2 * linux/fs/ext2/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
15 #include <linux/quotaops.h>
16 #include <linux/slab.h>
17 #include <linux/sched.h>
18 #include <linux/buffer_head.h>
19 #include <linux/capability.h>
22 * balloc.c contains the blocks allocation and deallocation routines
26 * The free blocks are managed by bitmaps. A file system contains several
27 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
28 * block for inodes, N blocks for the inode table and data blocks.
30 * The file system contains group descriptors which are located after the
31 * super block. Each descriptor contains the number of the bitmap block and
32 * the free blocks count in the block. The descriptors are loaded in memory
33 * when a file system is mounted (see ext2_fill_super).
37 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
39 struct ext2_group_desc
* ext2_get_group_desc(struct super_block
* sb
,
40 unsigned int block_group
,
41 struct buffer_head
** bh
)
43 unsigned long group_desc
;
45 struct ext2_group_desc
* desc
;
46 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
48 if (block_group
>= sbi
->s_groups_count
) {
49 ext2_error (sb
, "ext2_get_group_desc",
50 "block_group >= groups_count - "
51 "block_group = %d, groups_count = %lu",
52 block_group
, sbi
->s_groups_count
);
57 group_desc
= block_group
>> EXT2_DESC_PER_BLOCK_BITS(sb
);
58 offset
= block_group
& (EXT2_DESC_PER_BLOCK(sb
) - 1);
59 if (!sbi
->s_group_desc
[group_desc
]) {
60 ext2_error (sb
, "ext2_get_group_desc",
61 "Group descriptor not loaded - "
62 "block_group = %d, group_desc = %lu, desc = %lu",
63 block_group
, group_desc
, offset
);
67 desc
= (struct ext2_group_desc
*) sbi
->s_group_desc
[group_desc
]->b_data
;
69 *bh
= sbi
->s_group_desc
[group_desc
];
73 static int ext2_valid_block_bitmap(struct super_block
*sb
,
74 struct ext2_group_desc
*desc
,
75 unsigned int block_group
,
76 struct buffer_head
*bh
)
79 ext2_grpblk_t next_zero_bit
;
80 ext2_fsblk_t bitmap_blk
;
81 ext2_fsblk_t group_first_block
;
83 group_first_block
= ext2_group_first_block_no(sb
, block_group
);
85 /* check whether block bitmap block number is set */
86 bitmap_blk
= le32_to_cpu(desc
->bg_block_bitmap
);
87 offset
= bitmap_blk
- group_first_block
;
88 if (!ext2_test_bit(offset
, bh
->b_data
))
89 /* bad block bitmap */
92 /* check whether the inode bitmap block number is set */
93 bitmap_blk
= le32_to_cpu(desc
->bg_inode_bitmap
);
94 offset
= bitmap_blk
- group_first_block
;
95 if (!ext2_test_bit(offset
, bh
->b_data
))
96 /* bad block bitmap */
99 /* check whether the inode table block number is set */
100 bitmap_blk
= le32_to_cpu(desc
->bg_inode_table
);
101 offset
= bitmap_blk
- group_first_block
;
102 next_zero_bit
= ext2_find_next_zero_bit(bh
->b_data
,
103 offset
+ EXT2_SB(sb
)->s_itb_per_group
,
105 if (next_zero_bit
>= offset
+ EXT2_SB(sb
)->s_itb_per_group
)
106 /* good bitmap for inode tables */
110 ext2_error(sb
, __func__
,
111 "Invalid block bitmap - "
112 "block_group = %d, block = %lu",
113 block_group
, bitmap_blk
);
118 * Read the bitmap for a given block_group,and validate the
119 * bits for block/inode/inode tables are set in the bitmaps
121 * Return buffer_head on success or NULL in case of failure.
123 static struct buffer_head
*
124 read_block_bitmap(struct super_block
*sb
, unsigned int block_group
)
126 struct ext2_group_desc
* desc
;
127 struct buffer_head
* bh
= NULL
;
128 ext2_fsblk_t bitmap_blk
;
130 desc
= ext2_get_group_desc(sb
, block_group
, NULL
);
133 bitmap_blk
= le32_to_cpu(desc
->bg_block_bitmap
);
134 bh
= sb_getblk(sb
, bitmap_blk
);
136 ext2_error(sb
, __func__
,
137 "Cannot read block bitmap - "
138 "block_group = %d, block_bitmap = %u",
139 block_group
, le32_to_cpu(desc
->bg_block_bitmap
));
142 if (likely(bh_uptodate_or_lock(bh
)))
145 if (bh_submit_read(bh
) < 0) {
147 ext2_error(sb
, __func__
,
148 "Cannot read block bitmap - "
149 "block_group = %d, block_bitmap = %u",
150 block_group
, le32_to_cpu(desc
->bg_block_bitmap
));
154 ext2_valid_block_bitmap(sb
, desc
, block_group
, bh
);
156 * file system mounted not to panic on error, continue with corrupt
162 static void release_blocks(struct super_block
*sb
, int count
)
165 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
167 percpu_counter_add(&sbi
->s_freeblocks_counter
, count
);
171 static void group_adjust_blocks(struct super_block
*sb
, int group_no
,
172 struct ext2_group_desc
*desc
, struct buffer_head
*bh
, int count
)
175 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
176 unsigned free_blocks
;
178 spin_lock(sb_bgl_lock(sbi
, group_no
));
179 free_blocks
= le16_to_cpu(desc
->bg_free_blocks_count
);
180 desc
->bg_free_blocks_count
= cpu_to_le16(free_blocks
+ count
);
181 spin_unlock(sb_bgl_lock(sbi
, group_no
));
182 mark_buffer_dirty(bh
);
187 * The reservation window structure operations
188 * --------------------------------------------
189 * Operations include:
190 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
192 * We use a red-black tree to represent per-filesystem reservation
198 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
199 * @rb_root: root of per-filesystem reservation rb tree
200 * @verbose: verbose mode
201 * @fn: function which wishes to dump the reservation map
203 * If verbose is turned on, it will print the whole block reservation
204 * windows(start, end). Otherwise, it will only print out the "bad" windows,
205 * those windows that overlap with their immediate neighbors.
208 static void __rsv_window_dump(struct rb_root
*root
, int verbose
,
212 struct ext2_reserve_window_node
*rsv
, *prev
;
220 printk("Block Allocation Reservation Windows Map (%s):\n", fn
);
222 rsv
= rb_entry(n
, struct ext2_reserve_window_node
, rsv_node
);
224 printk("reservation window 0x%p "
225 "start: %lu, end: %lu\n",
226 rsv
, rsv
->rsv_start
, rsv
->rsv_end
);
227 if (rsv
->rsv_start
&& rsv
->rsv_start
>= rsv
->rsv_end
) {
228 printk("Bad reservation %p (start >= end)\n",
232 if (prev
&& prev
->rsv_end
>= rsv
->rsv_start
) {
233 printk("Bad reservation %p (prev->end >= start)\n",
239 printk("Restarting reservation walk in verbose mode\n");
247 printk("Window map complete.\n");
250 #define rsv_window_dump(root, verbose) \
251 __rsv_window_dump((root), (verbose), __func__)
253 #define rsv_window_dump(root, verbose) do {} while (0)
257 * goal_in_my_reservation()
258 * @rsv: inode's reservation window
259 * @grp_goal: given goal block relative to the allocation block group
260 * @group: the current allocation block group
261 * @sb: filesystem super block
263 * Test if the given goal block (group relative) is within the file's
264 * own block reservation window range.
266 * If the reservation window is outside the goal allocation group, return 0;
267 * grp_goal (given goal block) could be -1, which means no specific
268 * goal block. In this case, always return 1.
269 * If the goal block is within the reservation window, return 1;
270 * otherwise, return 0;
273 goal_in_my_reservation(struct ext2_reserve_window
*rsv
, ext2_grpblk_t grp_goal
,
274 unsigned int group
, struct super_block
* sb
)
276 ext2_fsblk_t group_first_block
, group_last_block
;
278 group_first_block
= ext2_group_first_block_no(sb
, group
);
279 group_last_block
= group_first_block
+ EXT2_BLOCKS_PER_GROUP(sb
) - 1;
281 if ((rsv
->_rsv_start
> group_last_block
) ||
282 (rsv
->_rsv_end
< group_first_block
))
284 if ((grp_goal
>= 0) && ((grp_goal
+ group_first_block
< rsv
->_rsv_start
)
285 || (grp_goal
+ group_first_block
> rsv
->_rsv_end
)))
291 * search_reserve_window()
292 * @rb_root: root of reservation tree
293 * @goal: target allocation block
295 * Find the reserved window which includes the goal, or the previous one
296 * if the goal is not in any window.
297 * Returns NULL if there are no windows or if all windows start after the goal.
299 static struct ext2_reserve_window_node
*
300 search_reserve_window(struct rb_root
*root
, ext2_fsblk_t goal
)
302 struct rb_node
*n
= root
->rb_node
;
303 struct ext2_reserve_window_node
*rsv
;
309 rsv
= rb_entry(n
, struct ext2_reserve_window_node
, rsv_node
);
311 if (goal
< rsv
->rsv_start
)
313 else if (goal
> rsv
->rsv_end
)
319 * We've fallen off the end of the tree: the goal wasn't inside
320 * any particular node. OK, the previous node must be to one
321 * side of the interval containing the goal. If it's the RHS,
322 * we need to back up one.
324 if (rsv
->rsv_start
> goal
) {
325 n
= rb_prev(&rsv
->rsv_node
);
326 rsv
= rb_entry(n
, struct ext2_reserve_window_node
, rsv_node
);
332 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
334 * @rsv: reservation window to add
336 * Must be called with rsv_lock held.
338 void ext2_rsv_window_add(struct super_block
*sb
,
339 struct ext2_reserve_window_node
*rsv
)
341 struct rb_root
*root
= &EXT2_SB(sb
)->s_rsv_window_root
;
342 struct rb_node
*node
= &rsv
->rsv_node
;
343 ext2_fsblk_t start
= rsv
->rsv_start
;
345 struct rb_node
** p
= &root
->rb_node
;
346 struct rb_node
* parent
= NULL
;
347 struct ext2_reserve_window_node
*this;
352 this = rb_entry(parent
, struct ext2_reserve_window_node
, rsv_node
);
354 if (start
< this->rsv_start
)
356 else if (start
> this->rsv_end
)
359 rsv_window_dump(root
, 1);
364 rb_link_node(node
, parent
, p
);
365 rb_insert_color(node
, root
);
369 * rsv_window_remove() -- unlink a window from the reservation rb tree
371 * @rsv: reservation window to remove
373 * Mark the block reservation window as not allocated, and unlink it
374 * from the filesystem reservation window rb tree. Must be called with
377 static void rsv_window_remove(struct super_block
*sb
,
378 struct ext2_reserve_window_node
*rsv
)
380 rsv
->rsv_start
= EXT2_RESERVE_WINDOW_NOT_ALLOCATED
;
381 rsv
->rsv_end
= EXT2_RESERVE_WINDOW_NOT_ALLOCATED
;
382 rsv
->rsv_alloc_hit
= 0;
383 rb_erase(&rsv
->rsv_node
, &EXT2_SB(sb
)->s_rsv_window_root
);
387 * rsv_is_empty() -- Check if the reservation window is allocated.
388 * @rsv: given reservation window to check
390 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
392 static inline int rsv_is_empty(struct ext2_reserve_window
*rsv
)
394 /* a valid reservation end block could not be 0 */
395 return (rsv
->_rsv_end
== EXT2_RESERVE_WINDOW_NOT_ALLOCATED
);
399 * ext2_init_block_alloc_info()
400 * @inode: file inode structure
402 * Allocate and initialize the reservation window structure, and
403 * link the window to the ext2 inode structure at last
405 * The reservation window structure is only dynamically allocated
406 * and linked to ext2 inode the first time the open file
407 * needs a new block. So, before every ext2_new_block(s) call, for
408 * regular files, we should check whether the reservation window
409 * structure exists or not. In the latter case, this function is called.
410 * Fail to do so will result in block reservation being turned off for that
413 * This function is called from ext2_get_blocks_handle(), also called
414 * when setting the reservation window size through ioctl before the file
415 * is open for write (needs block allocation).
417 * Needs truncate_mutex protection prior to calling this function.
419 void ext2_init_block_alloc_info(struct inode
*inode
)
421 struct ext2_inode_info
*ei
= EXT2_I(inode
);
422 struct ext2_block_alloc_info
*block_i
;
423 struct super_block
*sb
= inode
->i_sb
;
425 block_i
= kmalloc(sizeof(*block_i
), GFP_NOFS
);
427 struct ext2_reserve_window_node
*rsv
= &block_i
->rsv_window_node
;
429 rsv
->rsv_start
= EXT2_RESERVE_WINDOW_NOT_ALLOCATED
;
430 rsv
->rsv_end
= EXT2_RESERVE_WINDOW_NOT_ALLOCATED
;
433 * if filesystem is mounted with NORESERVATION, the goal
434 * reservation window size is set to zero to indicate
435 * block reservation is off
437 if (!test_opt(sb
, RESERVATION
))
438 rsv
->rsv_goal_size
= 0;
440 rsv
->rsv_goal_size
= EXT2_DEFAULT_RESERVE_BLOCKS
;
441 rsv
->rsv_alloc_hit
= 0;
442 block_i
->last_alloc_logical_block
= 0;
443 block_i
->last_alloc_physical_block
= 0;
445 ei
->i_block_alloc_info
= block_i
;
449 * ext2_discard_reservation()
452 * Discard(free) block reservation window on last file close, or truncate
455 * It is being called in three cases:
456 * ext2_release_file(): last writer closes the file
457 * ext2_clear_inode(): last iput(), when nobody links to this file.
458 * ext2_truncate(): when the block indirect map is about to change.
460 void ext2_discard_reservation(struct inode
*inode
)
462 struct ext2_inode_info
*ei
= EXT2_I(inode
);
463 struct ext2_block_alloc_info
*block_i
= ei
->i_block_alloc_info
;
464 struct ext2_reserve_window_node
*rsv
;
465 spinlock_t
*rsv_lock
= &EXT2_SB(inode
->i_sb
)->s_rsv_window_lock
;
470 rsv
= &block_i
->rsv_window_node
;
471 if (!rsv_is_empty(&rsv
->rsv_window
)) {
473 if (!rsv_is_empty(&rsv
->rsv_window
))
474 rsv_window_remove(inode
->i_sb
, rsv
);
475 spin_unlock(rsv_lock
);
480 * ext2_free_blocks() -- Free given blocks and update quota and i_blocks
482 * @block: start physical block to free
483 * @count: number of blocks to free
485 void ext2_free_blocks (struct inode
* inode
, unsigned long block
,
488 struct buffer_head
*bitmap_bh
= NULL
;
489 struct buffer_head
* bh2
;
490 unsigned long block_group
;
493 unsigned long overflow
;
494 struct super_block
* sb
= inode
->i_sb
;
495 struct ext2_sb_info
* sbi
= EXT2_SB(sb
);
496 struct ext2_group_desc
* desc
;
497 struct ext2_super_block
* es
= sbi
->s_es
;
498 unsigned freed
= 0, group_freed
;
500 if (block
< le32_to_cpu(es
->s_first_data_block
) ||
501 block
+ count
< block
||
502 block
+ count
> le32_to_cpu(es
->s_blocks_count
)) {
503 ext2_error (sb
, "ext2_free_blocks",
504 "Freeing blocks not in datazone - "
505 "block = %lu, count = %lu", block
, count
);
509 ext2_debug ("freeing block(s) %lu-%lu\n", block
, block
+ count
- 1);
513 block_group
= (block
- le32_to_cpu(es
->s_first_data_block
)) /
514 EXT2_BLOCKS_PER_GROUP(sb
);
515 bit
= (block
- le32_to_cpu(es
->s_first_data_block
)) %
516 EXT2_BLOCKS_PER_GROUP(sb
);
518 * Check to see if we are freeing blocks across a group
521 if (bit
+ count
> EXT2_BLOCKS_PER_GROUP(sb
)) {
522 overflow
= bit
+ count
- EXT2_BLOCKS_PER_GROUP(sb
);
526 bitmap_bh
= read_block_bitmap(sb
, block_group
);
530 desc
= ext2_get_group_desc (sb
, block_group
, &bh2
);
534 if (in_range (le32_to_cpu(desc
->bg_block_bitmap
), block
, count
) ||
535 in_range (le32_to_cpu(desc
->bg_inode_bitmap
), block
, count
) ||
536 in_range (block
, le32_to_cpu(desc
->bg_inode_table
),
537 sbi
->s_itb_per_group
) ||
538 in_range (block
+ count
- 1, le32_to_cpu(desc
->bg_inode_table
),
539 sbi
->s_itb_per_group
)) {
540 ext2_error (sb
, "ext2_free_blocks",
541 "Freeing blocks in system zones - "
542 "Block = %lu, count = %lu",
547 for (i
= 0, group_freed
= 0; i
< count
; i
++) {
548 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi
, block_group
),
549 bit
+ i
, bitmap_bh
->b_data
)) {
550 ext2_error(sb
, __func__
,
551 "bit already cleared for block %lu", block
+ i
);
557 mark_buffer_dirty(bitmap_bh
);
558 if (sb
->s_flags
& MS_SYNCHRONOUS
)
559 sync_dirty_buffer(bitmap_bh
);
561 group_adjust_blocks(sb
, block_group
, desc
, bh2
, group_freed
);
562 freed
+= group_freed
;
571 release_blocks(sb
, freed
);
572 dquot_free_block_nodirty(inode
, freed
);
576 * bitmap_search_next_usable_block()
577 * @start: the starting block (group relative) of the search
578 * @bh: bufferhead contains the block group bitmap
579 * @maxblocks: the ending block (group relative) of the reservation
581 * The bitmap search --- search forward through the actual bitmap on disk until
582 * we find a bit free.
585 bitmap_search_next_usable_block(ext2_grpblk_t start
, struct buffer_head
*bh
,
586 ext2_grpblk_t maxblocks
)
590 next
= ext2_find_next_zero_bit(bh
->b_data
, maxblocks
, start
);
591 if (next
>= maxblocks
)
597 * find_next_usable_block()
598 * @start: the starting block (group relative) to find next
599 * allocatable block in bitmap.
600 * @bh: bufferhead contains the block group bitmap
601 * @maxblocks: the ending block (group relative) for the search
603 * Find an allocatable block in a bitmap. We perform the "most
604 * appropriate allocation" algorithm of looking for a free block near
605 * the initial goal; then for a free byte somewhere in the bitmap;
606 * then for any free bit in the bitmap.
609 find_next_usable_block(int start
, struct buffer_head
*bh
, int maxblocks
)
611 ext2_grpblk_t here
, next
;
616 * The goal was occupied; search forward for a free
617 * block within the next XX blocks.
619 * end_goal is more or less random, but it has to be
620 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
621 * next 64-bit boundary is simple..
623 ext2_grpblk_t end_goal
= (start
+ 63) & ~63;
624 if (end_goal
> maxblocks
)
625 end_goal
= maxblocks
;
626 here
= ext2_find_next_zero_bit(bh
->b_data
, end_goal
, start
);
629 ext2_debug("Bit not found near goal\n");
636 p
= ((char *)bh
->b_data
) + (here
>> 3);
637 r
= memscan(p
, 0, ((maxblocks
+ 7) >> 3) - (here
>> 3));
638 next
= (r
- ((char *)bh
->b_data
)) << 3;
640 if (next
< maxblocks
&& next
>= here
)
643 here
= bitmap_search_next_usable_block(here
, bh
, maxblocks
);
648 * ext2_try_to_allocate()
650 * @group: given allocation block group
651 * @bitmap_bh: bufferhead holds the block bitmap
652 * @grp_goal: given target block within the group
653 * @count: target number of blocks to allocate
654 * @my_rsv: reservation window
656 * Attempt to allocate blocks within a give range. Set the range of allocation
657 * first, then find the first free bit(s) from the bitmap (within the range),
658 * and at last, allocate the blocks by claiming the found free bit as allocated.
660 * To set the range of this allocation:
661 * if there is a reservation window, only try to allocate block(s)
662 * from the file's own reservation window;
663 * Otherwise, the allocation range starts from the give goal block,
664 * ends at the block group's last block.
666 * If we failed to allocate the desired block then we may end up crossing to a
670 ext2_try_to_allocate(struct super_block
*sb
, int group
,
671 struct buffer_head
*bitmap_bh
, ext2_grpblk_t grp_goal
,
672 unsigned long *count
,
673 struct ext2_reserve_window
*my_rsv
)
675 ext2_fsblk_t group_first_block
;
676 ext2_grpblk_t start
, end
;
677 unsigned long num
= 0;
679 /* we do allocation within the reservation window if we have a window */
681 group_first_block
= ext2_group_first_block_no(sb
, group
);
682 if (my_rsv
->_rsv_start
>= group_first_block
)
683 start
= my_rsv
->_rsv_start
- group_first_block
;
685 /* reservation window cross group boundary */
687 end
= my_rsv
->_rsv_end
- group_first_block
+ 1;
688 if (end
> EXT2_BLOCKS_PER_GROUP(sb
))
689 /* reservation window crosses group boundary */
690 end
= EXT2_BLOCKS_PER_GROUP(sb
);
691 if ((start
<= grp_goal
) && (grp_goal
< end
))
700 end
= EXT2_BLOCKS_PER_GROUP(sb
);
703 BUG_ON(start
> EXT2_BLOCKS_PER_GROUP(sb
));
707 grp_goal
= find_next_usable_block(start
, bitmap_bh
, end
);
713 for (i
= 0; i
< 7 && grp_goal
> start
&&
714 !ext2_test_bit(grp_goal
- 1,
722 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb
), group
), grp_goal
,
723 bitmap_bh
->b_data
)) {
725 * The block was allocated by another thread, or it was
726 * allocated and then freed by another thread
736 while (num
< *count
&& grp_goal
< end
737 && !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb
), group
),
738 grp_goal
, bitmap_bh
->b_data
)) {
743 return grp_goal
- num
;
750 * find_next_reservable_window():
751 * find a reservable space within the given range.
752 * It does not allocate the reservation window for now:
753 * alloc_new_reservation() will do the work later.
755 * @search_head: the head of the searching list;
756 * This is not necessarily the list head of the whole filesystem
758 * We have both head and start_block to assist the search
759 * for the reservable space. The list starts from head,
760 * but we will shift to the place where start_block is,
761 * then start from there, when looking for a reservable space.
763 * @size: the target new reservation window size
765 * @group_first_block: the first block we consider to start
766 * the real search from
769 * the maximum block number that our goal reservable space
770 * could start from. This is normally the last block in this
771 * group. The search will end when we found the start of next
772 * possible reservable space is out of this boundary.
773 * This could handle the cross boundary reservation window
776 * basically we search from the given range, rather than the whole
777 * reservation double linked list, (start_block, last_block)
778 * to find a free region that is of my size and has not
782 static int find_next_reservable_window(
783 struct ext2_reserve_window_node
*search_head
,
784 struct ext2_reserve_window_node
*my_rsv
,
785 struct super_block
* sb
,
786 ext2_fsblk_t start_block
,
787 ext2_fsblk_t last_block
)
789 struct rb_node
*next
;
790 struct ext2_reserve_window_node
*rsv
, *prev
;
792 int size
= my_rsv
->rsv_goal_size
;
794 /* TODO: make the start of the reservation window byte-aligned */
795 /* cur = *start_block & ~7;*/
802 if (cur
<= rsv
->rsv_end
)
803 cur
= rsv
->rsv_end
+ 1;
806 * in the case we could not find a reservable space
807 * that is what is expected, during the re-search, we could
808 * remember what's the largest reservable space we could have
809 * and return that one.
811 * For now it will fail if we could not find the reservable
812 * space with expected-size (or more)...
814 if (cur
> last_block
)
815 return -1; /* fail */
818 next
= rb_next(&rsv
->rsv_node
);
819 rsv
= rb_entry(next
,struct ext2_reserve_window_node
,rsv_node
);
822 * Reached the last reservation, we can just append to the
828 if (cur
+ size
<= rsv
->rsv_start
) {
830 * Found a reserveable space big enough. We could
831 * have a reservation across the group boundary here
837 * we come here either :
838 * when we reach the end of the whole list,
839 * and there is empty reservable space after last entry in the list.
840 * append it to the end of the list.
842 * or we found one reservable space in the middle of the list,
843 * return the reservation window that we could append to.
847 if ((prev
!= my_rsv
) && (!rsv_is_empty(&my_rsv
->rsv_window
)))
848 rsv_window_remove(sb
, my_rsv
);
851 * Let's book the whole available window for now. We will check the
852 * disk bitmap later and then, if there are free blocks then we adjust
853 * the window size if it's larger than requested.
854 * Otherwise, we will remove this node from the tree next time
855 * call find_next_reservable_window.
857 my_rsv
->rsv_start
= cur
;
858 my_rsv
->rsv_end
= cur
+ size
- 1;
859 my_rsv
->rsv_alloc_hit
= 0;
862 ext2_rsv_window_add(sb
, my_rsv
);
868 * alloc_new_reservation()--allocate a new reservation window
870 * To make a new reservation, we search part of the filesystem
871 * reservation list (the list that inside the group). We try to
872 * allocate a new reservation window near the allocation goal,
873 * or the beginning of the group, if there is no goal.
875 * We first find a reservable space after the goal, then from
876 * there, we check the bitmap for the first free block after
877 * it. If there is no free block until the end of group, then the
878 * whole group is full, we failed. Otherwise, check if the free
879 * block is inside the expected reservable space, if so, we
881 * If the first free block is outside the reservable space, then
882 * start from the first free block, we search for next available
885 * on succeed, a new reservation will be found and inserted into the list
886 * It contains at least one free block, and it does not overlap with other
887 * reservation windows.
889 * failed: we failed to find a reservation window in this group
891 * @rsv: the reservation
893 * @grp_goal: The goal (group-relative). It is where the search for a
894 * free reservable space should start from.
895 * if we have a goal(goal >0 ), then start from there,
896 * no goal(goal = -1), we start from the first block
899 * @sb: the super block
900 * @group: the group we are trying to allocate in
901 * @bitmap_bh: the block group block bitmap
904 static int alloc_new_reservation(struct ext2_reserve_window_node
*my_rsv
,
905 ext2_grpblk_t grp_goal
, struct super_block
*sb
,
906 unsigned int group
, struct buffer_head
*bitmap_bh
)
908 struct ext2_reserve_window_node
*search_head
;
909 ext2_fsblk_t group_first_block
, group_end_block
, start_block
;
910 ext2_grpblk_t first_free_block
;
911 struct rb_root
*fs_rsv_root
= &EXT2_SB(sb
)->s_rsv_window_root
;
914 spinlock_t
*rsv_lock
= &EXT2_SB(sb
)->s_rsv_window_lock
;
916 group_first_block
= ext2_group_first_block_no(sb
, group
);
917 group_end_block
= group_first_block
+ (EXT2_BLOCKS_PER_GROUP(sb
) - 1);
920 start_block
= group_first_block
;
922 start_block
= grp_goal
+ group_first_block
;
924 size
= my_rsv
->rsv_goal_size
;
926 if (!rsv_is_empty(&my_rsv
->rsv_window
)) {
928 * if the old reservation is cross group boundary
929 * and if the goal is inside the old reservation window,
930 * we will come here when we just failed to allocate from
931 * the first part of the window. We still have another part
932 * that belongs to the next group. In this case, there is no
933 * point to discard our window and try to allocate a new one
934 * in this group(which will fail). we should
935 * keep the reservation window, just simply move on.
937 * Maybe we could shift the start block of the reservation
938 * window to the first block of next group.
941 if ((my_rsv
->rsv_start
<= group_end_block
) &&
942 (my_rsv
->rsv_end
> group_end_block
) &&
943 (start_block
>= my_rsv
->rsv_start
))
946 if ((my_rsv
->rsv_alloc_hit
>
947 (my_rsv
->rsv_end
- my_rsv
->rsv_start
+ 1) / 2)) {
949 * if the previously allocation hit ratio is
950 * greater than 1/2, then we double the size of
951 * the reservation window the next time,
952 * otherwise we keep the same size window
955 if (size
> EXT2_MAX_RESERVE_BLOCKS
)
956 size
= EXT2_MAX_RESERVE_BLOCKS
;
957 my_rsv
->rsv_goal_size
= size
;
963 * shift the search start to the window near the goal block
965 search_head
= search_reserve_window(fs_rsv_root
, start_block
);
968 * find_next_reservable_window() simply finds a reservable window
969 * inside the given range(start_block, group_end_block).
971 * To make sure the reservation window has a free bit inside it, we
972 * need to check the bitmap after we found a reservable window.
975 ret
= find_next_reservable_window(search_head
, my_rsv
, sb
,
976 start_block
, group_end_block
);
979 if (!rsv_is_empty(&my_rsv
->rsv_window
))
980 rsv_window_remove(sb
, my_rsv
);
981 spin_unlock(rsv_lock
);
986 * On success, find_next_reservable_window() returns the
987 * reservation window where there is a reservable space after it.
988 * Before we reserve this reservable space, we need
989 * to make sure there is at least a free block inside this region.
991 * Search the first free bit on the block bitmap. Search starts from
992 * the start block of the reservable space we just found.
994 spin_unlock(rsv_lock
);
995 first_free_block
= bitmap_search_next_usable_block(
996 my_rsv
->rsv_start
- group_first_block
,
997 bitmap_bh
, group_end_block
- group_first_block
+ 1);
999 if (first_free_block
< 0) {
1001 * no free block left on the bitmap, no point
1002 * to reserve the space. return failed.
1004 spin_lock(rsv_lock
);
1005 if (!rsv_is_empty(&my_rsv
->rsv_window
))
1006 rsv_window_remove(sb
, my_rsv
);
1007 spin_unlock(rsv_lock
);
1008 return -1; /* failed */
1011 start_block
= first_free_block
+ group_first_block
;
1013 * check if the first free block is within the
1014 * free space we just reserved
1016 if (start_block
>= my_rsv
->rsv_start
&& start_block
<= my_rsv
->rsv_end
)
1017 return 0; /* success */
1019 * if the first free bit we found is out of the reservable space
1020 * continue search for next reservable space,
1021 * start from where the free block is,
1022 * we also shift the list head to where we stopped last time
1024 search_head
= my_rsv
;
1025 spin_lock(rsv_lock
);
1030 * try_to_extend_reservation()
1031 * @my_rsv: given reservation window
1033 * @size: the delta to extend
1035 * Attempt to expand the reservation window large enough to have
1036 * required number of free blocks
1038 * Since ext2_try_to_allocate() will always allocate blocks within
1039 * the reservation window range, if the window size is too small,
1040 * multiple blocks allocation has to stop at the end of the reservation
1041 * window. To make this more efficient, given the total number of
1042 * blocks needed and the current size of the window, we try to
1043 * expand the reservation window size if necessary on a best-effort
1044 * basis before ext2_new_blocks() tries to allocate blocks.
1046 static void try_to_extend_reservation(struct ext2_reserve_window_node
*my_rsv
,
1047 struct super_block
*sb
, int size
)
1049 struct ext2_reserve_window_node
*next_rsv
;
1050 struct rb_node
*next
;
1051 spinlock_t
*rsv_lock
= &EXT2_SB(sb
)->s_rsv_window_lock
;
1053 if (!spin_trylock(rsv_lock
))
1056 next
= rb_next(&my_rsv
->rsv_node
);
1059 my_rsv
->rsv_end
+= size
;
1061 next_rsv
= rb_entry(next
, struct ext2_reserve_window_node
, rsv_node
);
1063 if ((next_rsv
->rsv_start
- my_rsv
->rsv_end
- 1) >= size
)
1064 my_rsv
->rsv_end
+= size
;
1066 my_rsv
->rsv_end
= next_rsv
->rsv_start
- 1;
1068 spin_unlock(rsv_lock
);
1072 * ext2_try_to_allocate_with_rsv()
1074 * @group: given allocation block group
1075 * @bitmap_bh: bufferhead holds the block bitmap
1076 * @grp_goal: given target block within the group
1077 * @count: target number of blocks to allocate
1078 * @my_rsv: reservation window
1080 * This is the main function used to allocate a new block and its reservation
1083 * Each time when a new block allocation is need, first try to allocate from
1084 * its own reservation. If it does not have a reservation window, instead of
1085 * looking for a free bit on bitmap first, then look up the reservation list to
1086 * see if it is inside somebody else's reservation window, we try to allocate a
1087 * reservation window for it starting from the goal first. Then do the block
1088 * allocation within the reservation window.
1090 * This will avoid keeping on searching the reservation list again and
1091 * again when somebody is looking for a free block (without
1092 * reservation), and there are lots of free blocks, but they are all
1095 * We use a red-black tree for the per-filesystem reservation list.
1097 static ext2_grpblk_t
1098 ext2_try_to_allocate_with_rsv(struct super_block
*sb
, unsigned int group
,
1099 struct buffer_head
*bitmap_bh
, ext2_grpblk_t grp_goal
,
1100 struct ext2_reserve_window_node
* my_rsv
,
1101 unsigned long *count
)
1103 ext2_fsblk_t group_first_block
, group_last_block
;
1104 ext2_grpblk_t ret
= 0;
1105 unsigned long num
= *count
;
1108 * we don't deal with reservation when
1109 * filesystem is mounted without reservation
1110 * or the file is not a regular file
1111 * or last attempt to allocate a block with reservation turned on failed
1113 if (my_rsv
== NULL
) {
1114 return ext2_try_to_allocate(sb
, group
, bitmap_bh
,
1115 grp_goal
, count
, NULL
);
1118 * grp_goal is a group relative block number (if there is a goal)
1119 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1120 * first block is a filesystem wide block number
1121 * first block is the block number of the first block in this group
1123 group_first_block
= ext2_group_first_block_no(sb
, group
);
1124 group_last_block
= group_first_block
+ (EXT2_BLOCKS_PER_GROUP(sb
) - 1);
1127 * Basically we will allocate a new block from inode's reservation
1130 * We need to allocate a new reservation window, if:
1131 * a) inode does not have a reservation window; or
1132 * b) last attempt to allocate a block from existing reservation
1134 * c) we come here with a goal and with a reservation window
1136 * We do not need to allocate a new reservation window if we come here
1137 * at the beginning with a goal and the goal is inside the window, or
1138 * we don't have a goal but already have a reservation window.
1139 * then we could go to allocate from the reservation window directly.
1142 if (rsv_is_empty(&my_rsv
->rsv_window
) || (ret
< 0) ||
1143 !goal_in_my_reservation(&my_rsv
->rsv_window
,
1144 grp_goal
, group
, sb
)) {
1145 if (my_rsv
->rsv_goal_size
< *count
)
1146 my_rsv
->rsv_goal_size
= *count
;
1147 ret
= alloc_new_reservation(my_rsv
, grp_goal
, sb
,
1152 if (!goal_in_my_reservation(&my_rsv
->rsv_window
,
1153 grp_goal
, group
, sb
))
1155 } else if (grp_goal
>= 0) {
1156 int curr
= my_rsv
->rsv_end
-
1157 (grp_goal
+ group_first_block
) + 1;
1160 try_to_extend_reservation(my_rsv
, sb
,
1164 if ((my_rsv
->rsv_start
> group_last_block
) ||
1165 (my_rsv
->rsv_end
< group_first_block
)) {
1166 rsv_window_dump(&EXT2_SB(sb
)->s_rsv_window_root
, 1);
1169 ret
= ext2_try_to_allocate(sb
, group
, bitmap_bh
, grp_goal
,
1170 &num
, &my_rsv
->rsv_window
);
1172 my_rsv
->rsv_alloc_hit
+= num
;
1174 break; /* succeed */
1182 * ext2_has_free_blocks()
1183 * @sbi: in-core super block structure.
1185 * Check if filesystem has at least 1 free block available for allocation.
1187 static int ext2_has_free_blocks(struct ext2_sb_info
*sbi
)
1189 ext2_fsblk_t free_blocks
, root_blocks
;
1191 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
1192 root_blocks
= le32_to_cpu(sbi
->s_es
->s_r_blocks_count
);
1193 if (free_blocks
< root_blocks
+ 1 && !capable(CAP_SYS_RESOURCE
) &&
1194 !uid_eq(sbi
->s_resuid
, current_fsuid()) &&
1195 (gid_eq(sbi
->s_resgid
, GLOBAL_ROOT_GID
) ||
1196 !in_group_p (sbi
->s_resgid
))) {
1203 * ext2_new_blocks() -- core block(s) allocation function
1204 * @inode: file inode
1205 * @goal: given target block(filesystem wide)
1206 * @count: target number of blocks to allocate
1209 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
1210 * free, or there is a free block within 32 blocks of the goal, that block
1211 * is allocated. Otherwise a forward search is made for a free block; within
1212 * each block group the search first looks for an entire free byte in the block
1213 * bitmap, and then for any free bit if that fails.
1214 * This function also updates quota and i_blocks field.
1216 ext2_fsblk_t
ext2_new_blocks(struct inode
*inode
, ext2_fsblk_t goal
,
1217 unsigned long *count
, int *errp
)
1219 struct buffer_head
*bitmap_bh
= NULL
;
1220 struct buffer_head
*gdp_bh
;
1223 ext2_grpblk_t grp_target_blk
; /* blockgroup relative goal block */
1224 ext2_grpblk_t grp_alloc_blk
; /* blockgroup-relative allocated block*/
1225 ext2_fsblk_t ret_block
; /* filesyetem-wide allocated block */
1226 int bgi
; /* blockgroup iteration index */
1227 int performed_allocation
= 0;
1228 ext2_grpblk_t free_blocks
; /* number of free blocks in a group */
1229 struct super_block
*sb
;
1230 struct ext2_group_desc
*gdp
;
1231 struct ext2_super_block
*es
;
1232 struct ext2_sb_info
*sbi
;
1233 struct ext2_reserve_window_node
*my_rsv
= NULL
;
1234 struct ext2_block_alloc_info
*block_i
;
1235 unsigned short windowsz
= 0;
1236 unsigned long ngroups
;
1237 unsigned long num
= *count
;
1243 printk("ext2_new_blocks: nonexistent device");
1248 * Check quota for allocation of this block.
1250 ret
= dquot_alloc_block(inode
, num
);
1257 es
= EXT2_SB(sb
)->s_es
;
1258 ext2_debug("goal=%lu.\n", goal
);
1260 * Allocate a block from reservation only when
1261 * filesystem is mounted with reservation(default,-o reservation), and
1262 * it's a regular file, and
1263 * the desired window size is greater than 0 (One could use ioctl
1264 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1265 * reservation on that particular file)
1267 block_i
= EXT2_I(inode
)->i_block_alloc_info
;
1269 windowsz
= block_i
->rsv_window_node
.rsv_goal_size
;
1271 my_rsv
= &block_i
->rsv_window_node
;
1274 if (!ext2_has_free_blocks(sbi
)) {
1280 * First, test whether the goal block is free.
1282 if (goal
< le32_to_cpu(es
->s_first_data_block
) ||
1283 goal
>= le32_to_cpu(es
->s_blocks_count
))
1284 goal
= le32_to_cpu(es
->s_first_data_block
);
1285 group_no
= (goal
- le32_to_cpu(es
->s_first_data_block
)) /
1286 EXT2_BLOCKS_PER_GROUP(sb
);
1287 goal_group
= group_no
;
1289 gdp
= ext2_get_group_desc(sb
, group_no
, &gdp_bh
);
1293 free_blocks
= le16_to_cpu(gdp
->bg_free_blocks_count
);
1295 * if there is not enough free blocks to make a new resevation
1296 * turn off reservation for this allocation
1298 if (my_rsv
&& (free_blocks
< windowsz
)
1299 && (free_blocks
> 0)
1300 && (rsv_is_empty(&my_rsv
->rsv_window
)))
1303 if (free_blocks
> 0) {
1304 grp_target_blk
= ((goal
- le32_to_cpu(es
->s_first_data_block
)) %
1305 EXT2_BLOCKS_PER_GROUP(sb
));
1306 bitmap_bh
= read_block_bitmap(sb
, group_no
);
1309 grp_alloc_blk
= ext2_try_to_allocate_with_rsv(sb
, group_no
,
1310 bitmap_bh
, grp_target_blk
,
1312 if (grp_alloc_blk
>= 0)
1316 ngroups
= EXT2_SB(sb
)->s_groups_count
;
1320 * Now search the rest of the groups. We assume that
1321 * group_no and gdp correctly point to the last group visited.
1323 for (bgi
= 0; bgi
< ngroups
; bgi
++) {
1325 if (group_no
>= ngroups
)
1327 gdp
= ext2_get_group_desc(sb
, group_no
, &gdp_bh
);
1331 free_blocks
= le16_to_cpu(gdp
->bg_free_blocks_count
);
1333 * skip this group (and avoid loading bitmap) if there
1334 * are no free blocks
1339 * skip this group if the number of
1340 * free blocks is less than half of the reservation
1343 if (my_rsv
&& (free_blocks
<= (windowsz
/2)))
1347 bitmap_bh
= read_block_bitmap(sb
, group_no
);
1351 * try to allocate block(s) from this group, without a goal(-1).
1353 grp_alloc_blk
= ext2_try_to_allocate_with_rsv(sb
, group_no
,
1354 bitmap_bh
, -1, my_rsv
, &num
);
1355 if (grp_alloc_blk
>= 0)
1359 * We may end up a bogus earlier ENOSPC error due to
1360 * filesystem is "full" of reservations, but
1361 * there maybe indeed free blocks available on disk
1362 * In this case, we just forget about the reservations
1363 * just do block allocation as without reservations.
1368 group_no
= goal_group
;
1371 /* No space left on the device */
1377 ext2_debug("using block group %d(%d)\n",
1378 group_no
, gdp
->bg_free_blocks_count
);
1380 ret_block
= grp_alloc_blk
+ ext2_group_first_block_no(sb
, group_no
);
1382 if (in_range(le32_to_cpu(gdp
->bg_block_bitmap
), ret_block
, num
) ||
1383 in_range(le32_to_cpu(gdp
->bg_inode_bitmap
), ret_block
, num
) ||
1384 in_range(ret_block
, le32_to_cpu(gdp
->bg_inode_table
),
1385 EXT2_SB(sb
)->s_itb_per_group
) ||
1386 in_range(ret_block
+ num
- 1, le32_to_cpu(gdp
->bg_inode_table
),
1387 EXT2_SB(sb
)->s_itb_per_group
)) {
1388 ext2_error(sb
, "ext2_new_blocks",
1389 "Allocating block in system zone - "
1390 "blocks from "E2FSBLK
", length %lu",
1393 * ext2_try_to_allocate marked the blocks we allocated as in
1394 * use. So we may want to selectively mark some of the blocks
1400 performed_allocation
= 1;
1402 if (ret_block
+ num
- 1 >= le32_to_cpu(es
->s_blocks_count
)) {
1403 ext2_error(sb
, "ext2_new_blocks",
1404 "block("E2FSBLK
") >= blocks count(%d) - "
1405 "block_group = %d, es == %p ", ret_block
,
1406 le32_to_cpu(es
->s_blocks_count
), group_no
, es
);
1410 group_adjust_blocks(sb
, group_no
, gdp
, gdp_bh
, -num
);
1411 percpu_counter_sub(&sbi
->s_freeblocks_counter
, num
);
1413 mark_buffer_dirty(bitmap_bh
);
1414 if (sb
->s_flags
& MS_SYNCHRONOUS
)
1415 sync_dirty_buffer(bitmap_bh
);
1419 dquot_free_block_nodirty(inode
, *count
-num
);
1420 mark_inode_dirty(inode
);
1428 * Undo the block allocation
1430 if (!performed_allocation
) {
1431 dquot_free_block_nodirty(inode
, *count
);
1432 mark_inode_dirty(inode
);
1438 ext2_fsblk_t
ext2_new_block(struct inode
*inode
, unsigned long goal
, int *errp
)
1440 unsigned long count
= 1;
1442 return ext2_new_blocks(inode
, goal
, &count
, errp
);
1447 unsigned long ext2_count_free(struct buffer_head
*map
, unsigned int numchars
)
1449 return numchars
* BITS_PER_BYTE
- memweight(map
->b_data
, numchars
);
1452 #endif /* EXT2FS_DEBUG */
1454 unsigned long ext2_count_free_blocks (struct super_block
* sb
)
1456 struct ext2_group_desc
* desc
;
1457 unsigned long desc_count
= 0;
1460 unsigned long bitmap_count
, x
;
1461 struct ext2_super_block
*es
;
1463 es
= EXT2_SB(sb
)->s_es
;
1467 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
1468 struct buffer_head
*bitmap_bh
;
1469 desc
= ext2_get_group_desc (sb
, i
, NULL
);
1472 desc_count
+= le16_to_cpu(desc
->bg_free_blocks_count
);
1473 bitmap_bh
= read_block_bitmap(sb
, i
);
1477 x
= ext2_count_free(bitmap_bh
, sb
->s_blocksize
);
1478 printk ("group %d: stored = %d, counted = %lu\n",
1479 i
, le16_to_cpu(desc
->bg_free_blocks_count
), x
);
1483 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1484 (long)le32_to_cpu(es
->s_free_blocks_count
),
1485 desc_count
, bitmap_count
);
1486 return bitmap_count
;
1488 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
1489 desc
= ext2_get_group_desc (sb
, i
, NULL
);
1492 desc_count
+= le16_to_cpu(desc
->bg_free_blocks_count
);
1498 static inline int test_root(int a
, int b
)
1507 static int ext2_group_sparse(int group
)
1511 return (test_root(group
, 3) || test_root(group
, 5) ||
1512 test_root(group
, 7));
1516 * ext2_bg_has_super - number of blocks used by the superblock in group
1517 * @sb: superblock for filesystem
1518 * @group: group number to check
1520 * Return the number of blocks used by the superblock (primary or backup)
1521 * in this group. Currently this will be only 0 or 1.
1523 int ext2_bg_has_super(struct super_block
*sb
, int group
)
1525 if (EXT2_HAS_RO_COMPAT_FEATURE(sb
,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER
)&&
1526 !ext2_group_sparse(group
))
1532 * ext2_bg_num_gdb - number of blocks used by the group table in group
1533 * @sb: superblock for filesystem
1534 * @group: group number to check
1536 * Return the number of blocks used by the group descriptor table
1537 * (primary or backup) in this group. In the future there may be a
1538 * different number of descriptor blocks in each group.
1540 unsigned long ext2_bg_num_gdb(struct super_block
*sb
, int group
)
1542 return ext2_bg_has_super(sb
, group
) ? EXT2_SB(sb
)->s_gdb_count
: 0;