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/sched.h>
17 #include <linux/buffer_head.h>
18 #include <linux/capability.h>
21 * balloc.c contains the blocks allocation and deallocation routines
25 * The free blocks are managed by bitmaps. A file system contains several
26 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
27 * block for inodes, N blocks for the inode table and data blocks.
29 * The file system contains group descriptors which are located after the
30 * super block. Each descriptor contains the number of the bitmap block and
31 * the free blocks count in the block. The descriptors are loaded in memory
32 * when a file system is mounted (see ext2_fill_super).
36 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
38 struct ext2_group_desc
* ext2_get_group_desc(struct super_block
* sb
,
39 unsigned int block_group
,
40 struct buffer_head
** bh
)
42 unsigned long group_desc
;
44 struct ext2_group_desc
* desc
;
45 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
47 if (block_group
>= sbi
->s_groups_count
) {
48 ext2_error (sb
, "ext2_get_group_desc",
49 "block_group >= groups_count - "
50 "block_group = %d, groups_count = %lu",
51 block_group
, sbi
->s_groups_count
);
56 group_desc
= block_group
>> EXT2_DESC_PER_BLOCK_BITS(sb
);
57 offset
= block_group
& (EXT2_DESC_PER_BLOCK(sb
) - 1);
58 if (!sbi
->s_group_desc
[group_desc
]) {
59 ext2_error (sb
, "ext2_get_group_desc",
60 "Group descriptor not loaded - "
61 "block_group = %d, group_desc = %lu, desc = %lu",
62 block_group
, group_desc
, offset
);
66 desc
= (struct ext2_group_desc
*) sbi
->s_group_desc
[group_desc
]->b_data
;
68 *bh
= sbi
->s_group_desc
[group_desc
];
73 * Read the bitmap for a given block_group, reading into the specified
74 * slot in the superblock's bitmap cache.
76 * Return buffer_head on success or NULL in case of failure.
78 static struct buffer_head
*
79 read_block_bitmap(struct super_block
*sb
, unsigned int block_group
)
81 struct ext2_group_desc
* desc
;
82 struct buffer_head
* bh
= NULL
;
84 desc
= ext2_get_group_desc (sb
, block_group
, NULL
);
87 bh
= sb_bread(sb
, le32_to_cpu(desc
->bg_block_bitmap
));
89 ext2_error (sb
, "read_block_bitmap",
90 "Cannot read block bitmap - "
91 "block_group = %d, block_bitmap = %u",
92 block_group
, le32_to_cpu(desc
->bg_block_bitmap
));
97 static void release_blocks(struct super_block
*sb
, int count
)
100 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
102 percpu_counter_add(&sbi
->s_freeblocks_counter
, count
);
107 static void group_adjust_blocks(struct super_block
*sb
, int group_no
,
108 struct ext2_group_desc
*desc
, struct buffer_head
*bh
, int count
)
111 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
112 unsigned free_blocks
;
114 spin_lock(sb_bgl_lock(sbi
, group_no
));
115 free_blocks
= le16_to_cpu(desc
->bg_free_blocks_count
);
116 desc
->bg_free_blocks_count
= cpu_to_le16(free_blocks
+ count
);
117 spin_unlock(sb_bgl_lock(sbi
, group_no
));
119 mark_buffer_dirty(bh
);
124 * The reservation window structure operations
125 * --------------------------------------------
126 * Operations include:
127 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
129 * We use a red-black tree to represent per-filesystem reservation
135 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
136 * @rb_root: root of per-filesystem reservation rb tree
137 * @verbose: verbose mode
138 * @fn: function which wishes to dump the reservation map
140 * If verbose is turned on, it will print the whole block reservation
141 * windows(start, end). Otherwise, it will only print out the "bad" windows,
142 * those windows that overlap with their immediate neighbors.
145 static void __rsv_window_dump(struct rb_root
*root
, int verbose
,
149 struct ext2_reserve_window_node
*rsv
, *prev
;
157 printk("Block Allocation Reservation Windows Map (%s):\n", fn
);
159 rsv
= rb_entry(n
, struct ext2_reserve_window_node
, rsv_node
);
161 printk("reservation window 0x%p "
162 "start: %lu, end: %lu\n",
163 rsv
, rsv
->rsv_start
, rsv
->rsv_end
);
164 if (rsv
->rsv_start
&& rsv
->rsv_start
>= rsv
->rsv_end
) {
165 printk("Bad reservation %p (start >= end)\n",
169 if (prev
&& prev
->rsv_end
>= rsv
->rsv_start
) {
170 printk("Bad reservation %p (prev->end >= start)\n",
176 printk("Restarting reservation walk in verbose mode\n");
184 printk("Window map complete.\n");
188 #define rsv_window_dump(root, verbose) \
189 __rsv_window_dump((root), (verbose), __FUNCTION__)
191 #define rsv_window_dump(root, verbose) do {} while (0)
195 * goal_in_my_reservation()
196 * @rsv: inode's reservation window
197 * @grp_goal: given goal block relative to the allocation block group
198 * @group: the current allocation block group
199 * @sb: filesystem super block
201 * Test if the given goal block (group relative) is within the file's
202 * own block reservation window range.
204 * If the reservation window is outside the goal allocation group, return 0;
205 * grp_goal (given goal block) could be -1, which means no specific
206 * goal block. In this case, always return 1.
207 * If the goal block is within the reservation window, return 1;
208 * otherwise, return 0;
211 goal_in_my_reservation(struct ext2_reserve_window
*rsv
, ext2_grpblk_t grp_goal
,
212 unsigned int group
, struct super_block
* sb
)
214 ext2_fsblk_t group_first_block
, group_last_block
;
216 group_first_block
= ext2_group_first_block_no(sb
, group
);
217 group_last_block
= group_first_block
+ EXT2_BLOCKS_PER_GROUP(sb
) - 1;
219 if ((rsv
->_rsv_start
> group_last_block
) ||
220 (rsv
->_rsv_end
< group_first_block
))
222 if ((grp_goal
>= 0) && ((grp_goal
+ group_first_block
< rsv
->_rsv_start
)
223 || (grp_goal
+ group_first_block
> rsv
->_rsv_end
)))
229 * search_reserve_window()
230 * @rb_root: root of reservation tree
231 * @goal: target allocation block
233 * Find the reserved window which includes the goal, or the previous one
234 * if the goal is not in any window.
235 * Returns NULL if there are no windows or if all windows start after the goal.
237 static struct ext2_reserve_window_node
*
238 search_reserve_window(struct rb_root
*root
, ext2_fsblk_t goal
)
240 struct rb_node
*n
= root
->rb_node
;
241 struct ext2_reserve_window_node
*rsv
;
247 rsv
= rb_entry(n
, struct ext2_reserve_window_node
, rsv_node
);
249 if (goal
< rsv
->rsv_start
)
251 else if (goal
> rsv
->rsv_end
)
257 * We've fallen off the end of the tree: the goal wasn't inside
258 * any particular node. OK, the previous node must be to one
259 * side of the interval containing the goal. If it's the RHS,
260 * we need to back up one.
262 if (rsv
->rsv_start
> goal
) {
263 n
= rb_prev(&rsv
->rsv_node
);
264 rsv
= rb_entry(n
, struct ext2_reserve_window_node
, rsv_node
);
270 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
272 * @rsv: reservation window to add
274 * Must be called with rsv_lock held.
276 void ext2_rsv_window_add(struct super_block
*sb
,
277 struct ext2_reserve_window_node
*rsv
)
279 struct rb_root
*root
= &EXT2_SB(sb
)->s_rsv_window_root
;
280 struct rb_node
*node
= &rsv
->rsv_node
;
281 ext2_fsblk_t start
= rsv
->rsv_start
;
283 struct rb_node
** p
= &root
->rb_node
;
284 struct rb_node
* parent
= NULL
;
285 struct ext2_reserve_window_node
*this;
290 this = rb_entry(parent
, struct ext2_reserve_window_node
, rsv_node
);
292 if (start
< this->rsv_start
)
294 else if (start
> this->rsv_end
)
297 rsv_window_dump(root
, 1);
302 rb_link_node(node
, parent
, p
);
303 rb_insert_color(node
, root
);
307 * rsv_window_remove() -- unlink a window from the reservation rb tree
309 * @rsv: reservation window to remove
311 * Mark the block reservation window as not allocated, and unlink it
312 * from the filesystem reservation window rb tree. Must be called with
315 static void rsv_window_remove(struct super_block
*sb
,
316 struct ext2_reserve_window_node
*rsv
)
318 rsv
->rsv_start
= EXT2_RESERVE_WINDOW_NOT_ALLOCATED
;
319 rsv
->rsv_end
= EXT2_RESERVE_WINDOW_NOT_ALLOCATED
;
320 rsv
->rsv_alloc_hit
= 0;
321 rb_erase(&rsv
->rsv_node
, &EXT2_SB(sb
)->s_rsv_window_root
);
325 * rsv_is_empty() -- Check if the reservation window is allocated.
326 * @rsv: given reservation window to check
328 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
330 static inline int rsv_is_empty(struct ext2_reserve_window
*rsv
)
332 /* a valid reservation end block could not be 0 */
333 return (rsv
->_rsv_end
== EXT2_RESERVE_WINDOW_NOT_ALLOCATED
);
337 * ext2_init_block_alloc_info()
338 * @inode: file inode structure
340 * Allocate and initialize the reservation window structure, and
341 * link the window to the ext2 inode structure at last
343 * The reservation window structure is only dynamically allocated
344 * and linked to ext2 inode the first time the open file
345 * needs a new block. So, before every ext2_new_block(s) call, for
346 * regular files, we should check whether the reservation window
347 * structure exists or not. In the latter case, this function is called.
348 * Fail to do so will result in block reservation being turned off for that
351 * This function is called from ext2_get_blocks_handle(), also called
352 * when setting the reservation window size through ioctl before the file
353 * is open for write (needs block allocation).
355 * Needs truncate_mutex protection prior to calling this function.
357 void ext2_init_block_alloc_info(struct inode
*inode
)
359 struct ext2_inode_info
*ei
= EXT2_I(inode
);
360 struct ext2_block_alloc_info
*block_i
= ei
->i_block_alloc_info
;
361 struct super_block
*sb
= inode
->i_sb
;
363 block_i
= kmalloc(sizeof(*block_i
), GFP_NOFS
);
365 struct ext2_reserve_window_node
*rsv
= &block_i
->rsv_window_node
;
367 rsv
->rsv_start
= EXT2_RESERVE_WINDOW_NOT_ALLOCATED
;
368 rsv
->rsv_end
= EXT2_RESERVE_WINDOW_NOT_ALLOCATED
;
371 * if filesystem is mounted with NORESERVATION, the goal
372 * reservation window size is set to zero to indicate
373 * block reservation is off
375 if (!test_opt(sb
, RESERVATION
))
376 rsv
->rsv_goal_size
= 0;
378 rsv
->rsv_goal_size
= EXT2_DEFAULT_RESERVE_BLOCKS
;
379 rsv
->rsv_alloc_hit
= 0;
380 block_i
->last_alloc_logical_block
= 0;
381 block_i
->last_alloc_physical_block
= 0;
383 ei
->i_block_alloc_info
= block_i
;
387 * ext2_discard_reservation()
390 * Discard(free) block reservation window on last file close, or truncate
393 * It is being called in three cases:
394 * ext2_release_file(): last writer closes the file
395 * ext2_clear_inode(): last iput(), when nobody links to this file.
396 * ext2_truncate(): when the block indirect map is about to change.
398 void ext2_discard_reservation(struct inode
*inode
)
400 struct ext2_inode_info
*ei
= EXT2_I(inode
);
401 struct ext2_block_alloc_info
*block_i
= ei
->i_block_alloc_info
;
402 struct ext2_reserve_window_node
*rsv
;
403 spinlock_t
*rsv_lock
= &EXT2_SB(inode
->i_sb
)->s_rsv_window_lock
;
408 rsv
= &block_i
->rsv_window_node
;
409 if (!rsv_is_empty(&rsv
->rsv_window
)) {
411 if (!rsv_is_empty(&rsv
->rsv_window
))
412 rsv_window_remove(inode
->i_sb
, rsv
);
413 spin_unlock(rsv_lock
);
418 * ext2_free_blocks_sb() -- Free given blocks and update quota and i_blocks
420 * @block: start physcial block to free
421 * @count: number of blocks to free
423 void ext2_free_blocks (struct inode
* inode
, unsigned long block
,
426 struct buffer_head
*bitmap_bh
= NULL
;
427 struct buffer_head
* bh2
;
428 unsigned long block_group
;
431 unsigned long overflow
;
432 struct super_block
* sb
= inode
->i_sb
;
433 struct ext2_sb_info
* sbi
= EXT2_SB(sb
);
434 struct ext2_group_desc
* desc
;
435 struct ext2_super_block
* es
= sbi
->s_es
;
436 unsigned freed
= 0, group_freed
;
438 if (block
< le32_to_cpu(es
->s_first_data_block
) ||
439 block
+ count
< block
||
440 block
+ count
> le32_to_cpu(es
->s_blocks_count
)) {
441 ext2_error (sb
, "ext2_free_blocks",
442 "Freeing blocks not in datazone - "
443 "block = %lu, count = %lu", block
, count
);
447 ext2_debug ("freeing block(s) %lu-%lu\n", block
, block
+ count
- 1);
451 block_group
= (block
- le32_to_cpu(es
->s_first_data_block
)) /
452 EXT2_BLOCKS_PER_GROUP(sb
);
453 bit
= (block
- le32_to_cpu(es
->s_first_data_block
)) %
454 EXT2_BLOCKS_PER_GROUP(sb
);
456 * Check to see if we are freeing blocks across a group
459 if (bit
+ count
> EXT2_BLOCKS_PER_GROUP(sb
)) {
460 overflow
= bit
+ count
- EXT2_BLOCKS_PER_GROUP(sb
);
464 bitmap_bh
= read_block_bitmap(sb
, block_group
);
468 desc
= ext2_get_group_desc (sb
, block_group
, &bh2
);
472 if (in_range (le32_to_cpu(desc
->bg_block_bitmap
), block
, count
) ||
473 in_range (le32_to_cpu(desc
->bg_inode_bitmap
), block
, count
) ||
474 in_range (block
, le32_to_cpu(desc
->bg_inode_table
),
475 sbi
->s_itb_per_group
) ||
476 in_range (block
+ count
- 1, le32_to_cpu(desc
->bg_inode_table
),
477 sbi
->s_itb_per_group
))
478 ext2_error (sb
, "ext2_free_blocks",
479 "Freeing blocks in system zones - "
480 "Block = %lu, count = %lu",
483 for (i
= 0, group_freed
= 0; i
< count
; i
++) {
484 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi
, block_group
),
485 bit
+ i
, bitmap_bh
->b_data
)) {
486 ext2_error(sb
, __FUNCTION__
,
487 "bit already cleared for block %lu", block
+ i
);
493 mark_buffer_dirty(bitmap_bh
);
494 if (sb
->s_flags
& MS_SYNCHRONOUS
)
495 sync_dirty_buffer(bitmap_bh
);
497 group_adjust_blocks(sb
, block_group
, desc
, bh2
, group_freed
);
498 freed
+= group_freed
;
507 release_blocks(sb
, freed
);
508 DQUOT_FREE_BLOCK(inode
, freed
);
512 * bitmap_search_next_usable_block()
513 * @start: the starting block (group relative) of the search
514 * @bh: bufferhead contains the block group bitmap
515 * @maxblocks: the ending block (group relative) of the reservation
517 * The bitmap search --- search forward through the actual bitmap on disk until
518 * we find a bit free.
521 bitmap_search_next_usable_block(ext2_grpblk_t start
, struct buffer_head
*bh
,
522 ext2_grpblk_t maxblocks
)
526 next
= ext2_find_next_zero_bit(bh
->b_data
, maxblocks
, start
);
527 if (next
>= maxblocks
)
533 * find_next_usable_block()
534 * @start: the starting block (group relative) to find next
535 * allocatable block in bitmap.
536 * @bh: bufferhead contains the block group bitmap
537 * @maxblocks: the ending block (group relative) for the search
539 * Find an allocatable block in a bitmap. We perform the "most
540 * appropriate allocation" algorithm of looking for a free block near
541 * the initial goal; then for a free byte somewhere in the bitmap;
542 * then for any free bit in the bitmap.
545 find_next_usable_block(int start
, struct buffer_head
*bh
, int maxblocks
)
547 ext2_grpblk_t here
, next
;
552 * The goal was occupied; search forward for a free
553 * block within the next XX blocks.
555 * end_goal is more or less random, but it has to be
556 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
557 * next 64-bit boundary is simple..
559 ext2_grpblk_t end_goal
= (start
+ 63) & ~63;
560 if (end_goal
> maxblocks
)
561 end_goal
= maxblocks
;
562 here
= ext2_find_next_zero_bit(bh
->b_data
, end_goal
, start
);
565 ext2_debug("Bit not found near goal\n");
572 p
= ((char *)bh
->b_data
) + (here
>> 3);
573 r
= memscan(p
, 0, ((maxblocks
+ 7) >> 3) - (here
>> 3));
574 next
= (r
- ((char *)bh
->b_data
)) << 3;
576 if (next
< maxblocks
&& next
>= here
)
579 here
= bitmap_search_next_usable_block(here
, bh
, maxblocks
);
584 * ext2_try_to_allocate()
586 * @handle: handle to this transaction
587 * @group: given allocation block group
588 * @bitmap_bh: bufferhead holds the block bitmap
589 * @grp_goal: given target block within the group
590 * @count: target number of blocks to allocate
591 * @my_rsv: reservation window
593 * Attempt to allocate blocks within a give range. Set the range of allocation
594 * first, then find the first free bit(s) from the bitmap (within the range),
595 * and at last, allocate the blocks by claiming the found free bit as allocated.
597 * To set the range of this allocation:
598 * if there is a reservation window, only try to allocate block(s)
599 * from the file's own reservation window;
600 * Otherwise, the allocation range starts from the give goal block,
601 * ends at the block group's last block.
603 * If we failed to allocate the desired block then we may end up crossing to a
607 ext2_try_to_allocate(struct super_block
*sb
, int group
,
608 struct buffer_head
*bitmap_bh
, ext2_grpblk_t grp_goal
,
609 unsigned long *count
,
610 struct ext2_reserve_window
*my_rsv
)
612 ext2_fsblk_t group_first_block
;
613 ext2_grpblk_t start
, end
;
614 unsigned long num
= 0;
616 /* we do allocation within the reservation window if we have a window */
618 group_first_block
= ext2_group_first_block_no(sb
, group
);
619 if (my_rsv
->_rsv_start
>= group_first_block
)
620 start
= my_rsv
->_rsv_start
- group_first_block
;
622 /* reservation window cross group boundary */
624 end
= my_rsv
->_rsv_end
- group_first_block
+ 1;
625 if (end
> EXT2_BLOCKS_PER_GROUP(sb
))
626 /* reservation window crosses group boundary */
627 end
= EXT2_BLOCKS_PER_GROUP(sb
);
628 if ((start
<= grp_goal
) && (grp_goal
< end
))
637 end
= EXT2_BLOCKS_PER_GROUP(sb
);
640 BUG_ON(start
> EXT2_BLOCKS_PER_GROUP(sb
));
644 grp_goal
= find_next_usable_block(start
, bitmap_bh
, end
);
650 for (i
= 0; i
< 7 && grp_goal
> start
&&
651 !ext2_test_bit(grp_goal
- 1,
659 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb
), group
), grp_goal
,
660 bitmap_bh
->b_data
)) {
662 * The block was allocated by another thread, or it was
663 * allocated and then freed by another thread
673 while (num
< *count
&& grp_goal
< end
674 && !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb
), group
),
675 grp_goal
, bitmap_bh
->b_data
)) {
680 return grp_goal
- num
;
687 * find_next_reservable_window():
688 * find a reservable space within the given range.
689 * It does not allocate the reservation window for now:
690 * alloc_new_reservation() will do the work later.
692 * @search_head: the head of the searching list;
693 * This is not necessarily the list head of the whole filesystem
695 * We have both head and start_block to assist the search
696 * for the reservable space. The list starts from head,
697 * but we will shift to the place where start_block is,
698 * then start from there, when looking for a reservable space.
700 * @size: the target new reservation window size
702 * @group_first_block: the first block we consider to start
703 * the real search from
706 * the maximum block number that our goal reservable space
707 * could start from. This is normally the last block in this
708 * group. The search will end when we found the start of next
709 * possible reservable space is out of this boundary.
710 * This could handle the cross boundary reservation window
713 * basically we search from the given range, rather than the whole
714 * reservation double linked list, (start_block, last_block)
715 * to find a free region that is of my size and has not
719 static int find_next_reservable_window(
720 struct ext2_reserve_window_node
*search_head
,
721 struct ext2_reserve_window_node
*my_rsv
,
722 struct super_block
* sb
,
723 ext2_fsblk_t start_block
,
724 ext2_fsblk_t last_block
)
726 struct rb_node
*next
;
727 struct ext2_reserve_window_node
*rsv
, *prev
;
729 int size
= my_rsv
->rsv_goal_size
;
731 /* TODO: make the start of the reservation window byte-aligned */
732 /* cur = *start_block & ~7;*/
739 if (cur
<= rsv
->rsv_end
)
740 cur
= rsv
->rsv_end
+ 1;
743 * in the case we could not find a reservable space
744 * that is what is expected, during the re-search, we could
745 * remember what's the largest reservable space we could have
746 * and return that one.
748 * For now it will fail if we could not find the reservable
749 * space with expected-size (or more)...
751 if (cur
> last_block
)
752 return -1; /* fail */
755 next
= rb_next(&rsv
->rsv_node
);
756 rsv
= rb_entry(next
,struct ext2_reserve_window_node
,rsv_node
);
759 * Reached the last reservation, we can just append to the
765 if (cur
+ size
<= rsv
->rsv_start
) {
767 * Found a reserveable space big enough. We could
768 * have a reservation across the group boundary here
774 * we come here either :
775 * when we reach the end of the whole list,
776 * and there is empty reservable space after last entry in the list.
777 * append it to the end of the list.
779 * or we found one reservable space in the middle of the list,
780 * return the reservation window that we could append to.
784 if ((prev
!= my_rsv
) && (!rsv_is_empty(&my_rsv
->rsv_window
)))
785 rsv_window_remove(sb
, my_rsv
);
788 * Let's book the whole avaliable window for now. We will check the
789 * disk bitmap later and then, if there are free blocks then we adjust
790 * the window size if it's larger than requested.
791 * Otherwise, we will remove this node from the tree next time
792 * call find_next_reservable_window.
794 my_rsv
->rsv_start
= cur
;
795 my_rsv
->rsv_end
= cur
+ size
- 1;
796 my_rsv
->rsv_alloc_hit
= 0;
799 ext2_rsv_window_add(sb
, my_rsv
);
805 * alloc_new_reservation()--allocate a new reservation window
807 * To make a new reservation, we search part of the filesystem
808 * reservation list (the list that inside the group). We try to
809 * allocate a new reservation window near the allocation goal,
810 * or the beginning of the group, if there is no goal.
812 * We first find a reservable space after the goal, then from
813 * there, we check the bitmap for the first free block after
814 * it. If there is no free block until the end of group, then the
815 * whole group is full, we failed. Otherwise, check if the free
816 * block is inside the expected reservable space, if so, we
818 * If the first free block is outside the reservable space, then
819 * start from the first free block, we search for next available
822 * on succeed, a new reservation will be found and inserted into the list
823 * It contains at least one free block, and it does not overlap with other
824 * reservation windows.
826 * failed: we failed to find a reservation window in this group
828 * @rsv: the reservation
830 * @grp_goal: The goal (group-relative). It is where the search for a
831 * free reservable space should start from.
832 * if we have a goal(goal >0 ), then start from there,
833 * no goal(goal = -1), we start from the first block
836 * @sb: the super block
837 * @group: the group we are trying to allocate in
838 * @bitmap_bh: the block group block bitmap
841 static int alloc_new_reservation(struct ext2_reserve_window_node
*my_rsv
,
842 ext2_grpblk_t grp_goal
, struct super_block
*sb
,
843 unsigned int group
, struct buffer_head
*bitmap_bh
)
845 struct ext2_reserve_window_node
*search_head
;
846 ext2_fsblk_t group_first_block
, group_end_block
, start_block
;
847 ext2_grpblk_t first_free_block
;
848 struct rb_root
*fs_rsv_root
= &EXT2_SB(sb
)->s_rsv_window_root
;
851 spinlock_t
*rsv_lock
= &EXT2_SB(sb
)->s_rsv_window_lock
;
853 group_first_block
= ext2_group_first_block_no(sb
, group
);
854 group_end_block
= group_first_block
+ (EXT2_BLOCKS_PER_GROUP(sb
) - 1);
857 start_block
= group_first_block
;
859 start_block
= grp_goal
+ group_first_block
;
861 size
= my_rsv
->rsv_goal_size
;
863 if (!rsv_is_empty(&my_rsv
->rsv_window
)) {
865 * if the old reservation is cross group boundary
866 * and if the goal is inside the old reservation window,
867 * we will come here when we just failed to allocate from
868 * the first part of the window. We still have another part
869 * that belongs to the next group. In this case, there is no
870 * point to discard our window and try to allocate a new one
871 * in this group(which will fail). we should
872 * keep the reservation window, just simply move on.
874 * Maybe we could shift the start block of the reservation
875 * window to the first block of next group.
878 if ((my_rsv
->rsv_start
<= group_end_block
) &&
879 (my_rsv
->rsv_end
> group_end_block
) &&
880 (start_block
>= my_rsv
->rsv_start
))
883 if ((my_rsv
->rsv_alloc_hit
>
884 (my_rsv
->rsv_end
- my_rsv
->rsv_start
+ 1) / 2)) {
886 * if the previously allocation hit ratio is
887 * greater than 1/2, then we double the size of
888 * the reservation window the next time,
889 * otherwise we keep the same size window
892 if (size
> EXT2_MAX_RESERVE_BLOCKS
)
893 size
= EXT2_MAX_RESERVE_BLOCKS
;
894 my_rsv
->rsv_goal_size
= size
;
900 * shift the search start to the window near the goal block
902 search_head
= search_reserve_window(fs_rsv_root
, start_block
);
905 * find_next_reservable_window() simply finds a reservable window
906 * inside the given range(start_block, group_end_block).
908 * To make sure the reservation window has a free bit inside it, we
909 * need to check the bitmap after we found a reservable window.
912 ret
= find_next_reservable_window(search_head
, my_rsv
, sb
,
913 start_block
, group_end_block
);
916 if (!rsv_is_empty(&my_rsv
->rsv_window
))
917 rsv_window_remove(sb
, my_rsv
);
918 spin_unlock(rsv_lock
);
923 * On success, find_next_reservable_window() returns the
924 * reservation window where there is a reservable space after it.
925 * Before we reserve this reservable space, we need
926 * to make sure there is at least a free block inside this region.
928 * Search the first free bit on the block bitmap. Search starts from
929 * the start block of the reservable space we just found.
931 spin_unlock(rsv_lock
);
932 first_free_block
= bitmap_search_next_usable_block(
933 my_rsv
->rsv_start
- group_first_block
,
934 bitmap_bh
, group_end_block
- group_first_block
+ 1);
936 if (first_free_block
< 0) {
938 * no free block left on the bitmap, no point
939 * to reserve the space. return failed.
942 if (!rsv_is_empty(&my_rsv
->rsv_window
))
943 rsv_window_remove(sb
, my_rsv
);
944 spin_unlock(rsv_lock
);
945 return -1; /* failed */
948 start_block
= first_free_block
+ group_first_block
;
950 * check if the first free block is within the
951 * free space we just reserved
953 if (start_block
>= my_rsv
->rsv_start
&& start_block
<= my_rsv
->rsv_end
)
954 return 0; /* success */
956 * if the first free bit we found is out of the reservable space
957 * continue search for next reservable space,
958 * start from where the free block is,
959 * we also shift the list head to where we stopped last time
961 search_head
= my_rsv
;
967 * try_to_extend_reservation()
968 * @my_rsv: given reservation window
970 * @size: the delta to extend
972 * Attempt to expand the reservation window large enough to have
973 * required number of free blocks
975 * Since ext2_try_to_allocate() will always allocate blocks within
976 * the reservation window range, if the window size is too small,
977 * multiple blocks allocation has to stop at the end of the reservation
978 * window. To make this more efficient, given the total number of
979 * blocks needed and the current size of the window, we try to
980 * expand the reservation window size if necessary on a best-effort
981 * basis before ext2_new_blocks() tries to allocate blocks.
983 static void try_to_extend_reservation(struct ext2_reserve_window_node
*my_rsv
,
984 struct super_block
*sb
, int size
)
986 struct ext2_reserve_window_node
*next_rsv
;
987 struct rb_node
*next
;
988 spinlock_t
*rsv_lock
= &EXT2_SB(sb
)->s_rsv_window_lock
;
990 if (!spin_trylock(rsv_lock
))
993 next
= rb_next(&my_rsv
->rsv_node
);
996 my_rsv
->rsv_end
+= size
;
998 next_rsv
= rb_entry(next
, struct ext2_reserve_window_node
, rsv_node
);
1000 if ((next_rsv
->rsv_start
- my_rsv
->rsv_end
- 1) >= size
)
1001 my_rsv
->rsv_end
+= size
;
1003 my_rsv
->rsv_end
= next_rsv
->rsv_start
- 1;
1005 spin_unlock(rsv_lock
);
1009 * ext2_try_to_allocate_with_rsv()
1011 * @group: given allocation block group
1012 * @bitmap_bh: bufferhead holds the block bitmap
1013 * @grp_goal: given target block within the group
1014 * @count: target number of blocks to allocate
1015 * @my_rsv: reservation window
1017 * This is the main function used to allocate a new block and its reservation
1020 * Each time when a new block allocation is need, first try to allocate from
1021 * its own reservation. If it does not have a reservation window, instead of
1022 * looking for a free bit on bitmap first, then look up the reservation list to
1023 * see if it is inside somebody else's reservation window, we try to allocate a
1024 * reservation window for it starting from the goal first. Then do the block
1025 * allocation within the reservation window.
1027 * This will avoid keeping on searching the reservation list again and
1028 * again when somebody is looking for a free block (without
1029 * reservation), and there are lots of free blocks, but they are all
1032 * We use a red-black tree for the per-filesystem reservation list.
1034 static ext2_grpblk_t
1035 ext2_try_to_allocate_with_rsv(struct super_block
*sb
, unsigned int group
,
1036 struct buffer_head
*bitmap_bh
, ext2_grpblk_t grp_goal
,
1037 struct ext2_reserve_window_node
* my_rsv
,
1038 unsigned long *count
)
1040 ext2_fsblk_t group_first_block
, group_last_block
;
1041 ext2_grpblk_t ret
= 0;
1042 unsigned long num
= *count
;
1045 * we don't deal with reservation when
1046 * filesystem is mounted without reservation
1047 * or the file is not a regular file
1048 * or last attempt to allocate a block with reservation turned on failed
1050 if (my_rsv
== NULL
) {
1051 return ext2_try_to_allocate(sb
, group
, bitmap_bh
,
1052 grp_goal
, count
, NULL
);
1055 * grp_goal is a group relative block number (if there is a goal)
1056 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1057 * first block is a filesystem wide block number
1058 * first block is the block number of the first block in this group
1060 group_first_block
= ext2_group_first_block_no(sb
, group
);
1061 group_last_block
= group_first_block
+ (EXT2_BLOCKS_PER_GROUP(sb
) - 1);
1064 * Basically we will allocate a new block from inode's reservation
1067 * We need to allocate a new reservation window, if:
1068 * a) inode does not have a reservation window; or
1069 * b) last attempt to allocate a block from existing reservation
1071 * c) we come here with a goal and with a reservation window
1073 * We do not need to allocate a new reservation window if we come here
1074 * at the beginning with a goal and the goal is inside the window, or
1075 * we don't have a goal but already have a reservation window.
1076 * then we could go to allocate from the reservation window directly.
1079 if (rsv_is_empty(&my_rsv
->rsv_window
) || (ret
< 0) ||
1080 !goal_in_my_reservation(&my_rsv
->rsv_window
,
1081 grp_goal
, group
, sb
)) {
1082 if (my_rsv
->rsv_goal_size
< *count
)
1083 my_rsv
->rsv_goal_size
= *count
;
1084 ret
= alloc_new_reservation(my_rsv
, grp_goal
, sb
,
1089 if (!goal_in_my_reservation(&my_rsv
->rsv_window
,
1090 grp_goal
, group
, sb
))
1092 } else if (grp_goal
>= 0) {
1093 int curr
= my_rsv
->rsv_end
-
1094 (grp_goal
+ group_first_block
) + 1;
1097 try_to_extend_reservation(my_rsv
, sb
,
1101 if ((my_rsv
->rsv_start
> group_last_block
) ||
1102 (my_rsv
->rsv_end
< group_first_block
)) {
1103 rsv_window_dump(&EXT2_SB(sb
)->s_rsv_window_root
, 1);
1106 ret
= ext2_try_to_allocate(sb
, group
, bitmap_bh
, grp_goal
,
1107 &num
, &my_rsv
->rsv_window
);
1109 my_rsv
->rsv_alloc_hit
+= num
;
1111 break; /* succeed */
1119 * ext2_has_free_blocks()
1120 * @sbi: in-core super block structure.
1122 * Check if filesystem has at least 1 free block available for allocation.
1124 static int ext2_has_free_blocks(struct ext2_sb_info
*sbi
)
1126 ext2_fsblk_t free_blocks
, root_blocks
;
1128 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
1129 root_blocks
= le32_to_cpu(sbi
->s_es
->s_r_blocks_count
);
1130 if (free_blocks
< root_blocks
+ 1 && !capable(CAP_SYS_RESOURCE
) &&
1131 sbi
->s_resuid
!= current
->fsuid
&&
1132 (sbi
->s_resgid
== 0 || !in_group_p (sbi
->s_resgid
))) {
1139 * ext2_new_blocks() -- core block(s) allocation function
1140 * @inode: file inode
1141 * @goal: given target block(filesystem wide)
1142 * @count: target number of blocks to allocate
1145 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
1146 * free, or there is a free block within 32 blocks of the goal, that block
1147 * is allocated. Otherwise a forward search is made for a free block; within
1148 * each block group the search first looks for an entire free byte in the block
1149 * bitmap, and then for any free bit if that fails.
1150 * This function also updates quota and i_blocks field.
1152 ext2_fsblk_t
ext2_new_blocks(struct inode
*inode
, ext2_fsblk_t goal
,
1153 unsigned long *count
, int *errp
)
1155 struct buffer_head
*bitmap_bh
= NULL
;
1156 struct buffer_head
*gdp_bh
;
1159 ext2_grpblk_t grp_target_blk
; /* blockgroup relative goal block */
1160 ext2_grpblk_t grp_alloc_blk
; /* blockgroup-relative allocated block*/
1161 ext2_fsblk_t ret_block
; /* filesyetem-wide allocated block */
1162 int bgi
; /* blockgroup iteration index */
1163 int performed_allocation
= 0;
1164 ext2_grpblk_t free_blocks
; /* number of free blocks in a group */
1165 struct super_block
*sb
;
1166 struct ext2_group_desc
*gdp
;
1167 struct ext2_super_block
*es
;
1168 struct ext2_sb_info
*sbi
;
1169 struct ext2_reserve_window_node
*my_rsv
= NULL
;
1170 struct ext2_block_alloc_info
*block_i
;
1171 unsigned short windowsz
= 0;
1172 unsigned long ngroups
;
1173 unsigned long num
= *count
;
1178 printk("ext2_new_blocks: nonexistent device");
1183 * Check quota for allocation of this block.
1185 if (DQUOT_ALLOC_BLOCK(inode
, num
)) {
1191 es
= EXT2_SB(sb
)->s_es
;
1192 ext2_debug("goal=%lu.\n", goal
);
1194 * Allocate a block from reservation only when
1195 * filesystem is mounted with reservation(default,-o reservation), and
1196 * it's a regular file, and
1197 * the desired window size is greater than 0 (One could use ioctl
1198 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1199 * reservation on that particular file)
1201 block_i
= EXT2_I(inode
)->i_block_alloc_info
;
1203 windowsz
= block_i
->rsv_window_node
.rsv_goal_size
;
1205 my_rsv
= &block_i
->rsv_window_node
;
1208 if (!ext2_has_free_blocks(sbi
)) {
1214 * First, test whether the goal block is free.
1216 if (goal
< le32_to_cpu(es
->s_first_data_block
) ||
1217 goal
>= le32_to_cpu(es
->s_blocks_count
))
1218 goal
= le32_to_cpu(es
->s_first_data_block
);
1219 group_no
= (goal
- le32_to_cpu(es
->s_first_data_block
)) /
1220 EXT2_BLOCKS_PER_GROUP(sb
);
1221 goal_group
= group_no
;
1223 gdp
= ext2_get_group_desc(sb
, group_no
, &gdp_bh
);
1227 free_blocks
= le16_to_cpu(gdp
->bg_free_blocks_count
);
1229 * if there is not enough free blocks to make a new resevation
1230 * turn off reservation for this allocation
1232 if (my_rsv
&& (free_blocks
< windowsz
)
1233 && (rsv_is_empty(&my_rsv
->rsv_window
)))
1236 if (free_blocks
> 0) {
1237 grp_target_blk
= ((goal
- le32_to_cpu(es
->s_first_data_block
)) %
1238 EXT2_BLOCKS_PER_GROUP(sb
));
1239 bitmap_bh
= read_block_bitmap(sb
, group_no
);
1242 grp_alloc_blk
= ext2_try_to_allocate_with_rsv(sb
, group_no
,
1243 bitmap_bh
, grp_target_blk
,
1245 if (grp_alloc_blk
>= 0)
1249 ngroups
= EXT2_SB(sb
)->s_groups_count
;
1253 * Now search the rest of the groups. We assume that
1254 * i and gdp correctly point to the last group visited.
1256 for (bgi
= 0; bgi
< ngroups
; bgi
++) {
1258 if (group_no
>= ngroups
)
1260 gdp
= ext2_get_group_desc(sb
, group_no
, &gdp_bh
);
1264 free_blocks
= le16_to_cpu(gdp
->bg_free_blocks_count
);
1266 * skip this group if the number of
1267 * free blocks is less than half of the reservation
1270 if (free_blocks
<= (windowsz
/2))
1274 bitmap_bh
= read_block_bitmap(sb
, group_no
);
1278 * try to allocate block(s) from this group, without a goal(-1).
1280 grp_alloc_blk
= ext2_try_to_allocate_with_rsv(sb
, group_no
,
1281 bitmap_bh
, -1, my_rsv
, &num
);
1282 if (grp_alloc_blk
>= 0)
1286 * We may end up a bogus ealier ENOSPC error due to
1287 * filesystem is "full" of reservations, but
1288 * there maybe indeed free blocks avaliable on disk
1289 * In this case, we just forget about the reservations
1290 * just do block allocation as without reservations.
1295 group_no
= goal_group
;
1298 /* No space left on the device */
1304 ext2_debug("using block group %d(%d)\n",
1305 group_no
, gdp
->bg_free_blocks_count
);
1307 ret_block
= grp_alloc_blk
+ ext2_group_first_block_no(sb
, group_no
);
1309 if (in_range(le32_to_cpu(gdp
->bg_block_bitmap
), ret_block
, num
) ||
1310 in_range(le32_to_cpu(gdp
->bg_inode_bitmap
), ret_block
, num
) ||
1311 in_range(ret_block
, le32_to_cpu(gdp
->bg_inode_table
),
1312 EXT2_SB(sb
)->s_itb_per_group
) ||
1313 in_range(ret_block
+ num
- 1, le32_to_cpu(gdp
->bg_inode_table
),
1314 EXT2_SB(sb
)->s_itb_per_group
))
1315 ext2_error(sb
, "ext2_new_blocks",
1316 "Allocating block in system zone - "
1317 "blocks from "E2FSBLK
", length %lu",
1320 performed_allocation
= 1;
1322 if (ret_block
+ num
- 1 >= le32_to_cpu(es
->s_blocks_count
)) {
1323 ext2_error(sb
, "ext2_new_blocks",
1324 "block("E2FSBLK
") >= blocks count(%d) - "
1325 "block_group = %d, es == %p ", ret_block
,
1326 le32_to_cpu(es
->s_blocks_count
), group_no
, es
);
1330 group_adjust_blocks(sb
, group_no
, gdp
, gdp_bh
, -num
);
1331 percpu_counter_sub(&sbi
->s_freeblocks_counter
, num
);
1333 mark_buffer_dirty(bitmap_bh
);
1334 if (sb
->s_flags
& MS_SYNCHRONOUS
)
1335 sync_dirty_buffer(bitmap_bh
);
1339 DQUOT_FREE_BLOCK(inode
, *count
-num
);
1347 * Undo the block allocation
1349 if (!performed_allocation
)
1350 DQUOT_FREE_BLOCK(inode
, *count
);
1355 ext2_fsblk_t
ext2_new_block(struct inode
*inode
, unsigned long goal
, int *errp
)
1357 unsigned long count
= 1;
1359 return ext2_new_blocks(inode
, goal
, &count
, errp
);
1364 static const int nibblemap
[] = {4, 3, 3, 2, 3, 2, 2, 1, 3, 2, 2, 1, 2, 1, 1, 0};
1366 unsigned long ext2_count_free (struct buffer_head
* map
, unsigned int numchars
)
1369 unsigned long sum
= 0;
1373 for (i
= 0; i
< numchars
; i
++)
1374 sum
+= nibblemap
[map
->b_data
[i
] & 0xf] +
1375 nibblemap
[(map
->b_data
[i
] >> 4) & 0xf];
1379 #endif /* EXT2FS_DEBUG */
1381 unsigned long ext2_count_free_blocks (struct super_block
* sb
)
1383 struct ext2_group_desc
* desc
;
1384 unsigned long desc_count
= 0;
1387 unsigned long bitmap_count
, x
;
1388 struct ext2_super_block
*es
;
1390 es
= EXT2_SB(sb
)->s_es
;
1394 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
1395 struct buffer_head
*bitmap_bh
;
1396 desc
= ext2_get_group_desc (sb
, i
, NULL
);
1399 desc_count
+= le16_to_cpu(desc
->bg_free_blocks_count
);
1400 bitmap_bh
= read_block_bitmap(sb
, i
);
1404 x
= ext2_count_free(bitmap_bh
, sb
->s_blocksize
);
1405 printk ("group %d: stored = %d, counted = %lu\n",
1406 i
, le16_to_cpu(desc
->bg_free_blocks_count
), x
);
1410 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1411 (long)le32_to_cpu(es
->s_free_blocks_count
),
1412 desc_count
, bitmap_count
);
1413 return bitmap_count
;
1415 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
1416 desc
= ext2_get_group_desc (sb
, i
, NULL
);
1419 desc_count
+= le16_to_cpu(desc
->bg_free_blocks_count
);
1425 static inline int test_root(int a
, int b
)
1434 static int ext2_group_sparse(int group
)
1438 return (test_root(group
, 3) || test_root(group
, 5) ||
1439 test_root(group
, 7));
1443 * ext2_bg_has_super - number of blocks used by the superblock in group
1444 * @sb: superblock for filesystem
1445 * @group: group number to check
1447 * Return the number of blocks used by the superblock (primary or backup)
1448 * in this group. Currently this will be only 0 or 1.
1450 int ext2_bg_has_super(struct super_block
*sb
, int group
)
1452 if (EXT2_HAS_RO_COMPAT_FEATURE(sb
,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER
)&&
1453 !ext2_group_sparse(group
))
1459 * ext2_bg_num_gdb - number of blocks used by the group table in group
1460 * @sb: superblock for filesystem
1461 * @group: group number to check
1463 * Return the number of blocks used by the group descriptor table
1464 * (primary or backup) in this group. In the future there may be a
1465 * different number of descriptor blocks in each group.
1467 unsigned long ext2_bg_num_gdb(struct super_block
*sb
, int group
)
1469 if (EXT2_HAS_RO_COMPAT_FEATURE(sb
,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER
)&&
1470 !ext2_group_sparse(group
))
1472 return EXT2_SB(sb
)->s_gdb_count
;