rsv_window_remove(sb
, my_rsv
);
1139 spin_unlock(rsv_lock
);
1140 return -1; /* failed */
1143 start_block
= first_free_block
+ group_first_block
;
1145 * check if the first free block is within the
1146 * free space we just reserved
1148 if (start_block
>= my_rsv
->rsv_start
&& start_block
< my_rsv
->rsv_end
)
1149 return 0; /* success */
1151 * if the first free bit we found is out of the reservable space
1152 * continue search for next reservable space,
1153 * start from where the free block is,
1154 * we also shift the list head to where we stopped last time
1156 search_head
= my_rsv
;
1157 spin_lock(rsv_lock
);
1162 * try_to_extend_reservation()
1163 * @my_rsv: given reservation window
1165 * @size: the delta to extend
1167 * Attempt to expand the reservation window large enough to have
1168 * required number of free blocks
1170 * Since ext4_try_to_allocate() will always allocate blocks within
1171 * the reservation window range, if the window size is too small,
1172 * multiple blocks allocation has to stop at the end of the reservation
1173 * window. To make this more efficient, given the total number of
1174 * blocks needed and the current size of the window, we try to
1175 * expand the reservation window size if necessary on a best-effort
1176 * basis before ext4_new_blocks() tries to allocate blocks,
1178 static void try_to_extend_reservation(struct ext4_reserve_window_node
*my_rsv
,
1179 struct super_block
*sb
, int size
)
1181 struct ext4_reserve_window_node
*next_rsv
;
1182 struct rb_node
*next
;
1183 spinlock_t
*rsv_lock
= &EXT4_SB(sb
)->s_rsv_window_lock
;
1185 if (!spin_trylock(rsv_lock
))
1188 next
= rb_next(&my_rsv
->rsv_node
);
1191 my_rsv
->rsv_end
+= size
;
1193 next_rsv
= list_entry(next
, struct ext4_reserve_window_node
, rsv_node
);
1195 if ((next_rsv
->rsv_start
- my_rsv
->rsv_end
- 1) >= size
)
1196 my_rsv
->rsv_end
+= size
;
1198 my_rsv
->rsv_end
= next_rsv
->rsv_start
- 1;
1200 spin_unlock(rsv_lock
);
1204 * ext4_try_to_allocate_with_rsv()
1206 * @handle: handle to this transaction
1207 * @group: given allocation block group
1208 * @bitmap_bh: bufferhead holds the block bitmap
1209 * @grp_goal: given target block within the group
1210 * @count: target number of blocks to allocate
1211 * @my_rsv: reservation window
1212 * @errp: pointer to store the error code
1214 * This is the main function used to allocate a new block and its reservation
1217 * Each time when a new block allocation is need, first try to allocate from
1218 * its own reservation. If it does not have a reservation window, instead of
1219 * looking for a free bit on bitmap first, then look up the reservation list to
1220 * see if it is inside somebody else's reservation window, we try to allocate a
1221 * reservation window for it starting from the goal first. Then do the block
1222 * allocation within the reservation window.
1224 * This will avoid keeping on searching the reservation list again and
1225 * again when somebody is looking for a free block (without
1226 * reservation), and there are lots of free blocks, but they are all
1229 * We use a red-black tree for the per-filesystem reservation list.
1232 static ext4_grpblk_t
1233 ext4_try_to_allocate_with_rsv(struct super_block
*sb
, handle_t
*handle
,
1234 unsigned int group
, struct buffer_head
*bitmap_bh
,
1235 ext4_grpblk_t grp_goal
,
1236 struct ext4_reserve_window_node
* my_rsv
,
1237 unsigned long *count
, int *errp
)
1239 ext4_fsblk_t group_first_block
, group_last_block
;
1240 ext4_grpblk_t ret
= 0;
1242 unsigned long num
= *count
;
1247 * Make sure we use undo access for the bitmap, because it is critical
1248 * that we do the frozen_data COW on bitmap buffers in all cases even
1249 * if the buffer is in BJ_Forget state in the committing transaction.
1251 BUFFER_TRACE(bitmap_bh
, "get undo access for new block");
1252 fatal
= ext4_journal_get_undo_access(handle
, bitmap_bh
);
1259 * we don't deal with reservation when
1260 * filesystem is mounted without reservation
1261 * or the file is not a regular file
1262 * or last attempt to allocate a block with reservation turned on failed
1264 if (my_rsv
== NULL
) {
1265 ret
= ext4_try_to_allocate(sb
, handle
, group
, bitmap_bh
,
1266 grp_goal
, count
, NULL
);
1270 * grp_goal is a group relative block number (if there is a goal)
1271 * 0 < grp_goal < EXT4_BLOCKS_PER_GROUP(sb)
1272 * first block is a filesystem wide block number
1273 * first block is the block number of the first block in this group
1275 group_first_block
= ext4_group_first_block_no(sb
, group
);
1276 group_last_block
= group_first_block
+ (EXT4_BLOCKS_PER_GROUP(sb
) - 1);
1279 * Basically we will allocate a new block from inode's reservation
1282 * We need to allocate a new reservation window, if:
1283 * a) inode does not have a reservation window; or
1284 * b) last attempt to allocate a block from existing reservation
1286 * c) we come here with a goal and with a reservation window
1288 * We do not need to allocate a new reservation window if we come here
1289 * at the beginning with a goal and the goal is inside the window, or
1290 * we don't have a goal but already have a reservation window.
1291 * then we could go to allocate from the reservation window directly.
1294 if (rsv_is_empty(&my_rsv
->rsv_window
) || (ret
< 0) ||
1295 !goal_in_my_reservation(&my_rsv
->rsv_window
,
1296 grp_goal
, group
, sb
)) {
1297 if (my_rsv
->rsv_goal_size
< *count
)
1298 my_rsv
->rsv_goal_size
= *count
;
1299 ret
= alloc_new_reservation(my_rsv
, grp_goal
, sb
,
1304 if (!goal_in_my_reservation(&my_rsv
->rsv_window
,
1305 grp_goal
, group
, sb
))
1307 } else if (grp_goal
> 0 &&
1308 (my_rsv
->rsv_end
-grp_goal
+1) < *count
)
1309 try_to_extend_reservation(my_rsv
, sb
,
1310 *count
-my_rsv
->rsv_end
+ grp_goal
- 1);
1312 if ((my_rsv
->rsv_start
> group_last_block
) ||
1313 (my_rsv
->rsv_end
< group_first_block
)) {
1314 rsv_window_dump(&EXT4_SB(sb
)->s_rsv_window_root
, 1);
1317 ret
= ext4_try_to_allocate(sb
, handle
, group
, bitmap_bh
,
1318 grp_goal
, &num
, &my_rsv
->rsv_window
);
1320 my_rsv
->rsv_alloc_hit
+= num
;
1322 break; /* succeed */
1328 BUFFER_TRACE(bitmap_bh
, "journal_dirty_metadata for "
1330 fatal
= ext4_journal_dirty_metadata(handle
, bitmap_bh
);
1338 BUFFER_TRACE(bitmap_bh
, "journal_release_buffer");
1339 ext4_journal_release_buffer(handle
, bitmap_bh
);
1344 * ext4_has_free_blocks()
1345 * @sbi: in-core super block structure.
1347 * Check if filesystem has at least 1 free block available for allocation.
1349 static int ext4_has_free_blocks(struct ext4_sb_info
*sbi
)
1351 ext4_fsblk_t free_blocks
, root_blocks
;
1353 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
1354 root_blocks
= ext4_r_blocks_count(sbi
->s_es
);
1355 if (free_blocks
< root_blocks
+ 1 && !capable(CAP_SYS_RESOURCE
) &&
1356 sbi
->s_resuid
!= current
->fsuid
&&
1357 (sbi
->s_resgid
== 0 || !in_group_p (sbi
->s_resgid
))) {
1364 * ext4_should_retry_alloc()
1366 * @retries number of attemps has been made
1368 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
1369 * it is profitable to retry the operation, this function will wait
1370 * for the current or commiting transaction to complete, and then
1373 * if the total number of retries exceed three times, return FALSE.
1375 int ext4_should_retry_alloc(struct super_block
*sb
, int *retries
)
1377 if (!ext4_has_free_blocks(EXT4_SB(sb
)) || (*retries
)++ > 3)
1380 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb
->s_id
);
1382 return jbd2_journal_force_commit_nested(EXT4_SB(sb
)->s_journal
);
1386 * ext4_new_blocks() -- core block(s) allocation function
1387 * @handle: handle to this transaction
1388 * @inode: file inode
1389 * @goal: given target block(filesystem wide)
1390 * @count: target number of blocks to allocate
1393 * ext4_new_blocks uses a goal block to assist allocation. It tries to
1394 * allocate block(s) from the block group contains the goal block first. If that
1395 * fails, it will try to allocate block(s) from other block groups without
1396 * any specific goal block.
1399 ext4_fsblk_t
ext4_new_blocks(handle_t
*handle
, struct inode
*inode
,
1400 ext4_fsblk_t goal
, unsigned long *count
, int *errp
)
1402 struct buffer_head
*bitmap_bh
= NULL
;
1403 struct buffer_head
*gdp_bh
;
1404 unsigned long group_no
;
1406 ext4_grpblk_t grp_target_blk
; /* blockgroup relative goal block */
1407 ext4_grpblk_t grp_alloc_blk
; /* blockgroup-relative allocated block*/
1408 ext4_fsblk_t ret_block
; /* filesyetem-wide allocated block */
1409 int bgi
; /* blockgroup iteration index */
1411 int performed_allocation
= 0;
1412 ext4_grpblk_t free_blocks
; /* number of free blocks in a group */
1413 struct super_block
*sb
;
1414 struct ext4_group_desc
*gdp
;
1415 struct ext4_super_block
*es
;
1416 struct ext4_sb_info
*sbi
;
1417 struct ext4_reserve_window_node
*my_rsv
= NULL
;
1418 struct ext4_block_alloc_info
*block_i
;
1419 unsigned short windowsz
= 0;
1421 static int goal_hits
, goal_attempts
;
1423 unsigned long ngroups
;
1424 unsigned long num
= *count
;
1429 printk("ext4_new_block: nonexistent device");
1434 * Check quota for allocation of this block.
1436 if (DQUOT_ALLOC_BLOCK(inode
, num
)) {
1442 es
= EXT4_SB(sb
)->s_es
;
1443 ext4_debug("goal=%lu.\n", goal
);
1445 * Allocate a block from reservation only when
1446 * filesystem is mounted with reservation(default,-o reservation), and
1447 * it's a regular file, and
1448 * the desired window size is greater than 0 (One could use ioctl
1449 * command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off
1450 * reservation on that particular file)
1452 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
1453 if (block_i
&& ((windowsz
= block_i
->rsv_window_node
.rsv_goal_size
) > 0))
1454 my_rsv
= &block_i
->rsv_window_node
;
1456 if (!ext4_has_free_blocks(sbi
)) {
1462 * First, test whether the goal block is free.
1464 if (goal
< le32_to_cpu(es
->s_first_data_block
) ||
1465 goal
>= ext4_blocks_count(es
))
1466 goal
= le32_to_cpu(es
->s_first_data_block
);
1467 ext4_get_group_no_and_offset(sb
, goal
, &group_no
, &grp_target_blk
);
1468 goal_group
= group_no
;
1470 gdp
= ext4_get_group_desc(sb
, group_no
, &gdp_bh
);
1474 free_blocks
= le16_to_cpu(gdp
->bg_free_blocks_count
);
1476 * if there is not enough free blocks to make a new resevation
1477 * turn off reservation for this allocation
1479 if (my_rsv
&& (free_blocks
< windowsz
)
1480 && (rsv_is_empty(&my_rsv
->rsv_window
)))
1483 if (free_blocks
> 0) {
1484 bitmap_bh
= read_block_bitmap(sb
, group_no
);
1487 grp_alloc_blk
= ext4_try_to_allocate_with_rsv(sb
, handle
,
1488 group_no
, bitmap_bh
, grp_target_blk
,
1489 my_rsv
, &num
, &fatal
);
1492 if (grp_alloc_blk
>= 0)
1496 ngroups
= EXT4_SB(sb
)->s_groups_count
;
1500 * Now search the rest of the groups. We assume that
1501 * i and gdp correctly point to the last group visited.
1503 for (bgi
= 0; bgi
< ngroups
; bgi
++) {
1505 if (group_no
>= ngroups
)
1507 gdp
= ext4_get_group_desc(sb
, group_no
, &gdp_bh
);
1512 free_blocks
= le16_to_cpu(gdp
->bg_free_blocks_count
);
1514 * skip this group if the number of
1515 * free blocks is less than half of the reservation
1518 if (free_blocks
<= (windowsz
/2))
1522 bitmap_bh
= read_block_bitmap(sb
, group_no
);
1526 * try to allocate block(s) from this group, without a goal(-1).
1528 grp_alloc_blk
= ext4_try_to_allocate_with_rsv(sb
, handle
,
1529 group_no
, bitmap_bh
, -1, my_rsv
,
1533 if (grp_alloc_blk
>= 0)
1537 * We may end up a bogus ealier ENOSPC error due to
1538 * filesystem is "full" of reservations, but
1539 * there maybe indeed free blocks avaliable on disk
1540 * In this case, we just forget about the reservations
1541 * just do block allocation as without reservations.
1545 group_no
= goal_group
;
1548 /* No space left on the device */
1554 ext4_debug("using block group %d(%d)\n",
1555 group_no
, gdp
->bg_free_blocks_count
);
1557 BUFFER_TRACE(gdp_bh
, "get_write_access");
1558 fatal
= ext4_journal_get_write_access(handle
, gdp_bh
);
1562 ret_block
= grp_alloc_blk
+ ext4_group_first_block_no(sb
, group_no
);
1564 if (in_range(ext4_block_bitmap(gdp
), ret_block
, num
) ||
1565 in_range(ext4_block_bitmap(gdp
), ret_block
, num
) ||
1566 in_range(ret_block
, ext4_inode_table(gdp
),
1567 EXT4_SB(sb
)->s_itb_per_group
) ||
1568 in_range(ret_block
+ num
- 1, ext4_inode_table(gdp
),
1569 EXT4_SB(sb
)->s_itb_per_group
))
1570 ext4_error(sb
, "ext4_new_block",
1571 "Allocating block in system zone - "
1572 "blocks from %llu, length %lu",
1575 performed_allocation
= 1;
1577 #ifdef CONFIG_JBD_DEBUG
1579 struct buffer_head
*debug_bh
;
1581 /* Record bitmap buffer state in the newly allocated block */
1582 debug_bh
= sb_find_get_block(sb
, ret_block
);
1584 BUFFER_TRACE(debug_bh
, "state when allocated");
1585 BUFFER_TRACE2(debug_bh
, bitmap_bh
, "bitmap state");
1589 jbd_lock_bh_state(bitmap_bh
);
1590 spin_lock(sb_bgl_lock(sbi
, group_no
));
1591 if (buffer_jbd(bitmap_bh
) && bh2jh(bitmap_bh
)->b_committed_data
) {
1594 for (i
= 0; i
< num
; i
++) {
1595 if (ext4_test_bit(grp_alloc_blk
+i
,
1596 bh2jh(bitmap_bh
)->b_committed_data
)) {
1597 printk("%s: block was unexpectedly set in "
1598 "b_committed_data\n", __FUNCTION__
);
1602 ext4_debug("found bit %d\n", grp_alloc_blk
);
1603 spin_unlock(sb_bgl_lock(sbi
, group_no
));
1604 jbd_unlock_bh_state(bitmap_bh
);
1607 if (ret_block
+ num
- 1 >= ext4_blocks_count(es
)) {
1608 ext4_error(sb
, "ext4_new_block",
1609 "block(%llu) >= blocks count(%llu) - "
1610 "block_group = %lu, es == %p ", ret_block
,
1611 ext4_blocks_count(es
), group_no
, es
);
1616 * It is up to the caller to add the new buffer to a journal
1617 * list of some description. We don't know in advance whether
1618 * the caller wants to use it as metadata or data.
1620 ext4_debug("allocating block %lu. Goal hits %d of %d.\n",
1621 ret_block
, goal_hits
, goal_attempts
);
1623 spin_lock(sb_bgl_lock(sbi
, group_no
));
1624 gdp
->bg_free_blocks_count
=
1625 cpu_to_le16(le16_to_cpu(gdp
->bg_free_blocks_count
)-num
);
1626 spin_unlock(sb_bgl_lock(sbi
, group_no
));
1627 percpu_counter_mod(&sbi
->s_freeblocks_counter
, -num
);
1629 BUFFER_TRACE(gdp_bh
, "journal_dirty_metadata for group descriptor");
1630 err
= ext4_journal_dirty_metadata(handle
, gdp_bh
);
1640 DQUOT_FREE_BLOCK(inode
, *count
-num
);
1649 ext4_std_error(sb
, fatal
);
1652 * Undo the block allocation
1654 if (!performed_allocation
)
1655 DQUOT_FREE_BLOCK(inode
, *count
);
1660 ext4_fsblk_t
ext4_new_block(handle_t
*handle
, struct inode
*inode
,
1661 ext4_fsblk_t goal
, int *errp
)
1663 unsigned long count
= 1;
1665 return ext4_new_blocks(handle
, inode
, goal
, &count
, errp
);
1669 * ext4_count_free_blocks() -- count filesystem free blocks
1672 * Adds up the number of free blocks from each block group.
1674 ext4_fsblk_t
ext4_count_free_blocks(struct super_block
*sb
)
1676 ext4_fsblk_t desc_count
;
1677 struct ext4_group_desc
*gdp
;
1679 unsigned long ngroups
= EXT4_SB(sb
)->s_groups_count
;
1681 struct ext4_super_block
*es
;
1682 ext4_fsblk_t bitmap_count
;
1684 struct buffer_head
*bitmap_bh
= NULL
;
1686 es
= EXT4_SB(sb
)->s_es
;
1692 for (i
= 0; i
< ngroups
; i
++) {
1693 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1696 desc_count
+= le16_to_cpu(gdp
->bg_free_blocks_count
);
1698 bitmap_bh
= read_block_bitmap(sb
, i
);
1699 if (bitmap_bh
== NULL
)
1702 x
= ext4_count_free(bitmap_bh
, sb
->s_blocksize
);
1703 printk("group %d: stored = %d, counted = %lu\n",
1704 i
, le16_to_cpu(gdp
->bg_free_blocks_count
), x
);
1708 printk("ext4_count_free_blocks: stored = %llu"
1709 ", computed = %llu, %llu\n",
1710 EXT4_FREE_BLOCKS_COUNT(es
),
1711 desc_count
, bitmap_count
);
1712 return bitmap_count
;
1716 for (i
= 0; i
< ngroups
; i
++) {
1717 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1720 desc_count
+= le16_to_cpu(gdp
->bg_free_blocks_count
);
1728 block_in_use(ext4_fsblk_t block
, struct super_block
*sb
, unsigned char *map
)
1730 ext4_grpblk_t offset
;
1732 ext4_get_group_no_and_offset(sb
, block
, NULL
, &offset
);
1733 return ext4_test_bit (offset
, map
);
1736 static inline int test_root(int a
, int b
)
1745 static int ext4_group_sparse(int group
)
1751 return (test_root(group
, 7) || test_root(group
, 5) ||
1752 test_root(group
, 3));
1756 * ext4_bg_has_super - number of blocks used by the superblock in group
1757 * @sb: superblock for filesystem
1758 * @group: group number to check
1760 * Return the number of blocks used by the superblock (primary or backup)
1761 * in this group. Currently this will be only 0 or 1.
1763 int ext4_bg_has_super(struct super_block
*sb
, int group
)
1765 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1766 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
) &&
1767 !ext4_group_sparse(group
))
1772 static unsigned long ext4_bg_num_gdb_meta(struct super_block
*sb
, int group
)
1774 unsigned long metagroup
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1775 unsigned long first
= metagroup
* EXT4_DESC_PER_BLOCK(sb
);
1776 unsigned long last
= first
+ EXT4_DESC_PER_BLOCK(sb
) - 1;
1778 if (group
== first
|| group
== first
+ 1 || group
== last
)
1783 static unsigned long ext4_bg_num_gdb_nometa(struct super_block
*sb
, int group
)
1785 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1786 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
) &&
1787 !ext4_group_sparse(group
))
1789 return EXT4_SB(sb
)->s_gdb_count
;
1793 * ext4_bg_num_gdb - number of blocks used by the group table in group
1794 * @sb: superblock for filesystem
1795 * @group: group number to check
1797 * Return the number of blocks used by the group descriptor table
1798 * (primary or backup) in this group. In the future there may be a
1799 * different number of descriptor blocks in each group.
1801 unsigned long ext4_bg_num_gdb(struct super_block
*sb
, int group
)
1803 unsigned long first_meta_bg
=
1804 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_meta_bg
);
1805 unsigned long metagroup
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1807 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
,EXT4_FEATURE_INCOMPAT_META_BG
) ||
1808 metagroup
< first_meta_bg
)
1809 return ext4_bg_num_gdb_nometa(sb
,group
);
1811 return ext4_bg_num_gdb_meta(sb
,group
);