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/quotaops.h>
19 #include <linux/buffer_head.h>
21 #include "ext4_jbd2.h"
25 * balloc.c contains the blocks allocation and deallocation routines
29 * Calculate the block group number and offset, given a block number
31 void ext4_get_group_no_and_offset(struct super_block
*sb
, ext4_fsblk_t blocknr
,
32 ext4_group_t
*blockgrpp
, ext4_grpblk_t
*offsetp
)
34 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
37 blocknr
= blocknr
- le32_to_cpu(es
->s_first_data_block
);
38 offset
= do_div(blocknr
, EXT4_BLOCKS_PER_GROUP(sb
));
46 static int ext4_block_in_group(struct super_block
*sb
, ext4_fsblk_t block
,
47 ext4_group_t block_group
)
49 ext4_group_t actual_group
;
50 ext4_get_group_no_and_offset(sb
, block
, &actual_group
, 0);
51 if (actual_group
== block_group
)
56 static int ext4_group_used_meta_blocks(struct super_block
*sb
,
57 ext4_group_t block_group
)
60 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
61 /* block bitmap, inode bitmap, and inode table blocks */
62 int used_blocks
= sbi
->s_itb_per_group
+ 2;
64 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FLEX_BG
)) {
65 struct ext4_group_desc
*gdp
;
66 struct buffer_head
*bh
;
68 gdp
= ext4_get_group_desc(sb
, block_group
, &bh
);
69 if (!ext4_block_in_group(sb
, ext4_block_bitmap(sb
, gdp
),
73 if (!ext4_block_in_group(sb
, ext4_inode_bitmap(sb
, gdp
),
77 tmp
= ext4_inode_table(sb
, gdp
);
78 for (; tmp
< ext4_inode_table(sb
, gdp
) +
79 sbi
->s_itb_per_group
; tmp
++) {
80 if (!ext4_block_in_group(sb
, tmp
, block_group
))
86 /* Initializes an uninitialized block bitmap if given, and returns the
87 * number of blocks free in the group. */
88 unsigned ext4_init_block_bitmap(struct super_block
*sb
, struct buffer_head
*bh
,
89 ext4_group_t block_group
, struct ext4_group_desc
*gdp
)
92 unsigned free_blocks
, group_blocks
;
93 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
96 J_ASSERT_BH(bh
, buffer_locked(bh
));
98 /* If checksum is bad mark all blocks used to prevent allocation
99 * essentially implementing a per-group read-only flag. */
100 if (!ext4_group_desc_csum_verify(sbi
, block_group
, gdp
)) {
101 ext4_error(sb
, __func__
,
102 "Checksum bad for group %lu\n", block_group
);
103 gdp
->bg_free_blocks_count
= 0;
104 gdp
->bg_free_inodes_count
= 0;
105 gdp
->bg_itable_unused
= 0;
106 memset(bh
->b_data
, 0xff, sb
->s_blocksize
);
109 memset(bh
->b_data
, 0, sb
->s_blocksize
);
112 /* Check for superblock and gdt backups in this group */
113 bit_max
= ext4_bg_has_super(sb
, block_group
);
115 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_META_BG
) ||
116 block_group
< le32_to_cpu(sbi
->s_es
->s_first_meta_bg
) *
117 sbi
->s_desc_per_block
) {
119 bit_max
+= ext4_bg_num_gdb(sb
, block_group
);
121 le16_to_cpu(sbi
->s_es
->s_reserved_gdt_blocks
);
123 } else { /* For META_BG_BLOCK_GROUPS */
124 int group_rel
= (block_group
-
125 le32_to_cpu(sbi
->s_es
->s_first_meta_bg
)) %
126 EXT4_DESC_PER_BLOCK(sb
);
127 if (group_rel
== 0 || group_rel
== 1 ||
128 (group_rel
== EXT4_DESC_PER_BLOCK(sb
) - 1))
132 if (block_group
== sbi
->s_groups_count
- 1) {
134 * Even though mke2fs always initialize first and last group
135 * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
136 * to make sure we calculate the right free blocks
138 group_blocks
= ext4_blocks_count(sbi
->s_es
) -
139 le32_to_cpu(sbi
->s_es
->s_first_data_block
) -
140 (EXT4_BLOCKS_PER_GROUP(sb
) * (sbi
->s_groups_count
-1));
142 group_blocks
= EXT4_BLOCKS_PER_GROUP(sb
);
145 free_blocks
= group_blocks
- bit_max
;
148 ext4_fsblk_t start
, tmp
;
151 for (bit
= 0; bit
< bit_max
; bit
++)
152 ext4_set_bit(bit
, bh
->b_data
);
154 start
= ext4_group_first_block_no(sb
, block_group
);
156 if (EXT4_HAS_INCOMPAT_FEATURE(sb
,
157 EXT4_FEATURE_INCOMPAT_FLEX_BG
))
160 /* Set bits for block and inode bitmaps, and inode table */
161 tmp
= ext4_block_bitmap(sb
, gdp
);
162 if (!flex_bg
|| ext4_block_in_group(sb
, tmp
, block_group
))
163 ext4_set_bit(tmp
- start
, bh
->b_data
);
165 tmp
= ext4_inode_bitmap(sb
, gdp
);
166 if (!flex_bg
|| ext4_block_in_group(sb
, tmp
, block_group
))
167 ext4_set_bit(tmp
- start
, bh
->b_data
);
169 tmp
= ext4_inode_table(sb
, gdp
);
170 for (; tmp
< ext4_inode_table(sb
, gdp
) +
171 sbi
->s_itb_per_group
; tmp
++) {
173 ext4_block_in_group(sb
, tmp
, block_group
))
174 ext4_set_bit(tmp
- start
, bh
->b_data
);
177 * Also if the number of blocks within the group is
178 * less than the blocksize * 8 ( which is the size
179 * of bitmap ), set rest of the block bitmap to 1
181 mark_bitmap_end(group_blocks
, sb
->s_blocksize
* 8, bh
->b_data
);
183 return free_blocks
- ext4_group_used_meta_blocks(sb
, block_group
);
188 * The free blocks are managed by bitmaps. A file system contains several
189 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
190 * block for inodes, N blocks for the inode table and data blocks.
192 * The file system contains group descriptors which are located after the
193 * super block. Each descriptor contains the number of the bitmap block and
194 * the free blocks count in the block. The descriptors are loaded in memory
195 * when a file system is mounted (see ext4_fill_super).
199 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
202 * ext4_get_group_desc() -- load group descriptor from disk
204 * @block_group: given block group
205 * @bh: pointer to the buffer head to store the block
208 struct ext4_group_desc
* ext4_get_group_desc(struct super_block
* sb
,
209 ext4_group_t block_group
,
210 struct buffer_head
** bh
)
212 unsigned long group_desc
;
213 unsigned long offset
;
214 struct ext4_group_desc
* desc
;
215 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
217 if (block_group
>= sbi
->s_groups_count
) {
218 ext4_error (sb
, "ext4_get_group_desc",
219 "block_group >= groups_count - "
220 "block_group = %lu, groups_count = %lu",
221 block_group
, sbi
->s_groups_count
);
227 group_desc
= block_group
>> EXT4_DESC_PER_BLOCK_BITS(sb
);
228 offset
= block_group
& (EXT4_DESC_PER_BLOCK(sb
) - 1);
229 if (!sbi
->s_group_desc
[group_desc
]) {
230 ext4_error (sb
, "ext4_get_group_desc",
231 "Group descriptor not loaded - "
232 "block_group = %lu, group_desc = %lu, desc = %lu",
233 block_group
, group_desc
, offset
);
237 desc
= (struct ext4_group_desc
*)(
238 (__u8
*)sbi
->s_group_desc
[group_desc
]->b_data
+
239 offset
* EXT4_DESC_SIZE(sb
));
241 *bh
= sbi
->s_group_desc
[group_desc
];
245 static int ext4_valid_block_bitmap(struct super_block
*sb
,
246 struct ext4_group_desc
*desc
,
247 unsigned int block_group
,
248 struct buffer_head
*bh
)
250 ext4_grpblk_t offset
;
251 ext4_grpblk_t next_zero_bit
;
252 ext4_fsblk_t bitmap_blk
;
253 ext4_fsblk_t group_first_block
;
255 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FLEX_BG
)) {
256 /* with FLEX_BG, the inode/block bitmaps and itable
257 * blocks may not be in the group at all
258 * so the bitmap validation will be skipped for those groups
259 * or it has to also read the block group where the bitmaps
260 * are located to verify they are set.
264 group_first_block
= ext4_group_first_block_no(sb
, block_group
);
266 /* check whether block bitmap block number is set */
267 bitmap_blk
= ext4_block_bitmap(sb
, desc
);
268 offset
= bitmap_blk
- group_first_block
;
269 if (!ext4_test_bit(offset
, bh
->b_data
))
270 /* bad block bitmap */
273 /* check whether the inode bitmap block number is set */
274 bitmap_blk
= ext4_inode_bitmap(sb
, desc
);
275 offset
= bitmap_blk
- group_first_block
;
276 if (!ext4_test_bit(offset
, bh
->b_data
))
277 /* bad block bitmap */
280 /* check whether the inode table block number is set */
281 bitmap_blk
= ext4_inode_table(sb
, desc
);
282 offset
= bitmap_blk
- group_first_block
;
283 next_zero_bit
= ext4_find_next_zero_bit(bh
->b_data
,
284 offset
+ EXT4_SB(sb
)->s_itb_per_group
,
286 if (next_zero_bit
>= offset
+ EXT4_SB(sb
)->s_itb_per_group
)
287 /* good bitmap for inode tables */
291 ext4_error(sb
, __func__
,
292 "Invalid block bitmap - "
293 "block_group = %d, block = %llu",
294 block_group
, bitmap_blk
);
298 * ext4_read_block_bitmap()
300 * @block_group: given block group
302 * Read the bitmap for a given block_group,and validate the
303 * bits for block/inode/inode tables are set in the bitmaps
305 * Return buffer_head on success or NULL in case of failure.
308 ext4_read_block_bitmap(struct super_block
*sb
, ext4_group_t block_group
)
310 struct ext4_group_desc
* desc
;
311 struct buffer_head
* bh
= NULL
;
312 ext4_fsblk_t bitmap_blk
;
314 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
317 bitmap_blk
= ext4_block_bitmap(sb
, desc
);
318 bh
= sb_getblk(sb
, bitmap_blk
);
320 ext4_error(sb
, __func__
,
321 "Cannot read block bitmap - "
322 "block_group = %d, block_bitmap = %llu",
323 (int)block_group
, (unsigned long long)bitmap_blk
);
326 if (bh_uptodate_or_lock(bh
))
329 if (desc
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
330 ext4_init_block_bitmap(sb
, bh
, block_group
, desc
);
331 set_buffer_uptodate(bh
);
335 if (bh_submit_read(bh
) < 0) {
337 ext4_error(sb
, __func__
,
338 "Cannot read block bitmap - "
339 "block_group = %d, block_bitmap = %llu",
340 (int)block_group
, (unsigned long long)bitmap_blk
);
343 ext4_valid_block_bitmap(sb
, desc
, block_group
, bh
);
345 * file system mounted not to panic on error,
346 * continue with corrupt bitmap
351 * The reservation window structure operations
352 * --------------------------------------------
353 * Operations include:
354 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
356 * We use a red-black tree to represent per-filesystem reservation
362 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
363 * @rb_root: root of per-filesystem reservation rb tree
364 * @verbose: verbose mode
365 * @fn: function which wishes to dump the reservation map
367 * If verbose is turned on, it will print the whole block reservation
368 * windows(start, end). Otherwise, it will only print out the "bad" windows,
369 * those windows that overlap with their immediate neighbors.
372 static void __rsv_window_dump(struct rb_root
*root
, int verbose
,
376 struct ext4_reserve_window_node
*rsv
, *prev
;
384 printk("Block Allocation Reservation Windows Map (%s):\n", fn
);
386 rsv
= rb_entry(n
, struct ext4_reserve_window_node
, rsv_node
);
388 printk("reservation window 0x%p "
389 "start: %llu, end: %llu\n",
390 rsv
, rsv
->rsv_start
, rsv
->rsv_end
);
391 if (rsv
->rsv_start
&& rsv
->rsv_start
>= rsv
->rsv_end
) {
392 printk("Bad reservation %p (start >= end)\n",
396 if (prev
&& prev
->rsv_end
>= rsv
->rsv_start
) {
397 printk("Bad reservation %p (prev->end >= start)\n",
403 printk("Restarting reservation walk in verbose mode\n");
411 printk("Window map complete.\n");
414 #define rsv_window_dump(root, verbose) \
415 __rsv_window_dump((root), (verbose), __func__)
417 #define rsv_window_dump(root, verbose) do {} while (0)
421 * goal_in_my_reservation()
422 * @rsv: inode's reservation window
423 * @grp_goal: given goal block relative to the allocation block group
424 * @group: the current allocation block group
425 * @sb: filesystem super block
427 * Test if the given goal block (group relative) is within the file's
428 * own block reservation window range.
430 * If the reservation window is outside the goal allocation group, return 0;
431 * grp_goal (given goal block) could be -1, which means no specific
432 * goal block. In this case, always return 1.
433 * If the goal block is within the reservation window, return 1;
434 * otherwise, return 0;
437 goal_in_my_reservation(struct ext4_reserve_window
*rsv
, ext4_grpblk_t grp_goal
,
438 ext4_group_t group
, struct super_block
*sb
)
440 ext4_fsblk_t group_first_block
, group_last_block
;
442 group_first_block
= ext4_group_first_block_no(sb
, group
);
443 group_last_block
= group_first_block
+ (EXT4_BLOCKS_PER_GROUP(sb
) - 1);
445 if ((rsv
->_rsv_start
> group_last_block
) ||
446 (rsv
->_rsv_end
< group_first_block
))
448 if ((grp_goal
>= 0) && ((grp_goal
+ group_first_block
< rsv
->_rsv_start
)
449 || (grp_goal
+ group_first_block
> rsv
->_rsv_end
)))
455 * search_reserve_window()
456 * @rb_root: root of reservation tree
457 * @goal: target allocation block
459 * Find the reserved window which includes the goal, or the previous one
460 * if the goal is not in any window.
461 * Returns NULL if there are no windows or if all windows start after the goal.
463 static struct ext4_reserve_window_node
*
464 search_reserve_window(struct rb_root
*root
, ext4_fsblk_t goal
)
466 struct rb_node
*n
= root
->rb_node
;
467 struct ext4_reserve_window_node
*rsv
;
473 rsv
= rb_entry(n
, struct ext4_reserve_window_node
, rsv_node
);
475 if (goal
< rsv
->rsv_start
)
477 else if (goal
> rsv
->rsv_end
)
483 * We've fallen off the end of the tree: the goal wasn't inside
484 * any particular node. OK, the previous node must be to one
485 * side of the interval containing the goal. If it's the RHS,
486 * we need to back up one.
488 if (rsv
->rsv_start
> goal
) {
489 n
= rb_prev(&rsv
->rsv_node
);
490 rsv
= rb_entry(n
, struct ext4_reserve_window_node
, rsv_node
);
496 * ext4_rsv_window_add() -- Insert a window to the block reservation rb tree.
498 * @rsv: reservation window to add
500 * Must be called with rsv_lock hold.
502 void ext4_rsv_window_add(struct super_block
*sb
,
503 struct ext4_reserve_window_node
*rsv
)
505 struct rb_root
*root
= &EXT4_SB(sb
)->s_rsv_window_root
;
506 struct rb_node
*node
= &rsv
->rsv_node
;
507 ext4_fsblk_t start
= rsv
->rsv_start
;
509 struct rb_node
** p
= &root
->rb_node
;
510 struct rb_node
* parent
= NULL
;
511 struct ext4_reserve_window_node
*this;
516 this = rb_entry(parent
, struct ext4_reserve_window_node
, rsv_node
);
518 if (start
< this->rsv_start
)
520 else if (start
> this->rsv_end
)
523 rsv_window_dump(root
, 1);
528 rb_link_node(node
, parent
, p
);
529 rb_insert_color(node
, root
);
533 * ext4_rsv_window_remove() -- unlink a window from the reservation rb tree
535 * @rsv: reservation window to remove
537 * Mark the block reservation window as not allocated, and unlink it
538 * from the filesystem reservation window rb tree. Must be called with
541 static void rsv_window_remove(struct super_block
*sb
,
542 struct ext4_reserve_window_node
*rsv
)
544 rsv
->rsv_start
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
545 rsv
->rsv_end
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
546 rsv
->rsv_alloc_hit
= 0;
547 rb_erase(&rsv
->rsv_node
, &EXT4_SB(sb
)->s_rsv_window_root
);
551 * rsv_is_empty() -- Check if the reservation window is allocated.
552 * @rsv: given reservation window to check
554 * returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED.
556 static inline int rsv_is_empty(struct ext4_reserve_window
*rsv
)
558 /* a valid reservation end block could not be 0 */
559 return rsv
->_rsv_end
== EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
563 * ext4_init_block_alloc_info()
564 * @inode: file inode structure
566 * Allocate and initialize the reservation window structure, and
567 * link the window to the ext4 inode structure at last
569 * The reservation window structure is only dynamically allocated
570 * and linked to ext4 inode the first time the open file
571 * needs a new block. So, before every ext4_new_block(s) call, for
572 * regular files, we should check whether the reservation window
573 * structure exists or not. In the latter case, this function is called.
574 * Fail to do so will result in block reservation being turned off for that
577 * This function is called from ext4_get_blocks_handle(), also called
578 * when setting the reservation window size through ioctl before the file
579 * is open for write (needs block allocation).
581 * Needs down_write(i_data_sem) protection prior to call this function.
583 void ext4_init_block_alloc_info(struct inode
*inode
)
585 struct ext4_inode_info
*ei
= EXT4_I(inode
);
586 struct ext4_block_alloc_info
*block_i
= ei
->i_block_alloc_info
;
587 struct super_block
*sb
= inode
->i_sb
;
589 block_i
= kmalloc(sizeof(*block_i
), GFP_NOFS
);
591 struct ext4_reserve_window_node
*rsv
= &block_i
->rsv_window_node
;
593 rsv
->rsv_start
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
594 rsv
->rsv_end
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
597 * if filesystem is mounted with NORESERVATION, the goal
598 * reservation window size is set to zero to indicate
599 * block reservation is off
601 if (!test_opt(sb
, RESERVATION
))
602 rsv
->rsv_goal_size
= 0;
604 rsv
->rsv_goal_size
= EXT4_DEFAULT_RESERVE_BLOCKS
;
605 rsv
->rsv_alloc_hit
= 0;
606 block_i
->last_alloc_logical_block
= 0;
607 block_i
->last_alloc_physical_block
= 0;
609 ei
->i_block_alloc_info
= block_i
;
613 * ext4_discard_reservation()
616 * Discard(free) block reservation window on last file close, or truncate
619 * It is being called in three cases:
620 * ext4_release_file(): last writer close the file
621 * ext4_clear_inode(): last iput(), when nobody link to this file.
622 * ext4_truncate(): when the block indirect map is about to change.
625 void ext4_discard_reservation(struct inode
*inode
)
627 struct ext4_inode_info
*ei
= EXT4_I(inode
);
628 struct ext4_block_alloc_info
*block_i
= ei
->i_block_alloc_info
;
629 struct ext4_reserve_window_node
*rsv
;
630 spinlock_t
*rsv_lock
= &EXT4_SB(inode
->i_sb
)->s_rsv_window_lock
;
632 ext4_mb_discard_inode_preallocations(inode
);
637 rsv
= &block_i
->rsv_window_node
;
638 if (!rsv_is_empty(&rsv
->rsv_window
)) {
640 if (!rsv_is_empty(&rsv
->rsv_window
))
641 rsv_window_remove(inode
->i_sb
, rsv
);
642 spin_unlock(rsv_lock
);
647 * ext4_free_blocks_sb() -- Free given blocks and update quota
648 * @handle: handle to this transaction
650 * @block: start physcial block to free
651 * @count: number of blocks to free
652 * @pdquot_freed_blocks: pointer to quota
654 void ext4_free_blocks_sb(handle_t
*handle
, struct super_block
*sb
,
655 ext4_fsblk_t block
, unsigned long count
,
656 unsigned long *pdquot_freed_blocks
)
658 struct buffer_head
*bitmap_bh
= NULL
;
659 struct buffer_head
*gd_bh
;
660 ext4_group_t block_group
;
663 unsigned long overflow
;
664 struct ext4_group_desc
* desc
;
665 struct ext4_super_block
* es
;
666 struct ext4_sb_info
*sbi
;
668 ext4_grpblk_t group_freed
;
670 *pdquot_freed_blocks
= 0;
673 if (block
< le32_to_cpu(es
->s_first_data_block
) ||
674 block
+ count
< block
||
675 block
+ count
> ext4_blocks_count(es
)) {
676 ext4_error (sb
, "ext4_free_blocks",
677 "Freeing blocks not in datazone - "
678 "block = %llu, count = %lu", block
, count
);
682 ext4_debug ("freeing block(s) %llu-%llu\n", block
, block
+ count
- 1);
686 ext4_get_group_no_and_offset(sb
, block
, &block_group
, &bit
);
688 * Check to see if we are freeing blocks across a group
691 if (bit
+ count
> EXT4_BLOCKS_PER_GROUP(sb
)) {
692 overflow
= bit
+ count
- EXT4_BLOCKS_PER_GROUP(sb
);
696 bitmap_bh
= ext4_read_block_bitmap(sb
, block_group
);
699 desc
= ext4_get_group_desc (sb
, block_group
, &gd_bh
);
703 if (in_range(ext4_block_bitmap(sb
, desc
), block
, count
) ||
704 in_range(ext4_inode_bitmap(sb
, desc
), block
, count
) ||
705 in_range(block
, ext4_inode_table(sb
, desc
), sbi
->s_itb_per_group
) ||
706 in_range(block
+ count
- 1, ext4_inode_table(sb
, desc
),
707 sbi
->s_itb_per_group
)) {
708 ext4_error (sb
, "ext4_free_blocks",
709 "Freeing blocks in system zones - "
710 "Block = %llu, count = %lu",
716 * We are about to start releasing blocks in the bitmap,
717 * so we need undo access.
719 /* @@@ check errors */
720 BUFFER_TRACE(bitmap_bh
, "getting undo access");
721 err
= ext4_journal_get_undo_access(handle
, bitmap_bh
);
726 * We are about to modify some metadata. Call the journal APIs
727 * to unshare ->b_data if a currently-committing transaction is
730 BUFFER_TRACE(gd_bh
, "get_write_access");
731 err
= ext4_journal_get_write_access(handle
, gd_bh
);
735 jbd_lock_bh_state(bitmap_bh
);
737 for (i
= 0, group_freed
= 0; i
< count
; i
++) {
739 * An HJ special. This is expensive...
741 #ifdef CONFIG_JBD2_DEBUG
742 jbd_unlock_bh_state(bitmap_bh
);
744 struct buffer_head
*debug_bh
;
745 debug_bh
= sb_find_get_block(sb
, block
+ i
);
747 BUFFER_TRACE(debug_bh
, "Deleted!");
748 if (!bh2jh(bitmap_bh
)->b_committed_data
)
749 BUFFER_TRACE(debug_bh
,
750 "No commited data in bitmap");
751 BUFFER_TRACE2(debug_bh
, bitmap_bh
, "bitmap");
755 jbd_lock_bh_state(bitmap_bh
);
757 if (need_resched()) {
758 jbd_unlock_bh_state(bitmap_bh
);
760 jbd_lock_bh_state(bitmap_bh
);
762 /* @@@ This prevents newly-allocated data from being
763 * freed and then reallocated within the same
766 * Ideally we would want to allow that to happen, but to
767 * do so requires making jbd2_journal_forget() capable of
768 * revoking the queued write of a data block, which
769 * implies blocking on the journal lock. *forget()
770 * cannot block due to truncate races.
772 * Eventually we can fix this by making jbd2_journal_forget()
773 * return a status indicating whether or not it was able
774 * to revoke the buffer. On successful revoke, it is
775 * safe not to set the allocation bit in the committed
776 * bitmap, because we know that there is no outstanding
777 * activity on the buffer any more and so it is safe to
780 BUFFER_TRACE(bitmap_bh
, "set in b_committed_data");
781 J_ASSERT_BH(bitmap_bh
,
782 bh2jh(bitmap_bh
)->b_committed_data
!= NULL
);
783 ext4_set_bit_atomic(sb_bgl_lock(sbi
, block_group
), bit
+ i
,
784 bh2jh(bitmap_bh
)->b_committed_data
);
787 * We clear the bit in the bitmap after setting the committed
788 * data bit, because this is the reverse order to that which
789 * the allocator uses.
791 BUFFER_TRACE(bitmap_bh
, "clear bit");
792 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi
, block_group
),
793 bit
+ i
, bitmap_bh
->b_data
)) {
794 jbd_unlock_bh_state(bitmap_bh
);
795 ext4_error(sb
, __func__
,
796 "bit already cleared for block %llu",
797 (ext4_fsblk_t
)(block
+ i
));
798 jbd_lock_bh_state(bitmap_bh
);
799 BUFFER_TRACE(bitmap_bh
, "bit already cleared");
804 jbd_unlock_bh_state(bitmap_bh
);
806 spin_lock(sb_bgl_lock(sbi
, block_group
));
807 le16_add_cpu(&desc
->bg_free_blocks_count
, group_freed
);
808 desc
->bg_checksum
= ext4_group_desc_csum(sbi
, block_group
, desc
);
809 spin_unlock(sb_bgl_lock(sbi
, block_group
));
810 percpu_counter_add(&sbi
->s_freeblocks_counter
, count
);
812 /* We dirtied the bitmap block */
813 BUFFER_TRACE(bitmap_bh
, "dirtied bitmap block");
814 err
= ext4_journal_dirty_metadata(handle
, bitmap_bh
);
816 /* And the group descriptor block */
817 BUFFER_TRACE(gd_bh
, "dirtied group descriptor block");
818 ret
= ext4_journal_dirty_metadata(handle
, gd_bh
);
820 *pdquot_freed_blocks
+= group_freed
;
822 if (overflow
&& !err
) {
830 ext4_std_error(sb
, err
);
835 * ext4_free_blocks() -- Free given blocks and update quota
836 * @handle: handle for this transaction
838 * @block: start physical block to free
839 * @count: number of blocks to count
840 * @metadata: Are these metadata blocks
842 void ext4_free_blocks(handle_t
*handle
, struct inode
*inode
,
843 ext4_fsblk_t block
, unsigned long count
,
846 struct super_block
* sb
;
847 unsigned long dquot_freed_blocks
;
849 /* this isn't the right place to decide whether block is metadata
850 * inode.c/extents.c knows better, but for safety ... */
851 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
) ||
852 ext4_should_journal_data(inode
))
857 if (!test_opt(sb
, MBALLOC
) || !EXT4_SB(sb
)->s_group_info
)
858 ext4_free_blocks_sb(handle
, sb
, block
, count
,
859 &dquot_freed_blocks
);
861 ext4_mb_free_blocks(handle
, inode
, block
, count
,
862 metadata
, &dquot_freed_blocks
);
863 if (dquot_freed_blocks
)
864 DQUOT_FREE_BLOCK(inode
, dquot_freed_blocks
);
869 * ext4_test_allocatable()
870 * @nr: given allocation block group
871 * @bh: bufferhead contains the bitmap of the given block group
873 * For ext4 allocations, we must not reuse any blocks which are
874 * allocated in the bitmap buffer's "last committed data" copy. This
875 * prevents deletes from freeing up the page for reuse until we have
876 * committed the delete transaction.
878 * If we didn't do this, then deleting something and reallocating it as
879 * data would allow the old block to be overwritten before the
880 * transaction committed (because we force data to disk before commit).
881 * This would lead to corruption if we crashed between overwriting the
882 * data and committing the delete.
884 * @@@ We may want to make this allocation behaviour conditional on
885 * data-writes at some point, and disable it for metadata allocations or
888 static int ext4_test_allocatable(ext4_grpblk_t nr
, struct buffer_head
*bh
)
891 struct journal_head
*jh
= bh2jh(bh
);
893 if (ext4_test_bit(nr
, bh
->b_data
))
896 jbd_lock_bh_state(bh
);
897 if (!jh
->b_committed_data
)
900 ret
= !ext4_test_bit(nr
, jh
->b_committed_data
);
901 jbd_unlock_bh_state(bh
);
906 * bitmap_search_next_usable_block()
907 * @start: the starting block (group relative) of the search
908 * @bh: bufferhead contains the block group bitmap
909 * @maxblocks: the ending block (group relative) of the reservation
911 * The bitmap search --- search forward alternately through the actual
912 * bitmap on disk and the last-committed copy in journal, until we find a
913 * bit free in both bitmaps.
916 bitmap_search_next_usable_block(ext4_grpblk_t start
, struct buffer_head
*bh
,
917 ext4_grpblk_t maxblocks
)
920 struct journal_head
*jh
= bh2jh(bh
);
922 while (start
< maxblocks
) {
923 next
= ext4_find_next_zero_bit(bh
->b_data
, maxblocks
, start
);
924 if (next
>= maxblocks
)
926 if (ext4_test_allocatable(next
, bh
))
928 jbd_lock_bh_state(bh
);
929 if (jh
->b_committed_data
)
930 start
= ext4_find_next_zero_bit(jh
->b_committed_data
,
932 jbd_unlock_bh_state(bh
);
938 * find_next_usable_block()
939 * @start: the starting block (group relative) to find next
940 * allocatable block in bitmap.
941 * @bh: bufferhead contains the block group bitmap
942 * @maxblocks: the ending block (group relative) for the search
944 * Find an allocatable block in a bitmap. We honor both the bitmap and
945 * its last-committed copy (if that exists), and perform the "most
946 * appropriate allocation" algorithm of looking for a free block near
947 * the initial goal; then for a free byte somewhere in the bitmap; then
948 * for any free bit in the bitmap.
951 find_next_usable_block(ext4_grpblk_t start
, struct buffer_head
*bh
,
952 ext4_grpblk_t maxblocks
)
954 ext4_grpblk_t here
, next
;
959 * The goal was occupied; search forward for a free
960 * block within the next XX blocks.
962 * end_goal is more or less random, but it has to be
963 * less than EXT4_BLOCKS_PER_GROUP. Aligning up to the
964 * next 64-bit boundary is simple..
966 ext4_grpblk_t end_goal
= (start
+ 63) & ~63;
967 if (end_goal
> maxblocks
)
968 end_goal
= maxblocks
;
969 here
= ext4_find_next_zero_bit(bh
->b_data
, end_goal
, start
);
970 if (here
< end_goal
&& ext4_test_allocatable(here
, bh
))
972 ext4_debug("Bit not found near goal\n");
979 p
= ((char *)bh
->b_data
) + (here
>> 3);
980 r
= memscan(p
, 0, ((maxblocks
+ 7) >> 3) - (here
>> 3));
981 next
= (r
- ((char *)bh
->b_data
)) << 3;
983 if (next
< maxblocks
&& next
>= start
&& ext4_test_allocatable(next
, bh
))
987 * The bitmap search --- search forward alternately through the actual
988 * bitmap and the last-committed copy until we find a bit free in
991 here
= bitmap_search_next_usable_block(here
, bh
, maxblocks
);
997 * @block: the free block (group relative) to allocate
998 * @bh: the bufferhead containts the block group bitmap
1000 * We think we can allocate this block in this bitmap. Try to set the bit.
1001 * If that succeeds then check that nobody has allocated and then freed the
1002 * block since we saw that is was not marked in b_committed_data. If it _was_
1003 * allocated and freed then clear the bit in the bitmap again and return
1007 claim_block(spinlock_t
*lock
, ext4_grpblk_t block
, struct buffer_head
*bh
)
1009 struct journal_head
*jh
= bh2jh(bh
);
1012 if (ext4_set_bit_atomic(lock
, block
, bh
->b_data
))
1014 jbd_lock_bh_state(bh
);
1015 if (jh
->b_committed_data
&& ext4_test_bit(block
,jh
->b_committed_data
)) {
1016 ext4_clear_bit_atomic(lock
, block
, bh
->b_data
);
1021 jbd_unlock_bh_state(bh
);
1026 * ext4_try_to_allocate()
1028 * @handle: handle to this transaction
1029 * @group: given allocation block group
1030 * @bitmap_bh: bufferhead holds the block bitmap
1031 * @grp_goal: given target block within the group
1032 * @count: target number of blocks to allocate
1033 * @my_rsv: reservation window
1035 * Attempt to allocate blocks within a give range. Set the range of allocation
1036 * first, then find the first free bit(s) from the bitmap (within the range),
1037 * and at last, allocate the blocks by claiming the found free bit as allocated.
1039 * To set the range of this allocation:
1040 * if there is a reservation window, only try to allocate block(s) from the
1041 * file's own reservation window;
1042 * Otherwise, the allocation range starts from the give goal block, ends at
1043 * the block group's last block.
1045 * If we failed to allocate the desired block then we may end up crossing to a
1046 * new bitmap. In that case we must release write access to the old one via
1047 * ext4_journal_release_buffer(), else we'll run out of credits.
1049 static ext4_grpblk_t
1050 ext4_try_to_allocate(struct super_block
*sb
, handle_t
*handle
,
1051 ext4_group_t group
, struct buffer_head
*bitmap_bh
,
1052 ext4_grpblk_t grp_goal
, unsigned long *count
,
1053 struct ext4_reserve_window
*my_rsv
)
1055 ext4_fsblk_t group_first_block
;
1056 ext4_grpblk_t start
, end
;
1057 unsigned long num
= 0;
1059 /* we do allocation within the reservation window if we have a window */
1061 group_first_block
= ext4_group_first_block_no(sb
, group
);
1062 if (my_rsv
->_rsv_start
>= group_first_block
)
1063 start
= my_rsv
->_rsv_start
- group_first_block
;
1065 /* reservation window cross group boundary */
1067 end
= my_rsv
->_rsv_end
- group_first_block
+ 1;
1068 if (end
> EXT4_BLOCKS_PER_GROUP(sb
))
1069 /* reservation window crosses group boundary */
1070 end
= EXT4_BLOCKS_PER_GROUP(sb
);
1071 if ((start
<= grp_goal
) && (grp_goal
< end
))
1080 end
= EXT4_BLOCKS_PER_GROUP(sb
);
1083 BUG_ON(start
> EXT4_BLOCKS_PER_GROUP(sb
));
1086 if (grp_goal
< 0 || !ext4_test_allocatable(grp_goal
, bitmap_bh
)) {
1087 grp_goal
= find_next_usable_block(start
, bitmap_bh
, end
);
1093 for (i
= 0; i
< 7 && grp_goal
> start
&&
1094 ext4_test_allocatable(grp_goal
- 1,
1102 if (!claim_block(sb_bgl_lock(EXT4_SB(sb
), group
),
1103 grp_goal
, bitmap_bh
)) {
1105 * The block was allocated by another thread, or it was
1106 * allocated and then freed by another thread
1116 while (num
< *count
&& grp_goal
< end
1117 && ext4_test_allocatable(grp_goal
, bitmap_bh
)
1118 && claim_block(sb_bgl_lock(EXT4_SB(sb
), group
),
1119 grp_goal
, bitmap_bh
)) {
1124 return grp_goal
- num
;
1131 * find_next_reservable_window():
1132 * find a reservable space within the given range.
1133 * It does not allocate the reservation window for now:
1134 * alloc_new_reservation() will do the work later.
1136 * @search_head: the head of the searching list;
1137 * This is not necessarily the list head of the whole filesystem
1139 * We have both head and start_block to assist the search
1140 * for the reservable space. The list starts from head,
1141 * but we will shift to the place where start_block is,
1142 * then start from there, when looking for a reservable space.
1144 * @size: the target new reservation window size
1146 * @group_first_block: the first block we consider to start
1147 * the real search from
1150 * the maximum block number that our goal reservable space
1151 * could start from. This is normally the last block in this
1152 * group. The search will end when we found the start of next
1153 * possible reservable space is out of this boundary.
1154 * This could handle the cross boundary reservation window
1157 * basically we search from the given range, rather than the whole
1158 * reservation double linked list, (start_block, last_block)
1159 * to find a free region that is of my size and has not
1163 static int find_next_reservable_window(
1164 struct ext4_reserve_window_node
*search_head
,
1165 struct ext4_reserve_window_node
*my_rsv
,
1166 struct super_block
* sb
,
1167 ext4_fsblk_t start_block
,
1168 ext4_fsblk_t last_block
)
1170 struct rb_node
*next
;
1171 struct ext4_reserve_window_node
*rsv
, *prev
;
1173 int size
= my_rsv
->rsv_goal_size
;
1175 /* TODO: make the start of the reservation window byte-aligned */
1176 /* cur = *start_block & ~7;*/
1183 if (cur
<= rsv
->rsv_end
)
1184 cur
= rsv
->rsv_end
+ 1;
1187 * in the case we could not find a reservable space
1188 * that is what is expected, during the re-search, we could
1189 * remember what's the largest reservable space we could have
1190 * and return that one.
1192 * For now it will fail if we could not find the reservable
1193 * space with expected-size (or more)...
1195 if (cur
> last_block
)
1196 return -1; /* fail */
1199 next
= rb_next(&rsv
->rsv_node
);
1200 rsv
= rb_entry(next
,struct ext4_reserve_window_node
,rsv_node
);
1203 * Reached the last reservation, we can just append to the
1209 if (cur
+ size
<= rsv
->rsv_start
) {
1211 * Found a reserveable space big enough. We could
1212 * have a reservation across the group boundary here
1218 * we come here either :
1219 * when we reach the end of the whole list,
1220 * and there is empty reservable space after last entry in the list.
1221 * append it to the end of the list.
1223 * or we found one reservable space in the middle of the list,
1224 * return the reservation window that we could append to.
1228 if ((prev
!= my_rsv
) && (!rsv_is_empty(&my_rsv
->rsv_window
)))
1229 rsv_window_remove(sb
, my_rsv
);
1232 * Let's book the whole avaliable window for now. We will check the
1233 * disk bitmap later and then, if there are free blocks then we adjust
1234 * the window size if it's larger than requested.
1235 * Otherwise, we will remove this node from the tree next time
1236 * call find_next_reservable_window.
1238 my_rsv
->rsv_start
= cur
;
1239 my_rsv
->rsv_end
= cur
+ size
- 1;
1240 my_rsv
->rsv_alloc_hit
= 0;
1243 ext4_rsv_window_add(sb
, my_rsv
);
1249 * alloc_new_reservation()--allocate a new reservation window
1251 * To make a new reservation, we search part of the filesystem
1252 * reservation list (the list that inside the group). We try to
1253 * allocate a new reservation window near the allocation goal,
1254 * or the beginning of the group, if there is no goal.
1256 * We first find a reservable space after the goal, then from
1257 * there, we check the bitmap for the first free block after
1258 * it. If there is no free block until the end of group, then the
1259 * whole group is full, we failed. Otherwise, check if the free
1260 * block is inside the expected reservable space, if so, we
1262 * If the first free block is outside the reservable space, then
1263 * start from the first free block, we search for next available
1266 * on succeed, a new reservation will be found and inserted into the list
1267 * It contains at least one free block, and it does not overlap with other
1268 * reservation windows.
1270 * failed: we failed to find a reservation window in this group
1272 * @rsv: the reservation
1274 * @grp_goal: The goal (group-relative). It is where the search for a
1275 * free reservable space should start from.
1276 * if we have a grp_goal(grp_goal >0 ), then start from there,
1277 * no grp_goal(grp_goal = -1), we start from the first block
1280 * @sb: the super block
1281 * @group: the group we are trying to allocate in
1282 * @bitmap_bh: the block group block bitmap
1285 static int alloc_new_reservation(struct ext4_reserve_window_node
*my_rsv
,
1286 ext4_grpblk_t grp_goal
, struct super_block
*sb
,
1287 ext4_group_t group
, struct buffer_head
*bitmap_bh
)
1289 struct ext4_reserve_window_node
*search_head
;
1290 ext4_fsblk_t group_first_block
, group_end_block
, start_block
;
1291 ext4_grpblk_t first_free_block
;
1292 struct rb_root
*fs_rsv_root
= &EXT4_SB(sb
)->s_rsv_window_root
;
1295 spinlock_t
*rsv_lock
= &EXT4_SB(sb
)->s_rsv_window_lock
;
1297 group_first_block
= ext4_group_first_block_no(sb
, group
);
1298 group_end_block
= group_first_block
+ (EXT4_BLOCKS_PER_GROUP(sb
) - 1);
1301 start_block
= group_first_block
;
1303 start_block
= grp_goal
+ group_first_block
;
1305 size
= my_rsv
->rsv_goal_size
;
1307 if (!rsv_is_empty(&my_rsv
->rsv_window
)) {
1309 * if the old reservation is cross group boundary
1310 * and if the goal is inside the old reservation window,
1311 * we will come here when we just failed to allocate from
1312 * the first part of the window. We still have another part
1313 * that belongs to the next group. In this case, there is no
1314 * point to discard our window and try to allocate a new one
1315 * in this group(which will fail). we should
1316 * keep the reservation window, just simply move on.
1318 * Maybe we could shift the start block of the reservation
1319 * window to the first block of next group.
1322 if ((my_rsv
->rsv_start
<= group_end_block
) &&
1323 (my_rsv
->rsv_end
> group_end_block
) &&
1324 (start_block
>= my_rsv
->rsv_start
))
1327 if ((my_rsv
->rsv_alloc_hit
>
1328 (my_rsv
->rsv_end
- my_rsv
->rsv_start
+ 1) / 2)) {
1330 * if the previously allocation hit ratio is
1331 * greater than 1/2, then we double the size of
1332 * the reservation window the next time,
1333 * otherwise we keep the same size window
1336 if (size
> EXT4_MAX_RESERVE_BLOCKS
)
1337 size
= EXT4_MAX_RESERVE_BLOCKS
;
1338 my_rsv
->rsv_goal_size
= size
;
1342 spin_lock(rsv_lock
);
1344 * shift the search start to the window near the goal block
1346 search_head
= search_reserve_window(fs_rsv_root
, start_block
);
1349 * find_next_reservable_window() simply finds a reservable window
1350 * inside the given range(start_block, group_end_block).
1352 * To make sure the reservation window has a free bit inside it, we
1353 * need to check the bitmap after we found a reservable window.
1356 ret
= find_next_reservable_window(search_head
, my_rsv
, sb
,
1357 start_block
, group_end_block
);
1360 if (!rsv_is_empty(&my_rsv
->rsv_window
))
1361 rsv_window_remove(sb
, my_rsv
);
1362 spin_unlock(rsv_lock
);
1367 * On success, find_next_reservable_window() returns the
1368 * reservation window where there is a reservable space after it.
1369 * Before we reserve this reservable space, we need
1370 * to make sure there is at least a free block inside this region.
1372 * searching the first free bit on the block bitmap and copy of
1373 * last committed bitmap alternatively, until we found a allocatable
1374 * block. Search start from the start block of the reservable space
1377 spin_unlock(rsv_lock
);
1378 first_free_block
= bitmap_search_next_usable_block(
1379 my_rsv
->rsv_start
- group_first_block
,
1380 bitmap_bh
, group_end_block
- group_first_block
+ 1);
1382 if (first_free_block
< 0) {
1384 * no free block left on the bitmap, no point
1385 * to reserve the space. return failed.
1387 spin_lock(rsv_lock
);
1388 if (!rsv_is_empty(&my_rsv
->rsv_window
))
1389 rsv_window_remove(sb
, my_rsv
);
1390 spin_unlock(rsv_lock
);
1391 return -1; /* failed */
1394 start_block
= first_free_block
+ group_first_block
;
1396 * check if the first free block is within the
1397 * free space we just reserved
1399 if (start_block
>= my_rsv
->rsv_start
&& start_block
<= my_rsv
->rsv_end
)
1400 return 0; /* success */
1402 * if the first free bit we found is out of the reservable space
1403 * continue search for next reservable space,
1404 * start from where the free block is,
1405 * we also shift the list head to where we stopped last time
1407 search_head
= my_rsv
;
1408 spin_lock(rsv_lock
);
1413 * try_to_extend_reservation()
1414 * @my_rsv: given reservation window
1416 * @size: the delta to extend
1418 * Attempt to expand the reservation window large enough to have
1419 * required number of free blocks
1421 * Since ext4_try_to_allocate() will always allocate blocks within
1422 * the reservation window range, if the window size is too small,
1423 * multiple blocks allocation has to stop at the end of the reservation
1424 * window. To make this more efficient, given the total number of
1425 * blocks needed and the current size of the window, we try to
1426 * expand the reservation window size if necessary on a best-effort
1427 * basis before ext4_new_blocks() tries to allocate blocks,
1429 static void try_to_extend_reservation(struct ext4_reserve_window_node
*my_rsv
,
1430 struct super_block
*sb
, int size
)
1432 struct ext4_reserve_window_node
*next_rsv
;
1433 struct rb_node
*next
;
1434 spinlock_t
*rsv_lock
= &EXT4_SB(sb
)->s_rsv_window_lock
;
1436 if (!spin_trylock(rsv_lock
))
1439 next
= rb_next(&my_rsv
->rsv_node
);
1442 my_rsv
->rsv_end
+= size
;
1444 next_rsv
= rb_entry(next
, struct ext4_reserve_window_node
, rsv_node
);
1446 if ((next_rsv
->rsv_start
- my_rsv
->rsv_end
- 1) >= size
)
1447 my_rsv
->rsv_end
+= size
;
1449 my_rsv
->rsv_end
= next_rsv
->rsv_start
- 1;
1451 spin_unlock(rsv_lock
);
1455 * ext4_try_to_allocate_with_rsv()
1457 * @handle: handle to this transaction
1458 * @group: given allocation block group
1459 * @bitmap_bh: bufferhead holds the block bitmap
1460 * @grp_goal: given target block within the group
1461 * @count: target number of blocks to allocate
1462 * @my_rsv: reservation window
1463 * @errp: pointer to store the error code
1465 * This is the main function used to allocate a new block and its reservation
1468 * Each time when a new block allocation is need, first try to allocate from
1469 * its own reservation. If it does not have a reservation window, instead of
1470 * looking for a free bit on bitmap first, then look up the reservation list to
1471 * see if it is inside somebody else's reservation window, we try to allocate a
1472 * reservation window for it starting from the goal first. Then do the block
1473 * allocation within the reservation window.
1475 * This will avoid keeping on searching the reservation list again and
1476 * again when somebody is looking for a free block (without
1477 * reservation), and there are lots of free blocks, but they are all
1480 * We use a red-black tree for the per-filesystem reservation list.
1483 static ext4_grpblk_t
1484 ext4_try_to_allocate_with_rsv(struct super_block
*sb
, handle_t
*handle
,
1485 ext4_group_t group
, struct buffer_head
*bitmap_bh
,
1486 ext4_grpblk_t grp_goal
,
1487 struct ext4_reserve_window_node
* my_rsv
,
1488 unsigned long *count
, int *errp
)
1490 ext4_fsblk_t group_first_block
, group_last_block
;
1491 ext4_grpblk_t ret
= 0;
1493 unsigned long num
= *count
;
1498 * Make sure we use undo access for the bitmap, because it is critical
1499 * that we do the frozen_data COW on bitmap buffers in all cases even
1500 * if the buffer is in BJ_Forget state in the committing transaction.
1502 BUFFER_TRACE(bitmap_bh
, "get undo access for new block");
1503 fatal
= ext4_journal_get_undo_access(handle
, bitmap_bh
);
1510 * we don't deal with reservation when
1511 * filesystem is mounted without reservation
1512 * or the file is not a regular file
1513 * or last attempt to allocate a block with reservation turned on failed
1515 if (my_rsv
== NULL
) {
1516 ret
= ext4_try_to_allocate(sb
, handle
, group
, bitmap_bh
,
1517 grp_goal
, count
, NULL
);
1521 * grp_goal is a group relative block number (if there is a goal)
1522 * 0 <= grp_goal < EXT4_BLOCKS_PER_GROUP(sb)
1523 * first block is a filesystem wide block number
1524 * first block is the block number of the first block in this group
1526 group_first_block
= ext4_group_first_block_no(sb
, group
);
1527 group_last_block
= group_first_block
+ (EXT4_BLOCKS_PER_GROUP(sb
) - 1);
1530 * Basically we will allocate a new block from inode's reservation
1533 * We need to allocate a new reservation window, if:
1534 * a) inode does not have a reservation window; or
1535 * b) last attempt to allocate a block from existing reservation
1537 * c) we come here with a goal and with a reservation window
1539 * We do not need to allocate a new reservation window if we come here
1540 * at the beginning with a goal and the goal is inside the window, or
1541 * we don't have a goal but already have a reservation window.
1542 * then we could go to allocate from the reservation window directly.
1545 if (rsv_is_empty(&my_rsv
->rsv_window
) || (ret
< 0) ||
1546 !goal_in_my_reservation(&my_rsv
->rsv_window
,
1547 grp_goal
, group
, sb
)) {
1548 if (my_rsv
->rsv_goal_size
< *count
)
1549 my_rsv
->rsv_goal_size
= *count
;
1550 ret
= alloc_new_reservation(my_rsv
, grp_goal
, sb
,
1555 if (!goal_in_my_reservation(&my_rsv
->rsv_window
,
1556 grp_goal
, group
, sb
))
1558 } else if (grp_goal
>= 0) {
1559 int curr
= my_rsv
->rsv_end
-
1560 (grp_goal
+ group_first_block
) + 1;
1563 try_to_extend_reservation(my_rsv
, sb
,
1567 if ((my_rsv
->rsv_start
> group_last_block
) ||
1568 (my_rsv
->rsv_end
< group_first_block
)) {
1569 rsv_window_dump(&EXT4_SB(sb
)->s_rsv_window_root
, 1);
1572 ret
= ext4_try_to_allocate(sb
, handle
, group
, bitmap_bh
,
1573 grp_goal
, &num
, &my_rsv
->rsv_window
);
1575 my_rsv
->rsv_alloc_hit
+= num
;
1577 break; /* succeed */
1583 BUFFER_TRACE(bitmap_bh
, "journal_dirty_metadata for "
1585 fatal
= ext4_journal_dirty_metadata(handle
, bitmap_bh
);
1593 BUFFER_TRACE(bitmap_bh
, "journal_release_buffer");
1594 ext4_journal_release_buffer(handle
, bitmap_bh
);
1599 * ext4_has_free_blocks()
1600 * @sbi: in-core super block structure.
1602 * Check if filesystem has at least 1 free block available for allocation.
1604 static int ext4_has_free_blocks(struct ext4_sb_info
*sbi
)
1606 ext4_fsblk_t free_blocks
, root_blocks
;
1608 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
1609 root_blocks
= ext4_r_blocks_count(sbi
->s_es
);
1610 if (free_blocks
< root_blocks
+ 1 && !capable(CAP_SYS_RESOURCE
) &&
1611 sbi
->s_resuid
!= current
->fsuid
&&
1612 (sbi
->s_resgid
== 0 || !in_group_p (sbi
->s_resgid
))) {
1619 * ext4_should_retry_alloc()
1621 * @retries number of attemps has been made
1623 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
1624 * it is profitable to retry the operation, this function will wait
1625 * for the current or commiting transaction to complete, and then
1628 * if the total number of retries exceed three times, return FALSE.
1630 int ext4_should_retry_alloc(struct super_block
*sb
, int *retries
)
1632 if (!ext4_has_free_blocks(EXT4_SB(sb
)) || (*retries
)++ > 3)
1635 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb
->s_id
);
1637 return jbd2_journal_force_commit_nested(EXT4_SB(sb
)->s_journal
);
1641 * ext4_new_blocks_old() -- core block(s) allocation function
1642 * @handle: handle to this transaction
1643 * @inode: file inode
1644 * @goal: given target block(filesystem wide)
1645 * @count: target number of blocks to allocate
1648 * ext4_new_blocks uses a goal block to assist allocation. It tries to
1649 * allocate block(s) from the block group contains the goal block first. If that
1650 * fails, it will try to allocate block(s) from other block groups without
1651 * any specific goal block.
1654 ext4_fsblk_t
ext4_new_blocks_old(handle_t
*handle
, struct inode
*inode
,
1655 ext4_fsblk_t goal
, unsigned long *count
, int *errp
)
1657 struct buffer_head
*bitmap_bh
= NULL
;
1658 struct buffer_head
*gdp_bh
;
1659 ext4_group_t group_no
;
1660 ext4_group_t goal_group
;
1661 ext4_grpblk_t grp_target_blk
; /* blockgroup relative goal block */
1662 ext4_grpblk_t grp_alloc_blk
; /* blockgroup-relative allocated block*/
1663 ext4_fsblk_t ret_block
; /* filesyetem-wide allocated block */
1664 ext4_group_t bgi
; /* blockgroup iteration index */
1666 int performed_allocation
= 0;
1667 ext4_grpblk_t free_blocks
; /* number of free blocks in a group */
1668 struct super_block
*sb
;
1669 struct ext4_group_desc
*gdp
;
1670 struct ext4_super_block
*es
;
1671 struct ext4_sb_info
*sbi
;
1672 struct ext4_reserve_window_node
*my_rsv
= NULL
;
1673 struct ext4_block_alloc_info
*block_i
;
1674 unsigned short windowsz
= 0;
1675 ext4_group_t ngroups
;
1676 unsigned long num
= *count
;
1681 printk("ext4_new_block: nonexistent device");
1686 * Check quota for allocation of this block.
1688 if (DQUOT_ALLOC_BLOCK(inode
, num
)) {
1694 es
= EXT4_SB(sb
)->s_es
;
1695 ext4_debug("goal=%llu.\n", goal
);
1697 * Allocate a block from reservation only when
1698 * filesystem is mounted with reservation(default,-o reservation), and
1699 * it's a regular file, and
1700 * the desired window size is greater than 0 (One could use ioctl
1701 * command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off
1702 * reservation on that particular file)
1704 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
1705 if (block_i
&& ((windowsz
= block_i
->rsv_window_node
.rsv_goal_size
) > 0))
1706 my_rsv
= &block_i
->rsv_window_node
;
1708 if (!ext4_has_free_blocks(sbi
)) {
1714 * First, test whether the goal block is free.
1716 if (goal
< le32_to_cpu(es
->s_first_data_block
) ||
1717 goal
>= ext4_blocks_count(es
))
1718 goal
= le32_to_cpu(es
->s_first_data_block
);
1719 ext4_get_group_no_and_offset(sb
, goal
, &group_no
, &grp_target_blk
);
1720 goal_group
= group_no
;
1722 gdp
= ext4_get_group_desc(sb
, group_no
, &gdp_bh
);
1726 free_blocks
= le16_to_cpu(gdp
->bg_free_blocks_count
);
1728 * if there is not enough free blocks to make a new resevation
1729 * turn off reservation for this allocation
1731 if (my_rsv
&& (free_blocks
< windowsz
)
1732 && (rsv_is_empty(&my_rsv
->rsv_window
)))
1735 if (free_blocks
> 0) {
1736 bitmap_bh
= ext4_read_block_bitmap(sb
, group_no
);
1739 grp_alloc_blk
= ext4_try_to_allocate_with_rsv(sb
, handle
,
1740 group_no
, bitmap_bh
, grp_target_blk
,
1741 my_rsv
, &num
, &fatal
);
1744 if (grp_alloc_blk
>= 0)
1748 ngroups
= EXT4_SB(sb
)->s_groups_count
;
1752 * Now search the rest of the groups. We assume that
1753 * group_no and gdp correctly point to the last group visited.
1755 for (bgi
= 0; bgi
< ngroups
; bgi
++) {
1757 if (group_no
>= ngroups
)
1759 gdp
= ext4_get_group_desc(sb
, group_no
, &gdp_bh
);
1762 free_blocks
= le16_to_cpu(gdp
->bg_free_blocks_count
);
1764 * skip this group if the number of
1765 * free blocks is less than half of the reservation
1768 if (free_blocks
<= (windowsz
/2))
1772 bitmap_bh
= ext4_read_block_bitmap(sb
, group_no
);
1776 * try to allocate block(s) from this group, without a goal(-1).
1778 grp_alloc_blk
= ext4_try_to_allocate_with_rsv(sb
, handle
,
1779 group_no
, bitmap_bh
, -1, my_rsv
,
1783 if (grp_alloc_blk
>= 0)
1787 * We may end up a bogus ealier ENOSPC error due to
1788 * filesystem is "full" of reservations, but
1789 * there maybe indeed free blocks avaliable on disk
1790 * In this case, we just forget about the reservations
1791 * just do block allocation as without reservations.
1796 group_no
= goal_group
;
1799 /* No space left on the device */
1805 ext4_debug("using block group %lu(%d)\n",
1806 group_no
, gdp
->bg_free_blocks_count
);
1808 BUFFER_TRACE(gdp_bh
, "get_write_access");
1809 fatal
= ext4_journal_get_write_access(handle
, gdp_bh
);
1813 ret_block
= grp_alloc_blk
+ ext4_group_first_block_no(sb
, group_no
);
1815 if (in_range(ext4_block_bitmap(sb
, gdp
), ret_block
, num
) ||
1816 in_range(ext4_inode_bitmap(sb
, gdp
), ret_block
, num
) ||
1817 in_range(ret_block
, ext4_inode_table(sb
, gdp
),
1818 EXT4_SB(sb
)->s_itb_per_group
) ||
1819 in_range(ret_block
+ num
- 1, ext4_inode_table(sb
, gdp
),
1820 EXT4_SB(sb
)->s_itb_per_group
)) {
1821 ext4_error(sb
, "ext4_new_block",
1822 "Allocating block in system zone - "
1823 "blocks from %llu, length %lu",
1826 * claim_block marked the blocks we allocated
1827 * as in use. So we may want to selectively
1828 * mark some of the blocks as free
1833 performed_allocation
= 1;
1835 #ifdef CONFIG_JBD2_DEBUG
1837 struct buffer_head
*debug_bh
;
1839 /* Record bitmap buffer state in the newly allocated block */
1840 debug_bh
= sb_find_get_block(sb
, ret_block
);
1842 BUFFER_TRACE(debug_bh
, "state when allocated");
1843 BUFFER_TRACE2(debug_bh
, bitmap_bh
, "bitmap state");
1847 jbd_lock_bh_state(bitmap_bh
);
1848 spin_lock(sb_bgl_lock(sbi
, group_no
));
1849 if (buffer_jbd(bitmap_bh
) && bh2jh(bitmap_bh
)->b_committed_data
) {
1852 for (i
= 0; i
< num
; i
++) {
1853 if (ext4_test_bit(grp_alloc_blk
+i
,
1854 bh2jh(bitmap_bh
)->b_committed_data
)) {
1855 printk("%s: block was unexpectedly set in "
1856 "b_committed_data\n", __func__
);
1860 ext4_debug("found bit %d\n", grp_alloc_blk
);
1861 spin_unlock(sb_bgl_lock(sbi
, group_no
));
1862 jbd_unlock_bh_state(bitmap_bh
);
1865 if (ret_block
+ num
- 1 >= ext4_blocks_count(es
)) {
1866 ext4_error(sb
, "ext4_new_block",
1867 "block(%llu) >= blocks count(%llu) - "
1868 "block_group = %lu, es == %p ", ret_block
,
1869 ext4_blocks_count(es
), group_no
, es
);
1874 * It is up to the caller to add the new buffer to a journal
1875 * list of some description. We don't know in advance whether
1876 * the caller wants to use it as metadata or data.
1878 spin_lock(sb_bgl_lock(sbi
, group_no
));
1879 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
))
1880 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT
);
1881 le16_add_cpu(&gdp
->bg_free_blocks_count
, -num
);
1882 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group_no
, gdp
);
1883 spin_unlock(sb_bgl_lock(sbi
, group_no
));
1884 percpu_counter_sub(&sbi
->s_freeblocks_counter
, num
);
1886 BUFFER_TRACE(gdp_bh
, "journal_dirty_metadata for group descriptor");
1887 err
= ext4_journal_dirty_metadata(handle
, gdp_bh
);
1897 DQUOT_FREE_BLOCK(inode
, *count
-num
);
1906 ext4_std_error(sb
, fatal
);
1909 * Undo the block allocation
1911 if (!performed_allocation
)
1912 DQUOT_FREE_BLOCK(inode
, *count
);
1917 ext4_fsblk_t
ext4_new_block(handle_t
*handle
, struct inode
*inode
,
1918 ext4_fsblk_t goal
, int *errp
)
1920 struct ext4_allocation_request ar
;
1923 if (!test_opt(inode
->i_sb
, MBALLOC
)) {
1924 unsigned long count
= 1;
1925 ret
= ext4_new_blocks_old(handle
, inode
, goal
, &count
, errp
);
1929 memset(&ar
, 0, sizeof(ar
));
1933 ret
= ext4_mb_new_blocks(handle
, &ar
, errp
);
1937 ext4_fsblk_t
ext4_new_blocks(handle_t
*handle
, struct inode
*inode
,
1938 ext4_fsblk_t goal
, unsigned long *count
, int *errp
)
1940 struct ext4_allocation_request ar
;
1943 if (!test_opt(inode
->i_sb
, MBALLOC
)) {
1944 ret
= ext4_new_blocks_old(handle
, inode
, goal
, count
, errp
);
1948 memset(&ar
, 0, sizeof(ar
));
1952 ret
= ext4_mb_new_blocks(handle
, &ar
, errp
);
1959 * ext4_count_free_blocks() -- count filesystem free blocks
1962 * Adds up the number of free blocks from each block group.
1964 ext4_fsblk_t
ext4_count_free_blocks(struct super_block
*sb
)
1966 ext4_fsblk_t desc_count
;
1967 struct ext4_group_desc
*gdp
;
1969 ext4_group_t ngroups
= EXT4_SB(sb
)->s_groups_count
;
1971 struct ext4_super_block
*es
;
1972 ext4_fsblk_t bitmap_count
;
1974 struct buffer_head
*bitmap_bh
= NULL
;
1976 es
= EXT4_SB(sb
)->s_es
;
1982 for (i
= 0; i
< ngroups
; i
++) {
1983 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1986 desc_count
+= le16_to_cpu(gdp
->bg_free_blocks_count
);
1988 bitmap_bh
= ext4_read_block_bitmap(sb
, i
);
1989 if (bitmap_bh
== NULL
)
1992 x
= ext4_count_free(bitmap_bh
, sb
->s_blocksize
);
1993 printk(KERN_DEBUG
"group %lu: stored = %d, counted = %lu\n",
1994 i
, le16_to_cpu(gdp
->bg_free_blocks_count
), x
);
1998 printk("ext4_count_free_blocks: stored = %llu"
1999 ", computed = %llu, %llu\n",
2000 ext4_free_blocks_count(es
),
2001 desc_count
, bitmap_count
);
2002 return bitmap_count
;
2006 for (i
= 0; i
< ngroups
; i
++) {
2007 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
2010 desc_count
+= le16_to_cpu(gdp
->bg_free_blocks_count
);
2017 static inline int test_root(ext4_group_t a
, int b
)
2026 static int ext4_group_sparse(ext4_group_t group
)
2032 return (test_root(group
, 7) || test_root(group
, 5) ||
2033 test_root(group
, 3));
2037 * ext4_bg_has_super - number of blocks used by the superblock in group
2038 * @sb: superblock for filesystem
2039 * @group: group number to check
2041 * Return the number of blocks used by the superblock (primary or backup)
2042 * in this group. Currently this will be only 0 or 1.
2044 int ext4_bg_has_super(struct super_block
*sb
, ext4_group_t group
)
2046 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
2047 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
) &&
2048 !ext4_group_sparse(group
))
2053 static unsigned long ext4_bg_num_gdb_meta(struct super_block
*sb
,
2056 unsigned long metagroup
= group
/ EXT4_DESC_PER_BLOCK(sb
);
2057 ext4_group_t first
= metagroup
* EXT4_DESC_PER_BLOCK(sb
);
2058 ext4_group_t last
= first
+ EXT4_DESC_PER_BLOCK(sb
) - 1;
2060 if (group
== first
|| group
== first
+ 1 || group
== last
)
2065 static unsigned long ext4_bg_num_gdb_nometa(struct super_block
*sb
,
2068 return ext4_bg_has_super(sb
, group
) ? EXT4_SB(sb
)->s_gdb_count
: 0;
2072 * ext4_bg_num_gdb - number of blocks used by the group table in group
2073 * @sb: superblock for filesystem
2074 * @group: group number to check
2076 * Return the number of blocks used by the group descriptor table
2077 * (primary or backup) in this group. In the future there may be a
2078 * different number of descriptor blocks in each group.
2080 unsigned long ext4_bg_num_gdb(struct super_block
*sb
, ext4_group_t group
)
2082 unsigned long first_meta_bg
=
2083 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_meta_bg
);
2084 unsigned long metagroup
= group
/ EXT4_DESC_PER_BLOCK(sb
);
2086 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
,EXT4_FEATURE_INCOMPAT_META_BG
) ||
2087 metagroup
< first_meta_bg
)
2088 return ext4_bg_num_gdb_nometa(sb
,group
);
2090 return ext4_bg_num_gdb_meta(sb
,group
);