2 * linux/fs/ext4/balloc.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
9 * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
10 * Big-endian to little-endian byte-swapping/bitmaps by
11 * David S. Miller (davem@caip.rutgers.edu), 1995
14 #include <linux/time.h>
15 #include <linux/capability.h>
17 #include <linux/jbd2.h>
18 #include <linux/ext4_fs.h>
19 #include <linux/ext4_jbd2.h>
20 #include <linux/quotaops.h>
21 #include <linux/buffer_head.h>
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 /* Initializes an uninitialized block bitmap if given, and returns the
47 * number of blocks free in the group. */
48 unsigned ext4_init_block_bitmap(struct super_block
*sb
, struct buffer_head
*bh
,
49 ext4_group_t block_group
, struct ext4_group_desc
*gdp
)
53 unsigned free_blocks
, group_blocks
;
54 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
57 J_ASSERT_BH(bh
, buffer_locked(bh
));
59 /* If checksum is bad mark all blocks used to prevent allocation
60 * essentially implementing a per-group read-only flag. */
61 if (!ext4_group_desc_csum_verify(sbi
, block_group
, gdp
)) {
62 ext4_error(sb
, __FUNCTION__
,
63 "Checksum bad for group %lu\n", block_group
);
64 gdp
->bg_free_blocks_count
= 0;
65 gdp
->bg_free_inodes_count
= 0;
66 gdp
->bg_itable_unused
= 0;
67 memset(bh
->b_data
, 0xff, sb
->s_blocksize
);
70 memset(bh
->b_data
, 0, sb
->s_blocksize
);
73 /* Check for superblock and gdt backups in this group */
74 bit_max
= ext4_bg_has_super(sb
, block_group
);
76 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_META_BG
) ||
77 block_group
< le32_to_cpu(sbi
->s_es
->s_first_meta_bg
) *
78 sbi
->s_desc_per_block
) {
80 bit_max
+= ext4_bg_num_gdb(sb
, block_group
);
82 le16_to_cpu(sbi
->s_es
->s_reserved_gdt_blocks
);
84 } else { /* For META_BG_BLOCK_GROUPS */
85 int group_rel
= (block_group
-
86 le32_to_cpu(sbi
->s_es
->s_first_meta_bg
)) %
87 EXT4_DESC_PER_BLOCK(sb
);
88 if (group_rel
== 0 || group_rel
== 1 ||
89 (group_rel
== EXT4_DESC_PER_BLOCK(sb
) - 1))
93 if (block_group
== sbi
->s_groups_count
- 1) {
95 * Even though mke2fs always initialize first and last group
96 * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
97 * to make sure we calculate the right free blocks
99 group_blocks
= ext4_blocks_count(sbi
->s_es
) -
100 le32_to_cpu(sbi
->s_es
->s_first_data_block
) -
101 (EXT4_BLOCKS_PER_GROUP(sb
) * (sbi
->s_groups_count
-1));
103 group_blocks
= EXT4_BLOCKS_PER_GROUP(sb
);
106 free_blocks
= group_blocks
- bit_max
;
109 for (bit
= 0; bit
< bit_max
; bit
++)
110 ext4_set_bit(bit
, bh
->b_data
);
112 start
= block_group
* EXT4_BLOCKS_PER_GROUP(sb
) +
113 le32_to_cpu(sbi
->s_es
->s_first_data_block
);
115 /* Set bits for block and inode bitmaps, and inode table */
116 ext4_set_bit(ext4_block_bitmap(sb
, gdp
) - start
, bh
->b_data
);
117 ext4_set_bit(ext4_inode_bitmap(sb
, gdp
) - start
, bh
->b_data
);
118 for (bit
= (ext4_inode_table(sb
, gdp
) - start
),
119 bit_max
= bit
+ sbi
->s_itb_per_group
; bit
< bit_max
; bit
++)
120 ext4_set_bit(bit
, bh
->b_data
);
123 * Also if the number of blocks within the group is
124 * less than the blocksize * 8 ( which is the size
125 * of bitmap ), set rest of the block bitmap to 1
127 mark_bitmap_end(group_blocks
, sb
->s_blocksize
* 8, bh
->b_data
);
130 return free_blocks
- sbi
->s_itb_per_group
- 2;
135 * The free blocks are managed by bitmaps. A file system contains several
136 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
137 * block for inodes, N blocks for the inode table and data blocks.
139 * The file system contains group descriptors which are located after the
140 * super block. Each descriptor contains the number of the bitmap block and
141 * the free blocks count in the block. The descriptors are loaded in memory
142 * when a file system is mounted (see ext4_fill_super).
146 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
149 * ext4_get_group_desc() -- load group descriptor from disk
151 * @block_group: given block group
152 * @bh: pointer to the buffer head to store the block
155 struct ext4_group_desc
* ext4_get_group_desc(struct super_block
* sb
,
156 ext4_group_t block_group
,
157 struct buffer_head
** bh
)
159 unsigned long group_desc
;
160 unsigned long offset
;
161 struct ext4_group_desc
* desc
;
162 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
164 if (block_group
>= sbi
->s_groups_count
) {
165 ext4_error (sb
, "ext4_get_group_desc",
166 "block_group >= groups_count - "
167 "block_group = %lu, groups_count = %lu",
168 block_group
, sbi
->s_groups_count
);
174 group_desc
= block_group
>> EXT4_DESC_PER_BLOCK_BITS(sb
);
175 offset
= block_group
& (EXT4_DESC_PER_BLOCK(sb
) - 1);
176 if (!sbi
->s_group_desc
[group_desc
]) {
177 ext4_error (sb
, "ext4_get_group_desc",
178 "Group descriptor not loaded - "
179 "block_group = %lu, group_desc = %lu, desc = %lu",
180 block_group
, group_desc
, offset
);
184 desc
= (struct ext4_group_desc
*)(
185 (__u8
*)sbi
->s_group_desc
[group_desc
]->b_data
+
186 offset
* EXT4_DESC_SIZE(sb
));
188 *bh
= sbi
->s_group_desc
[group_desc
];
192 static int ext4_valid_block_bitmap(struct super_block
*sb
,
193 struct ext4_group_desc
*desc
,
194 unsigned int block_group
,
195 struct buffer_head
*bh
)
197 ext4_grpblk_t offset
;
198 ext4_grpblk_t next_zero_bit
;
199 ext4_fsblk_t bitmap_blk
;
200 ext4_fsblk_t group_first_block
;
202 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FLEX_BG
)) {
203 /* with FLEX_BG, the inode/block bitmaps and itable
204 * blocks may not be in the group at all
205 * so the bitmap validation will be skipped for those groups
206 * or it has to also read the block group where the bitmaps
207 * are located to verify they are set.
211 group_first_block
= ext4_group_first_block_no(sb
, block_group
);
213 /* check whether block bitmap block number is set */
214 bitmap_blk
= ext4_block_bitmap(sb
, desc
);
215 offset
= bitmap_blk
- group_first_block
;
216 if (!ext4_test_bit(offset
, bh
->b_data
))
217 /* bad block bitmap */
220 /* check whether the inode bitmap block number is set */
221 bitmap_blk
= ext4_inode_bitmap(sb
, desc
);
222 offset
= bitmap_blk
- group_first_block
;
223 if (!ext4_test_bit(offset
, bh
->b_data
))
224 /* bad block bitmap */
227 /* check whether the inode table block number is set */
228 bitmap_blk
= ext4_inode_table(sb
, desc
);
229 offset
= bitmap_blk
- group_first_block
;
230 next_zero_bit
= ext4_find_next_zero_bit(bh
->b_data
,
231 offset
+ EXT4_SB(sb
)->s_itb_per_group
,
233 if (next_zero_bit
>= offset
+ EXT4_SB(sb
)->s_itb_per_group
)
234 /* good bitmap for inode tables */
238 ext4_error(sb
, __FUNCTION__
,
239 "Invalid block bitmap - "
240 "block_group = %d, block = %llu",
241 block_group
, bitmap_blk
);
245 * read_block_bitmap()
247 * @block_group: given block group
249 * Read the bitmap for a given block_group,and validate the
250 * bits for block/inode/inode tables are set in the bitmaps
252 * Return buffer_head on success or NULL in case of failure.
255 read_block_bitmap(struct super_block
*sb
, ext4_group_t block_group
)
257 struct ext4_group_desc
* desc
;
258 struct buffer_head
* bh
= NULL
;
259 ext4_fsblk_t bitmap_blk
;
261 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
264 bitmap_blk
= ext4_block_bitmap(sb
, desc
);
265 bh
= sb_getblk(sb
, bitmap_blk
);
267 ext4_error(sb
, __FUNCTION__
,
268 "Cannot read block bitmap - "
269 "block_group = %d, block_bitmap = %llu",
270 (int)block_group
, (unsigned long long)bitmap_blk
);
273 if (bh_uptodate_or_lock(bh
))
276 if (desc
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
277 ext4_init_block_bitmap(sb
, bh
, block_group
, desc
);
278 set_buffer_uptodate(bh
);
282 if (bh_submit_read(bh
) < 0) {
284 ext4_error(sb
, __FUNCTION__
,
285 "Cannot read block bitmap - "
286 "block_group = %d, block_bitmap = %llu",
287 (int)block_group
, (unsigned long long)bitmap_blk
);
290 if (!ext4_valid_block_bitmap(sb
, desc
, block_group
, bh
)) {
298 * The reservation window structure operations
299 * --------------------------------------------
300 * Operations include:
301 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
303 * We use a red-black tree to represent per-filesystem reservation
309 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
310 * @rb_root: root of per-filesystem reservation rb tree
311 * @verbose: verbose mode
312 * @fn: function which wishes to dump the reservation map
314 * If verbose is turned on, it will print the whole block reservation
315 * windows(start, end). Otherwise, it will only print out the "bad" windows,
316 * those windows that overlap with their immediate neighbors.
319 static void __rsv_window_dump(struct rb_root
*root
, int verbose
,
323 struct ext4_reserve_window_node
*rsv
, *prev
;
331 printk("Block Allocation Reservation Windows Map (%s):\n", fn
);
333 rsv
= rb_entry(n
, struct ext4_reserve_window_node
, rsv_node
);
335 printk("reservation window 0x%p "
336 "start: %llu, end: %llu\n",
337 rsv
, rsv
->rsv_start
, rsv
->rsv_end
);
338 if (rsv
->rsv_start
&& rsv
->rsv_start
>= rsv
->rsv_end
) {
339 printk("Bad reservation %p (start >= end)\n",
343 if (prev
&& prev
->rsv_end
>= rsv
->rsv_start
) {
344 printk("Bad reservation %p (prev->end >= start)\n",
350 printk("Restarting reservation walk in verbose mode\n");
358 printk("Window map complete.\n");
362 #define rsv_window_dump(root, verbose) \
363 __rsv_window_dump((root), (verbose), __FUNCTION__)
365 #define rsv_window_dump(root, verbose) do {} while (0)
369 * goal_in_my_reservation()
370 * @rsv: inode's reservation window
371 * @grp_goal: given goal block relative to the allocation block group
372 * @group: the current allocation block group
373 * @sb: filesystem super block
375 * Test if the given goal block (group relative) is within the file's
376 * own block reservation window range.
378 * If the reservation window is outside the goal allocation group, return 0;
379 * grp_goal (given goal block) could be -1, which means no specific
380 * goal block. In this case, always return 1.
381 * If the goal block is within the reservation window, return 1;
382 * otherwise, return 0;
385 goal_in_my_reservation(struct ext4_reserve_window
*rsv
, ext4_grpblk_t grp_goal
,
386 ext4_group_t group
, struct super_block
*sb
)
388 ext4_fsblk_t group_first_block
, group_last_block
;
390 group_first_block
= ext4_group_first_block_no(sb
, group
);
391 group_last_block
= group_first_block
+ (EXT4_BLOCKS_PER_GROUP(sb
) - 1);
393 if ((rsv
->_rsv_start
> group_last_block
) ||
394 (rsv
->_rsv_end
< group_first_block
))
396 if ((grp_goal
>= 0) && ((grp_goal
+ group_first_block
< rsv
->_rsv_start
)
397 || (grp_goal
+ group_first_block
> rsv
->_rsv_end
)))
403 * search_reserve_window()
404 * @rb_root: root of reservation tree
405 * @goal: target allocation block
407 * Find the reserved window which includes the goal, or the previous one
408 * if the goal is not in any window.
409 * Returns NULL if there are no windows or if all windows start after the goal.
411 static struct ext4_reserve_window_node
*
412 search_reserve_window(struct rb_root
*root
, ext4_fsblk_t goal
)
414 struct rb_node
*n
= root
->rb_node
;
415 struct ext4_reserve_window_node
*rsv
;
421 rsv
= rb_entry(n
, struct ext4_reserve_window_node
, rsv_node
);
423 if (goal
< rsv
->rsv_start
)
425 else if (goal
> rsv
->rsv_end
)
431 * We've fallen off the end of the tree: the goal wasn't inside
432 * any particular node. OK, the previous node must be to one
433 * side of the interval containing the goal. If it's the RHS,
434 * we need to back up one.
436 if (rsv
->rsv_start
> goal
) {
437 n
= rb_prev(&rsv
->rsv_node
);
438 rsv
= rb_entry(n
, struct ext4_reserve_window_node
, rsv_node
);
444 * ext4_rsv_window_add() -- Insert a window to the block reservation rb tree.
446 * @rsv: reservation window to add
448 * Must be called with rsv_lock hold.
450 void ext4_rsv_window_add(struct super_block
*sb
,
451 struct ext4_reserve_window_node
*rsv
)
453 struct rb_root
*root
= &EXT4_SB(sb
)->s_rsv_window_root
;
454 struct rb_node
*node
= &rsv
->rsv_node
;
455 ext4_fsblk_t start
= rsv
->rsv_start
;
457 struct rb_node
** p
= &root
->rb_node
;
458 struct rb_node
* parent
= NULL
;
459 struct ext4_reserve_window_node
*this;
464 this = rb_entry(parent
, struct ext4_reserve_window_node
, rsv_node
);
466 if (start
< this->rsv_start
)
468 else if (start
> this->rsv_end
)
471 rsv_window_dump(root
, 1);
476 rb_link_node(node
, parent
, p
);
477 rb_insert_color(node
, root
);
481 * ext4_rsv_window_remove() -- unlink a window from the reservation rb tree
483 * @rsv: reservation window to remove
485 * Mark the block reservation window as not allocated, and unlink it
486 * from the filesystem reservation window rb tree. Must be called with
489 static void rsv_window_remove(struct super_block
*sb
,
490 struct ext4_reserve_window_node
*rsv
)
492 rsv
->rsv_start
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
493 rsv
->rsv_end
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
494 rsv
->rsv_alloc_hit
= 0;
495 rb_erase(&rsv
->rsv_node
, &EXT4_SB(sb
)->s_rsv_window_root
);
499 * rsv_is_empty() -- Check if the reservation window is allocated.
500 * @rsv: given reservation window to check
502 * returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED.
504 static inline int rsv_is_empty(struct ext4_reserve_window
*rsv
)
506 /* a valid reservation end block could not be 0 */
507 return rsv
->_rsv_end
== EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
511 * ext4_init_block_alloc_info()
512 * @inode: file inode structure
514 * Allocate and initialize the reservation window structure, and
515 * link the window to the ext4 inode structure at last
517 * The reservation window structure is only dynamically allocated
518 * and linked to ext4 inode the first time the open file
519 * needs a new block. So, before every ext4_new_block(s) call, for
520 * regular files, we should check whether the reservation window
521 * structure exists or not. In the latter case, this function is called.
522 * Fail to do so will result in block reservation being turned off for that
525 * This function is called from ext4_get_blocks_handle(), also called
526 * when setting the reservation window size through ioctl before the file
527 * is open for write (needs block allocation).
529 * Needs down_write(i_data_sem) protection prior to call this function.
531 void ext4_init_block_alloc_info(struct inode
*inode
)
533 struct ext4_inode_info
*ei
= EXT4_I(inode
);
534 struct ext4_block_alloc_info
*block_i
= ei
->i_block_alloc_info
;
535 struct super_block
*sb
= inode
->i_sb
;
537 block_i
= kmalloc(sizeof(*block_i
), GFP_NOFS
);
539 struct ext4_reserve_window_node
*rsv
= &block_i
->rsv_window_node
;
541 rsv
->rsv_start
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
542 rsv
->rsv_end
= EXT4_RESERVE_WINDOW_NOT_ALLOCATED
;
545 * if filesystem is mounted with NORESERVATION, the goal
546 * reservation window size is set to zero to indicate
547 * block reservation is off
549 if (!test_opt(sb
, RESERVATION
))
550 rsv
->rsv_goal_size
= 0;
552 rsv
->rsv_goal_size
= EXT4_DEFAULT_RESERVE_BLOCKS
;
553 rsv
->rsv_alloc_hit
= 0;
554 block_i
->last_alloc_logical_block
= 0;
555 block_i
->last_alloc_physical_block
= 0;
557 ei
->i_block_alloc_info
= block_i
;
561 * ext4_discard_reservation()
564 * Discard(free) block reservation window on last file close, or truncate
567 * It is being called in three cases:
568 * ext4_release_file(): last writer close the file
569 * ext4_clear_inode(): last iput(), when nobody link to this file.
570 * ext4_truncate(): when the block indirect map is about to change.
573 void ext4_discard_reservation(struct inode
*inode
)
575 struct ext4_inode_info
*ei
= EXT4_I(inode
);
576 struct ext4_block_alloc_info
*block_i
= ei
->i_block_alloc_info
;
577 struct ext4_reserve_window_node
*rsv
;
578 spinlock_t
*rsv_lock
= &EXT4_SB(inode
->i_sb
)->s_rsv_window_lock
;
580 ext4_mb_discard_inode_preallocations(inode
);
585 rsv
= &block_i
->rsv_window_node
;
586 if (!rsv_is_empty(&rsv
->rsv_window
)) {
588 if (!rsv_is_empty(&rsv
->rsv_window
))
589 rsv_window_remove(inode
->i_sb
, rsv
);
590 spin_unlock(rsv_lock
);
595 * ext4_free_blocks_sb() -- Free given blocks and update quota
596 * @handle: handle to this transaction
598 * @block: start physcial block to free
599 * @count: number of blocks to free
600 * @pdquot_freed_blocks: pointer to quota
602 void ext4_free_blocks_sb(handle_t
*handle
, struct super_block
*sb
,
603 ext4_fsblk_t block
, unsigned long count
,
604 unsigned long *pdquot_freed_blocks
)
606 struct buffer_head
*bitmap_bh
= NULL
;
607 struct buffer_head
*gd_bh
;
608 ext4_group_t block_group
;
611 unsigned long overflow
;
612 struct ext4_group_desc
* desc
;
613 struct ext4_super_block
* es
;
614 struct ext4_sb_info
*sbi
;
616 ext4_grpblk_t group_freed
;
618 *pdquot_freed_blocks
= 0;
621 if (block
< le32_to_cpu(es
->s_first_data_block
) ||
622 block
+ count
< block
||
623 block
+ count
> ext4_blocks_count(es
)) {
624 ext4_error (sb
, "ext4_free_blocks",
625 "Freeing blocks not in datazone - "
626 "block = %llu, count = %lu", block
, count
);
630 ext4_debug ("freeing block(s) %llu-%llu\n", block
, block
+ count
- 1);
634 ext4_get_group_no_and_offset(sb
, block
, &block_group
, &bit
);
636 * Check to see if we are freeing blocks across a group
639 if (bit
+ count
> EXT4_BLOCKS_PER_GROUP(sb
)) {
640 overflow
= bit
+ count
- EXT4_BLOCKS_PER_GROUP(sb
);
644 bitmap_bh
= read_block_bitmap(sb
, block_group
);
647 desc
= ext4_get_group_desc (sb
, block_group
, &gd_bh
);
651 if (in_range(ext4_block_bitmap(sb
, desc
), block
, count
) ||
652 in_range(ext4_inode_bitmap(sb
, desc
), block
, count
) ||
653 in_range(block
, ext4_inode_table(sb
, desc
), sbi
->s_itb_per_group
) ||
654 in_range(block
+ count
- 1, ext4_inode_table(sb
, desc
),
655 sbi
->s_itb_per_group
)) {
656 ext4_error (sb
, "ext4_free_blocks",
657 "Freeing blocks in system zones - "
658 "Block = %llu, count = %lu",
664 * We are about to start releasing blocks in the bitmap,
665 * so we need undo access.
667 /* @@@ check errors */
668 BUFFER_TRACE(bitmap_bh
, "getting undo access");
669 err
= ext4_journal_get_undo_access(handle
, bitmap_bh
);
674 * We are about to modify some metadata. Call the journal APIs
675 * to unshare ->b_data if a currently-committing transaction is
678 BUFFER_TRACE(gd_bh
, "get_write_access");
679 err
= ext4_journal_get_write_access(handle
, gd_bh
);
683 jbd_lock_bh_state(bitmap_bh
);
685 for (i
= 0, group_freed
= 0; i
< count
; i
++) {
687 * An HJ special. This is expensive...
689 #ifdef CONFIG_JBD2_DEBUG
690 jbd_unlock_bh_state(bitmap_bh
);
692 struct buffer_head
*debug_bh
;
693 debug_bh
= sb_find_get_block(sb
, block
+ i
);
695 BUFFER_TRACE(debug_bh
, "Deleted!");
696 if (!bh2jh(bitmap_bh
)->b_committed_data
)
697 BUFFER_TRACE(debug_bh
,
698 "No commited data in bitmap");
699 BUFFER_TRACE2(debug_bh
, bitmap_bh
, "bitmap");
703 jbd_lock_bh_state(bitmap_bh
);
705 if (need_resched()) {
706 jbd_unlock_bh_state(bitmap_bh
);
708 jbd_lock_bh_state(bitmap_bh
);
710 /* @@@ This prevents newly-allocated data from being
711 * freed and then reallocated within the same
714 * Ideally we would want to allow that to happen, but to
715 * do so requires making jbd2_journal_forget() capable of
716 * revoking the queued write of a data block, which
717 * implies blocking on the journal lock. *forget()
718 * cannot block due to truncate races.
720 * Eventually we can fix this by making jbd2_journal_forget()
721 * return a status indicating whether or not it was able
722 * to revoke the buffer. On successful revoke, it is
723 * safe not to set the allocation bit in the committed
724 * bitmap, because we know that there is no outstanding
725 * activity on the buffer any more and so it is safe to
728 BUFFER_TRACE(bitmap_bh
, "set in b_committed_data");
729 J_ASSERT_BH(bitmap_bh
,
730 bh2jh(bitmap_bh
)->b_committed_data
!= NULL
);
731 ext4_set_bit_atomic(sb_bgl_lock(sbi
, block_group
), bit
+ i
,
732 bh2jh(bitmap_bh
)->b_committed_data
);
735 * We clear the bit in the bitmap after setting the committed
736 * data bit, because this is the reverse order to that which
737 * the allocator uses.
739 BUFFER_TRACE(bitmap_bh
, "clear bit");
740 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi
, block_group
),
741 bit
+ i
, bitmap_bh
->b_data
)) {
742 jbd_unlock_bh_state(bitmap_bh
);
743 ext4_error(sb
, __FUNCTION__
,
744 "bit already cleared for block %llu",
745 (ext4_fsblk_t
)(block
+ i
));
746 jbd_lock_bh_state(bitmap_bh
);
747 BUFFER_TRACE(bitmap_bh
, "bit already cleared");
752 jbd_unlock_bh_state(bitmap_bh
);
754 spin_lock(sb_bgl_lock(sbi
, block_group
));
755 desc
->bg_free_blocks_count
=
756 cpu_to_le16(le16_to_cpu(desc
->bg_free_blocks_count
) +
758 desc
->bg_checksum
= ext4_group_desc_csum(sbi
, block_group
, desc
);
759 spin_unlock(sb_bgl_lock(sbi
, block_group
));
760 percpu_counter_add(&sbi
->s_freeblocks_counter
, count
);
762 /* We dirtied the bitmap block */
763 BUFFER_TRACE(bitmap_bh
, "dirtied bitmap block");
764 err
= ext4_journal_dirty_metadata(handle
, bitmap_bh
);
766 /* And the group descriptor block */
767 BUFFER_TRACE(gd_bh
, "dirtied group descriptor block");
768 ret
= ext4_journal_dirty_metadata(handle
, gd_bh
);
770 *pdquot_freed_blocks
+= group_freed
;
772 if (overflow
&& !err
) {
780 ext4_std_error(sb
, err
);
785 * ext4_free_blocks() -- Free given blocks and update quota
786 * @handle: handle for this transaction
788 * @block: start physical block to free
789 * @count: number of blocks to count
790 * @metadata: Are these metadata blocks
792 void ext4_free_blocks(handle_t
*handle
, struct inode
*inode
,
793 ext4_fsblk_t block
, unsigned long count
,
796 struct super_block
* sb
;
797 unsigned long dquot_freed_blocks
;
799 /* this isn't the right place to decide whether block is metadata
800 * inode.c/extents.c knows better, but for safety ... */
801 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
) ||
802 ext4_should_journal_data(inode
))
807 if (!test_opt(sb
, MBALLOC
) || !EXT4_SB(sb
)->s_group_info
)
808 ext4_free_blocks_sb(handle
, sb
, block
, count
,
809 &dquot_freed_blocks
);
811 ext4_mb_free_blocks(handle
, inode
, block
, count
,
812 metadata
, &dquot_freed_blocks
);
813 if (dquot_freed_blocks
)
814 DQUOT_FREE_BLOCK(inode
, dquot_freed_blocks
);
819 * ext4_test_allocatable()
820 * @nr: given allocation block group
821 * @bh: bufferhead contains the bitmap of the given block group
823 * For ext4 allocations, we must not reuse any blocks which are
824 * allocated in the bitmap buffer's "last committed data" copy. This
825 * prevents deletes from freeing up the page for reuse until we have
826 * committed the delete transaction.
828 * If we didn't do this, then deleting something and reallocating it as
829 * data would allow the old block to be overwritten before the
830 * transaction committed (because we force data to disk before commit).
831 * This would lead to corruption if we crashed between overwriting the
832 * data and committing the delete.
834 * @@@ We may want to make this allocation behaviour conditional on
835 * data-writes at some point, and disable it for metadata allocations or
838 static int ext4_test_allocatable(ext4_grpblk_t nr
, struct buffer_head
*bh
)
841 struct journal_head
*jh
= bh2jh(bh
);
843 if (ext4_test_bit(nr
, bh
->b_data
))
846 jbd_lock_bh_state(bh
);
847 if (!jh
->b_committed_data
)
850 ret
= !ext4_test_bit(nr
, jh
->b_committed_data
);
851 jbd_unlock_bh_state(bh
);
856 * bitmap_search_next_usable_block()
857 * @start: the starting block (group relative) of the search
858 * @bh: bufferhead contains the block group bitmap
859 * @maxblocks: the ending block (group relative) of the reservation
861 * The bitmap search --- search forward alternately through the actual
862 * bitmap on disk and the last-committed copy in journal, until we find a
863 * bit free in both bitmaps.
866 bitmap_search_next_usable_block(ext4_grpblk_t start
, struct buffer_head
*bh
,
867 ext4_grpblk_t maxblocks
)
870 struct journal_head
*jh
= bh2jh(bh
);
872 while (start
< maxblocks
) {
873 next
= ext4_find_next_zero_bit(bh
->b_data
, maxblocks
, start
);
874 if (next
>= maxblocks
)
876 if (ext4_test_allocatable(next
, bh
))
878 jbd_lock_bh_state(bh
);
879 if (jh
->b_committed_data
)
880 start
= ext4_find_next_zero_bit(jh
->b_committed_data
,
882 jbd_unlock_bh_state(bh
);
888 * find_next_usable_block()
889 * @start: the starting block (group relative) to find next
890 * allocatable block in bitmap.
891 * @bh: bufferhead contains the block group bitmap
892 * @maxblocks: the ending block (group relative) for the search
894 * Find an allocatable block in a bitmap. We honor both the bitmap and
895 * its last-committed copy (if that exists), and perform the "most
896 * appropriate allocation" algorithm of looking for a free block near
897 * the initial goal; then for a free byte somewhere in the bitmap; then
898 * for any free bit in the bitmap.
901 find_next_usable_block(ext4_grpblk_t start
, struct buffer_head
*bh
,
902 ext4_grpblk_t maxblocks
)
904 ext4_grpblk_t here
, next
;
909 * The goal was occupied; search forward for a free
910 * block within the next XX blocks.
912 * end_goal is more or less random, but it has to be
913 * less than EXT4_BLOCKS_PER_GROUP. Aligning up to the
914 * next 64-bit boundary is simple..
916 ext4_grpblk_t end_goal
= (start
+ 63) & ~63;
917 if (end_goal
> maxblocks
)
918 end_goal
= maxblocks
;
919 here
= ext4_find_next_zero_bit(bh
->b_data
, end_goal
, start
);
920 if (here
< end_goal
&& ext4_test_allocatable(here
, bh
))
922 ext4_debug("Bit not found near goal\n");
929 p
= ((char *)bh
->b_data
) + (here
>> 3);
930 r
= memscan(p
, 0, ((maxblocks
+ 7) >> 3) - (here
>> 3));
931 next
= (r
- ((char *)bh
->b_data
)) << 3;
933 if (next
< maxblocks
&& next
>= start
&& ext4_test_allocatable(next
, bh
))
937 * The bitmap search --- search forward alternately through the actual
938 * bitmap and the last-committed copy until we find a bit free in
941 here
= bitmap_search_next_usable_block(here
, bh
, maxblocks
);
947 * @block: the free block (group relative) to allocate
948 * @bh: the bufferhead containts the block group bitmap
950 * We think we can allocate this block in this bitmap. Try to set the bit.
951 * If that succeeds then check that nobody has allocated and then freed the
952 * block since we saw that is was not marked in b_committed_data. If it _was_
953 * allocated and freed then clear the bit in the bitmap again and return
957 claim_block(spinlock_t
*lock
, ext4_grpblk_t block
, struct buffer_head
*bh
)
959 struct journal_head
*jh
= bh2jh(bh
);
962 if (ext4_set_bit_atomic(lock
, block
, bh
->b_data
))
964 jbd_lock_bh_state(bh
);
965 if (jh
->b_committed_data
&& ext4_test_bit(block
,jh
->b_committed_data
)) {
966 ext4_clear_bit_atomic(lock
, block
, bh
->b_data
);
971 jbd_unlock_bh_state(bh
);
976 * ext4_try_to_allocate()
978 * @handle: handle to this transaction
979 * @group: given allocation block group
980 * @bitmap_bh: bufferhead holds the block bitmap
981 * @grp_goal: given target block within the group
982 * @count: target number of blocks to allocate
983 * @my_rsv: reservation window
985 * Attempt to allocate blocks within a give range. Set the range of allocation
986 * first, then find the first free bit(s) from the bitmap (within the range),
987 * and at last, allocate the blocks by claiming the found free bit as allocated.
989 * To set the range of this allocation:
990 * if there is a reservation window, only try to allocate block(s) from the
991 * file's own reservation window;
992 * Otherwise, the allocation range starts from the give goal block, ends at
993 * the block group's last block.
995 * If we failed to allocate the desired block then we may end up crossing to a
996 * new bitmap. In that case we must release write access to the old one via
997 * ext4_journal_release_buffer(), else we'll run out of credits.
1000 ext4_try_to_allocate(struct super_block
*sb
, handle_t
*handle
,
1001 ext4_group_t group
, struct buffer_head
*bitmap_bh
,
1002 ext4_grpblk_t grp_goal
, unsigned long *count
,
1003 struct ext4_reserve_window
*my_rsv
)
1005 ext4_fsblk_t group_first_block
;
1006 ext4_grpblk_t start
, end
;
1007 unsigned long num
= 0;
1009 /* we do allocation within the reservation window if we have a window */
1011 group_first_block
= ext4_group_first_block_no(sb
, group
);
1012 if (my_rsv
->_rsv_start
>= group_first_block
)
1013 start
= my_rsv
->_rsv_start
- group_first_block
;
1015 /* reservation window cross group boundary */
1017 end
= my_rsv
->_rsv_end
- group_first_block
+ 1;
1018 if (end
> EXT4_BLOCKS_PER_GROUP(sb
))
1019 /* reservation window crosses group boundary */
1020 end
= EXT4_BLOCKS_PER_GROUP(sb
);
1021 if ((start
<= grp_goal
) && (grp_goal
< end
))
1030 end
= EXT4_BLOCKS_PER_GROUP(sb
);
1033 BUG_ON(start
> EXT4_BLOCKS_PER_GROUP(sb
));
1036 if (grp_goal
< 0 || !ext4_test_allocatable(grp_goal
, bitmap_bh
)) {
1037 grp_goal
= find_next_usable_block(start
, bitmap_bh
, end
);
1043 for (i
= 0; i
< 7 && grp_goal
> start
&&
1044 ext4_test_allocatable(grp_goal
- 1,
1052 if (!claim_block(sb_bgl_lock(EXT4_SB(sb
), group
),
1053 grp_goal
, bitmap_bh
)) {
1055 * The block was allocated by another thread, or it was
1056 * allocated and then freed by another thread
1066 while (num
< *count
&& grp_goal
< end
1067 && ext4_test_allocatable(grp_goal
, bitmap_bh
)
1068 && claim_block(sb_bgl_lock(EXT4_SB(sb
), group
),
1069 grp_goal
, bitmap_bh
)) {
1074 return grp_goal
- num
;
1081 * find_next_reservable_window():
1082 * find a reservable space within the given range.
1083 * It does not allocate the reservation window for now:
1084 * alloc_new_reservation() will do the work later.
1086 * @search_head: the head of the searching list;
1087 * This is not necessarily the list head of the whole filesystem
1089 * We have both head and start_block to assist the search
1090 * for the reservable space. The list starts from head,
1091 * but we will shift to the place where start_block is,
1092 * then start from there, when looking for a reservable space.
1094 * @size: the target new reservation window size
1096 * @group_first_block: the first block we consider to start
1097 * the real search from
1100 * the maximum block number that our goal reservable space
1101 * could start from. This is normally the last block in this
1102 * group. The search will end when we found the start of next
1103 * possible reservable space is out of this boundary.
1104 * This could handle the cross boundary reservation window
1107 * basically we search from the given range, rather than the whole
1108 * reservation double linked list, (start_block, last_block)
1109 * to find a free region that is of my size and has not
1113 static int find_next_reservable_window(
1114 struct ext4_reserve_window_node
*search_head
,
1115 struct ext4_reserve_window_node
*my_rsv
,
1116 struct super_block
* sb
,
1117 ext4_fsblk_t start_block
,
1118 ext4_fsblk_t last_block
)
1120 struct rb_node
*next
;
1121 struct ext4_reserve_window_node
*rsv
, *prev
;
1123 int size
= my_rsv
->rsv_goal_size
;
1125 /* TODO: make the start of the reservation window byte-aligned */
1126 /* cur = *start_block & ~7;*/
1133 if (cur
<= rsv
->rsv_end
)
1134 cur
= rsv
->rsv_end
+ 1;
1137 * in the case we could not find a reservable space
1138 * that is what is expected, during the re-search, we could
1139 * remember what's the largest reservable space we could have
1140 * and return that one.
1142 * For now it will fail if we could not find the reservable
1143 * space with expected-size (or more)...
1145 if (cur
> last_block
)
1146 return -1; /* fail */
1149 next
= rb_next(&rsv
->rsv_node
);
1150 rsv
= rb_entry(next
,struct ext4_reserve_window_node
,rsv_node
);
1153 * Reached the last reservation, we can just append to the
1159 if (cur
+ size
<= rsv
->rsv_start
) {
1161 * Found a reserveable space big enough. We could
1162 * have a reservation across the group boundary here
1168 * we come here either :
1169 * when we reach the end of the whole list,
1170 * and there is empty reservable space after last entry in the list.
1171 * append it to the end of the list.
1173 * or we found one reservable space in the middle of the list,
1174 * return the reservation window that we could append to.
1178 if ((prev
!= my_rsv
) && (!rsv_is_empty(&my_rsv
->rsv_window
)))
1179 rsv_window_remove(sb
, my_rsv
);
1182 * Let's book the whole avaliable window for now. We will check the
1183 * disk bitmap later and then, if there are free blocks then we adjust
1184 * the window size if it's larger than requested.
1185 * Otherwise, we will remove this node from the tree next time
1186 * call find_next_reservable_window.
1188 my_rsv
->rsv_start
= cur
;
1189 my_rsv
->rsv_end
= cur
+ size
- 1;
1190 my_rsv
->rsv_alloc_hit
= 0;
1193 ext4_rsv_window_add(sb
, my_rsv
);
1199 * alloc_new_reservation()--allocate a new reservation window
1201 * To make a new reservation, we search part of the filesystem
1202 * reservation list (the list that inside the group). We try to
1203 * allocate a new reservation window near the allocation goal,
1204 * or the beginning of the group, if there is no goal.
1206 * We first find a reservable space after the goal, then from
1207 * there, we check the bitmap for the first free block after
1208 * it. If there is no free block until the end of group, then the
1209 * whole group is full, we failed. Otherwise, check if the free
1210 * block is inside the expected reservable space, if so, we
1212 * If the first free block is outside the reservable space, then
1213 * start from the first free block, we search for next available
1216 * on succeed, a new reservation will be found and inserted into the list
1217 * It contains at least one free block, and it does not overlap with other
1218 * reservation windows.
1220 * failed: we failed to find a reservation window in this group
1222 * @rsv: the reservation
1224 * @grp_goal: The goal (group-relative). It is where the search for a
1225 * free reservable space should start from.
1226 * if we have a grp_goal(grp_goal >0 ), then start from there,
1227 * no grp_goal(grp_goal = -1), we start from the first block
1230 * @sb: the super block
1231 * @group: the group we are trying to allocate in
1232 * @bitmap_bh: the block group block bitmap
1235 static int alloc_new_reservation(struct ext4_reserve_window_node
*my_rsv
,
1236 ext4_grpblk_t grp_goal
, struct super_block
*sb
,
1237 ext4_group_t group
, struct buffer_head
*bitmap_bh
)
1239 struct ext4_reserve_window_node
*search_head
;
1240 ext4_fsblk_t group_first_block
, group_end_block
, start_block
;
1241 ext4_grpblk_t first_free_block
;
1242 struct rb_root
*fs_rsv_root
= &EXT4_SB(sb
)->s_rsv_window_root
;
1245 spinlock_t
*rsv_lock
= &EXT4_SB(sb
)->s_rsv_window_lock
;
1247 group_first_block
= ext4_group_first_block_no(sb
, group
);
1248 group_end_block
= group_first_block
+ (EXT4_BLOCKS_PER_GROUP(sb
) - 1);
1251 start_block
= group_first_block
;
1253 start_block
= grp_goal
+ group_first_block
;
1255 size
= my_rsv
->rsv_goal_size
;
1257 if (!rsv_is_empty(&my_rsv
->rsv_window
)) {
1259 * if the old reservation is cross group boundary
1260 * and if the goal is inside the old reservation window,
1261 * we will come here when we just failed to allocate from
1262 * the first part of the window. We still have another part
1263 * that belongs to the next group. In this case, there is no
1264 * point to discard our window and try to allocate a new one
1265 * in this group(which will fail). we should
1266 * keep the reservation window, just simply move on.
1268 * Maybe we could shift the start block of the reservation
1269 * window to the first block of next group.
1272 if ((my_rsv
->rsv_start
<= group_end_block
) &&
1273 (my_rsv
->rsv_end
> group_end_block
) &&
1274 (start_block
>= my_rsv
->rsv_start
))
1277 if ((my_rsv
->rsv_alloc_hit
>
1278 (my_rsv
->rsv_end
- my_rsv
->rsv_start
+ 1) / 2)) {
1280 * if the previously allocation hit ratio is
1281 * greater than 1/2, then we double the size of
1282 * the reservation window the next time,
1283 * otherwise we keep the same size window
1286 if (size
> EXT4_MAX_RESERVE_BLOCKS
)
1287 size
= EXT4_MAX_RESERVE_BLOCKS
;
1288 my_rsv
->rsv_goal_size
= size
;
1292 spin_lock(rsv_lock
);
1294 * shift the search start to the window near the goal block
1296 search_head
= search_reserve_window(fs_rsv_root
, start_block
);
1299 * find_next_reservable_window() simply finds a reservable window
1300 * inside the given range(start_block, group_end_block).
1302 * To make sure the reservation window has a free bit inside it, we
1303 * need to check the bitmap after we found a reservable window.
1306 ret
= find_next_reservable_window(search_head
, my_rsv
, sb
,
1307 start_block
, group_end_block
);
1310 if (!rsv_is_empty(&my_rsv
->rsv_window
))
1311 rsv_window_remove(sb
, my_rsv
);
1312 spin_unlock(rsv_lock
);
1317 * On success, find_next_reservable_window() returns the
1318 * reservation window where there is a reservable space after it.
1319 * Before we reserve this reservable space, we need
1320 * to make sure there is at least a free block inside this region.
1322 * searching the first free bit on the block bitmap and copy of
1323 * last committed bitmap alternatively, until we found a allocatable
1324 * block. Search start from the start block of the reservable space
1327 spin_unlock(rsv_lock
);
1328 first_free_block
= bitmap_search_next_usable_block(
1329 my_rsv
->rsv_start
- group_first_block
,
1330 bitmap_bh
, group_end_block
- group_first_block
+ 1);
1332 if (first_free_block
< 0) {
1334 * no free block left on the bitmap, no point
1335 * to reserve the space. return failed.
1337 spin_lock(rsv_lock
);
1338 if (!rsv_is_empty(&my_rsv
->rsv_window
))
1339 rsv_window_remove(sb
, my_rsv
);
1340 spin_unlock(rsv_lock
);
1341 return -1; /* failed */
1344 start_block
= first_free_block
+ group_first_block
;
1346 * check if the first free block is within the
1347 * free space we just reserved
1349 if (start_block
>= my_rsv
->rsv_start
&& start_block
<= my_rsv
->rsv_end
)
1350 return 0; /* success */
1352 * if the first free bit we found is out of the reservable space
1353 * continue search for next reservable space,
1354 * start from where the free block is,
1355 * we also shift the list head to where we stopped last time
1357 search_head
= my_rsv
;
1358 spin_lock(rsv_lock
);
1363 * try_to_extend_reservation()
1364 * @my_rsv: given reservation window
1366 * @size: the delta to extend
1368 * Attempt to expand the reservation window large enough to have
1369 * required number of free blocks
1371 * Since ext4_try_to_allocate() will always allocate blocks within
1372 * the reservation window range, if the window size is too small,
1373 * multiple blocks allocation has to stop at the end of the reservation
1374 * window. To make this more efficient, given the total number of
1375 * blocks needed and the current size of the window, we try to
1376 * expand the reservation window size if necessary on a best-effort
1377 * basis before ext4_new_blocks() tries to allocate blocks,
1379 static void try_to_extend_reservation(struct ext4_reserve_window_node
*my_rsv
,
1380 struct super_block
*sb
, int size
)
1382 struct ext4_reserve_window_node
*next_rsv
;
1383 struct rb_node
*next
;
1384 spinlock_t
*rsv_lock
= &EXT4_SB(sb
)->s_rsv_window_lock
;
1386 if (!spin_trylock(rsv_lock
))
1389 next
= rb_next(&my_rsv
->rsv_node
);
1392 my_rsv
->rsv_end
+= size
;
1394 next_rsv
= rb_entry(next
, struct ext4_reserve_window_node
, rsv_node
);
1396 if ((next_rsv
->rsv_start
- my_rsv
->rsv_end
- 1) >= size
)
1397 my_rsv
->rsv_end
+= size
;
1399 my_rsv
->rsv_end
= next_rsv
->rsv_start
- 1;
1401 spin_unlock(rsv_lock
);
1405 * ext4_try_to_allocate_with_rsv()
1407 * @handle: handle to this transaction
1408 * @group: given allocation block group
1409 * @bitmap_bh: bufferhead holds the block bitmap
1410 * @grp_goal: given target block within the group
1411 * @count: target number of blocks to allocate
1412 * @my_rsv: reservation window
1413 * @errp: pointer to store the error code
1415 * This is the main function used to allocate a new block and its reservation
1418 * Each time when a new block allocation is need, first try to allocate from
1419 * its own reservation. If it does not have a reservation window, instead of
1420 * looking for a free bit on bitmap first, then look up the reservation list to
1421 * see if it is inside somebody else's reservation window, we try to allocate a
1422 * reservation window for it starting from the goal first. Then do the block
1423 * allocation within the reservation window.
1425 * This will avoid keeping on searching the reservation list again and
1426 * again when somebody is looking for a free block (without
1427 * reservation), and there are lots of free blocks, but they are all
1430 * We use a red-black tree for the per-filesystem reservation list.
1433 static ext4_grpblk_t
1434 ext4_try_to_allocate_with_rsv(struct super_block
*sb
, handle_t
*handle
,
1435 ext4_group_t group
, struct buffer_head
*bitmap_bh
,
1436 ext4_grpblk_t grp_goal
,
1437 struct ext4_reserve_window_node
* my_rsv
,
1438 unsigned long *count
, int *errp
)
1440 ext4_fsblk_t group_first_block
, group_last_block
;
1441 ext4_grpblk_t ret
= 0;
1443 unsigned long num
= *count
;
1448 * Make sure we use undo access for the bitmap, because it is critical
1449 * that we do the frozen_data COW on bitmap buffers in all cases even
1450 * if the buffer is in BJ_Forget state in the committing transaction.
1452 BUFFER_TRACE(bitmap_bh
, "get undo access for new block");
1453 fatal
= ext4_journal_get_undo_access(handle
, bitmap_bh
);
1460 * we don't deal with reservation when
1461 * filesystem is mounted without reservation
1462 * or the file is not a regular file
1463 * or last attempt to allocate a block with reservation turned on failed
1465 if (my_rsv
== NULL
) {
1466 ret
= ext4_try_to_allocate(sb
, handle
, group
, bitmap_bh
,
1467 grp_goal
, count
, NULL
);
1471 * grp_goal is a group relative block number (if there is a goal)
1472 * 0 <= grp_goal < EXT4_BLOCKS_PER_GROUP(sb)
1473 * first block is a filesystem wide block number
1474 * first block is the block number of the first block in this group
1476 group_first_block
= ext4_group_first_block_no(sb
, group
);
1477 group_last_block
= group_first_block
+ (EXT4_BLOCKS_PER_GROUP(sb
) - 1);
1480 * Basically we will allocate a new block from inode's reservation
1483 * We need to allocate a new reservation window, if:
1484 * a) inode does not have a reservation window; or
1485 * b) last attempt to allocate a block from existing reservation
1487 * c) we come here with a goal and with a reservation window
1489 * We do not need to allocate a new reservation window if we come here
1490 * at the beginning with a goal and the goal is inside the window, or
1491 * we don't have a goal but already have a reservation window.
1492 * then we could go to allocate from the reservation window directly.
1495 if (rsv_is_empty(&my_rsv
->rsv_window
) || (ret
< 0) ||
1496 !goal_in_my_reservation(&my_rsv
->rsv_window
,
1497 grp_goal
, group
, sb
)) {
1498 if (my_rsv
->rsv_goal_size
< *count
)
1499 my_rsv
->rsv_goal_size
= *count
;
1500 ret
= alloc_new_reservation(my_rsv
, grp_goal
, sb
,
1505 if (!goal_in_my_reservation(&my_rsv
->rsv_window
,
1506 grp_goal
, group
, sb
))
1508 } else if (grp_goal
>= 0) {
1509 int curr
= my_rsv
->rsv_end
-
1510 (grp_goal
+ group_first_block
) + 1;
1513 try_to_extend_reservation(my_rsv
, sb
,
1517 if ((my_rsv
->rsv_start
> group_last_block
) ||
1518 (my_rsv
->rsv_end
< group_first_block
)) {
1519 rsv_window_dump(&EXT4_SB(sb
)->s_rsv_window_root
, 1);
1522 ret
= ext4_try_to_allocate(sb
, handle
, group
, bitmap_bh
,
1523 grp_goal
, &num
, &my_rsv
->rsv_window
);
1525 my_rsv
->rsv_alloc_hit
+= num
;
1527 break; /* succeed */
1533 BUFFER_TRACE(bitmap_bh
, "journal_dirty_metadata for "
1535 fatal
= ext4_journal_dirty_metadata(handle
, bitmap_bh
);
1543 BUFFER_TRACE(bitmap_bh
, "journal_release_buffer");
1544 ext4_journal_release_buffer(handle
, bitmap_bh
);
1549 * ext4_has_free_blocks()
1550 * @sbi: in-core super block structure.
1552 * Check if filesystem has at least 1 free block available for allocation.
1554 static int ext4_has_free_blocks(struct ext4_sb_info
*sbi
)
1556 ext4_fsblk_t free_blocks
, root_blocks
;
1558 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
1559 root_blocks
= ext4_r_blocks_count(sbi
->s_es
);
1560 if (free_blocks
< root_blocks
+ 1 && !capable(CAP_SYS_RESOURCE
) &&
1561 sbi
->s_resuid
!= current
->fsuid
&&
1562 (sbi
->s_resgid
== 0 || !in_group_p (sbi
->s_resgid
))) {
1569 * ext4_should_retry_alloc()
1571 * @retries number of attemps has been made
1573 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
1574 * it is profitable to retry the operation, this function will wait
1575 * for the current or commiting transaction to complete, and then
1578 * if the total number of retries exceed three times, return FALSE.
1580 int ext4_should_retry_alloc(struct super_block
*sb
, int *retries
)
1582 if (!ext4_has_free_blocks(EXT4_SB(sb
)) || (*retries
)++ > 3)
1585 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb
->s_id
);
1587 return jbd2_journal_force_commit_nested(EXT4_SB(sb
)->s_journal
);
1591 * ext4_new_blocks_old() -- core block(s) allocation function
1592 * @handle: handle to this transaction
1593 * @inode: file inode
1594 * @goal: given target block(filesystem wide)
1595 * @count: target number of blocks to allocate
1598 * ext4_new_blocks uses a goal block to assist allocation. It tries to
1599 * allocate block(s) from the block group contains the goal block first. If that
1600 * fails, it will try to allocate block(s) from other block groups without
1601 * any specific goal block.
1604 ext4_fsblk_t
ext4_new_blocks_old(handle_t
*handle
, struct inode
*inode
,
1605 ext4_fsblk_t goal
, unsigned long *count
, int *errp
)
1607 struct buffer_head
*bitmap_bh
= NULL
;
1608 struct buffer_head
*gdp_bh
;
1609 ext4_group_t group_no
;
1610 ext4_group_t goal_group
;
1611 ext4_grpblk_t grp_target_blk
; /* blockgroup relative goal block */
1612 ext4_grpblk_t grp_alloc_blk
; /* blockgroup-relative allocated block*/
1613 ext4_fsblk_t ret_block
; /* filesyetem-wide allocated block */
1614 ext4_group_t bgi
; /* blockgroup iteration index */
1616 int performed_allocation
= 0;
1617 ext4_grpblk_t free_blocks
; /* number of free blocks in a group */
1618 struct super_block
*sb
;
1619 struct ext4_group_desc
*gdp
;
1620 struct ext4_super_block
*es
;
1621 struct ext4_sb_info
*sbi
;
1622 struct ext4_reserve_window_node
*my_rsv
= NULL
;
1623 struct ext4_block_alloc_info
*block_i
;
1624 unsigned short windowsz
= 0;
1625 ext4_group_t ngroups
;
1626 unsigned long num
= *count
;
1631 printk("ext4_new_block: nonexistent device");
1636 * Check quota for allocation of this block.
1638 if (DQUOT_ALLOC_BLOCK(inode
, num
)) {
1644 es
= EXT4_SB(sb
)->s_es
;
1645 ext4_debug("goal=%llu.\n", goal
);
1647 * Allocate a block from reservation only when
1648 * filesystem is mounted with reservation(default,-o reservation), and
1649 * it's a regular file, and
1650 * the desired window size is greater than 0 (One could use ioctl
1651 * command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off
1652 * reservation on that particular file)
1654 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
1655 if (block_i
&& ((windowsz
= block_i
->rsv_window_node
.rsv_goal_size
) > 0))
1656 my_rsv
= &block_i
->rsv_window_node
;
1658 if (!ext4_has_free_blocks(sbi
)) {
1664 * First, test whether the goal block is free.
1666 if (goal
< le32_to_cpu(es
->s_first_data_block
) ||
1667 goal
>= ext4_blocks_count(es
))
1668 goal
= le32_to_cpu(es
->s_first_data_block
);
1669 ext4_get_group_no_and_offset(sb
, goal
, &group_no
, &grp_target_blk
);
1670 goal_group
= group_no
;
1672 gdp
= ext4_get_group_desc(sb
, group_no
, &gdp_bh
);
1676 free_blocks
= le16_to_cpu(gdp
->bg_free_blocks_count
);
1678 * if there is not enough free blocks to make a new resevation
1679 * turn off reservation for this allocation
1681 if (my_rsv
&& (free_blocks
< windowsz
)
1682 && (rsv_is_empty(&my_rsv
->rsv_window
)))
1685 if (free_blocks
> 0) {
1686 bitmap_bh
= read_block_bitmap(sb
, group_no
);
1689 grp_alloc_blk
= ext4_try_to_allocate_with_rsv(sb
, handle
,
1690 group_no
, bitmap_bh
, grp_target_blk
,
1691 my_rsv
, &num
, &fatal
);
1694 if (grp_alloc_blk
>= 0)
1698 ngroups
= EXT4_SB(sb
)->s_groups_count
;
1702 * Now search the rest of the groups. We assume that
1703 * group_no and gdp correctly point to the last group visited.
1705 for (bgi
= 0; bgi
< ngroups
; bgi
++) {
1707 if (group_no
>= ngroups
)
1709 gdp
= ext4_get_group_desc(sb
, group_no
, &gdp_bh
);
1712 free_blocks
= le16_to_cpu(gdp
->bg_free_blocks_count
);
1714 * skip this group if the number of
1715 * free blocks is less than half of the reservation
1718 if (free_blocks
<= (windowsz
/2))
1722 bitmap_bh
= read_block_bitmap(sb
, group_no
);
1726 * try to allocate block(s) from this group, without a goal(-1).
1728 grp_alloc_blk
= ext4_try_to_allocate_with_rsv(sb
, handle
,
1729 group_no
, bitmap_bh
, -1, my_rsv
,
1733 if (grp_alloc_blk
>= 0)
1737 * We may end up a bogus ealier ENOSPC error due to
1738 * filesystem is "full" of reservations, but
1739 * there maybe indeed free blocks avaliable on disk
1740 * In this case, we just forget about the reservations
1741 * just do block allocation as without reservations.
1746 group_no
= goal_group
;
1749 /* No space left on the device */
1755 ext4_debug("using block group %lu(%d)\n",
1756 group_no
, gdp
->bg_free_blocks_count
);
1758 BUFFER_TRACE(gdp_bh
, "get_write_access");
1759 fatal
= ext4_journal_get_write_access(handle
, gdp_bh
);
1763 ret_block
= grp_alloc_blk
+ ext4_group_first_block_no(sb
, group_no
);
1765 if (in_range(ext4_block_bitmap(sb
, gdp
), ret_block
, num
) ||
1766 in_range(ext4_inode_bitmap(sb
, gdp
), ret_block
, num
) ||
1767 in_range(ret_block
, ext4_inode_table(sb
, gdp
),
1768 EXT4_SB(sb
)->s_itb_per_group
) ||
1769 in_range(ret_block
+ num
- 1, ext4_inode_table(sb
, gdp
),
1770 EXT4_SB(sb
)->s_itb_per_group
)) {
1771 ext4_error(sb
, "ext4_new_block",
1772 "Allocating block in system zone - "
1773 "blocks from %llu, length %lu",
1778 performed_allocation
= 1;
1780 #ifdef CONFIG_JBD2_DEBUG
1782 struct buffer_head
*debug_bh
;
1784 /* Record bitmap buffer state in the newly allocated block */
1785 debug_bh
= sb_find_get_block(sb
, ret_block
);
1787 BUFFER_TRACE(debug_bh
, "state when allocated");
1788 BUFFER_TRACE2(debug_bh
, bitmap_bh
, "bitmap state");
1792 jbd_lock_bh_state(bitmap_bh
);
1793 spin_lock(sb_bgl_lock(sbi
, group_no
));
1794 if (buffer_jbd(bitmap_bh
) && bh2jh(bitmap_bh
)->b_committed_data
) {
1797 for (i
= 0; i
< num
; i
++) {
1798 if (ext4_test_bit(grp_alloc_blk
+i
,
1799 bh2jh(bitmap_bh
)->b_committed_data
)) {
1800 printk("%s: block was unexpectedly set in "
1801 "b_committed_data\n", __FUNCTION__
);
1805 ext4_debug("found bit %d\n", grp_alloc_blk
);
1806 spin_unlock(sb_bgl_lock(sbi
, group_no
));
1807 jbd_unlock_bh_state(bitmap_bh
);
1810 if (ret_block
+ num
- 1 >= ext4_blocks_count(es
)) {
1811 ext4_error(sb
, "ext4_new_block",
1812 "block(%llu) >= blocks count(%llu) - "
1813 "block_group = %lu, es == %p ", ret_block
,
1814 ext4_blocks_count(es
), group_no
, es
);
1819 * It is up to the caller to add the new buffer to a journal
1820 * list of some description. We don't know in advance whether
1821 * the caller wants to use it as metadata or data.
1823 spin_lock(sb_bgl_lock(sbi
, group_no
));
1824 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
))
1825 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT
);
1826 gdp
->bg_free_blocks_count
=
1827 cpu_to_le16(le16_to_cpu(gdp
->bg_free_blocks_count
)-num
);
1828 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group_no
, gdp
);
1829 spin_unlock(sb_bgl_lock(sbi
, group_no
));
1830 percpu_counter_sub(&sbi
->s_freeblocks_counter
, num
);
1832 BUFFER_TRACE(gdp_bh
, "journal_dirty_metadata for group descriptor");
1833 err
= ext4_journal_dirty_metadata(handle
, gdp_bh
);
1843 DQUOT_FREE_BLOCK(inode
, *count
-num
);
1852 ext4_std_error(sb
, fatal
);
1855 * Undo the block allocation
1857 if (!performed_allocation
)
1858 DQUOT_FREE_BLOCK(inode
, *count
);
1863 ext4_fsblk_t
ext4_new_block(handle_t
*handle
, struct inode
*inode
,
1864 ext4_fsblk_t goal
, int *errp
)
1866 struct ext4_allocation_request ar
;
1869 if (!test_opt(inode
->i_sb
, MBALLOC
)) {
1870 unsigned long count
= 1;
1871 ret
= ext4_new_blocks_old(handle
, inode
, goal
, &count
, errp
);
1875 memset(&ar
, 0, sizeof(ar
));
1879 ret
= ext4_mb_new_blocks(handle
, &ar
, errp
);
1883 ext4_fsblk_t
ext4_new_blocks(handle_t
*handle
, struct inode
*inode
,
1884 ext4_fsblk_t goal
, unsigned long *count
, int *errp
)
1886 struct ext4_allocation_request ar
;
1889 if (!test_opt(inode
->i_sb
, MBALLOC
)) {
1890 ret
= ext4_new_blocks_old(handle
, inode
, goal
, count
, errp
);
1894 memset(&ar
, 0, sizeof(ar
));
1898 ret
= ext4_mb_new_blocks(handle
, &ar
, errp
);
1905 * ext4_count_free_blocks() -- count filesystem free blocks
1908 * Adds up the number of free blocks from each block group.
1910 ext4_fsblk_t
ext4_count_free_blocks(struct super_block
*sb
)
1912 ext4_fsblk_t desc_count
;
1913 struct ext4_group_desc
*gdp
;
1915 ext4_group_t ngroups
= EXT4_SB(sb
)->s_groups_count
;
1917 struct ext4_super_block
*es
;
1918 ext4_fsblk_t bitmap_count
;
1920 struct buffer_head
*bitmap_bh
= NULL
;
1922 es
= EXT4_SB(sb
)->s_es
;
1928 for (i
= 0; i
< ngroups
; i
++) {
1929 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1932 desc_count
+= le16_to_cpu(gdp
->bg_free_blocks_count
);
1934 bitmap_bh
= read_block_bitmap(sb
, i
);
1935 if (bitmap_bh
== NULL
)
1938 x
= ext4_count_free(bitmap_bh
, sb
->s_blocksize
);
1939 printk(KERN_DEBUG
"group %lu: stored = %d, counted = %lu\n",
1940 i
, le16_to_cpu(gdp
->bg_free_blocks_count
), x
);
1944 printk("ext4_count_free_blocks: stored = %llu"
1945 ", computed = %llu, %llu\n",
1946 ext4_free_blocks_count(es
),
1947 desc_count
, bitmap_count
);
1948 return bitmap_count
;
1952 for (i
= 0; i
< ngroups
; i
++) {
1953 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1956 desc_count
+= le16_to_cpu(gdp
->bg_free_blocks_count
);
1963 static inline int test_root(ext4_group_t a
, int b
)
1972 static int ext4_group_sparse(ext4_group_t group
)
1978 return (test_root(group
, 7) || test_root(group
, 5) ||
1979 test_root(group
, 3));
1983 * ext4_bg_has_super - number of blocks used by the superblock in group
1984 * @sb: superblock for filesystem
1985 * @group: group number to check
1987 * Return the number of blocks used by the superblock (primary or backup)
1988 * in this group. Currently this will be only 0 or 1.
1990 int ext4_bg_has_super(struct super_block
*sb
, ext4_group_t group
)
1992 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1993 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
) &&
1994 !ext4_group_sparse(group
))
1999 static unsigned long ext4_bg_num_gdb_meta(struct super_block
*sb
,
2002 unsigned long metagroup
= group
/ EXT4_DESC_PER_BLOCK(sb
);
2003 ext4_group_t first
= metagroup
* EXT4_DESC_PER_BLOCK(sb
);
2004 ext4_group_t last
= first
+ EXT4_DESC_PER_BLOCK(sb
) - 1;
2006 if (group
== first
|| group
== first
+ 1 || group
== last
)
2011 static unsigned long ext4_bg_num_gdb_nometa(struct super_block
*sb
,
2014 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
2015 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
) &&
2016 !ext4_group_sparse(group
))
2018 return EXT4_SB(sb
)->s_gdb_count
;
2022 * ext4_bg_num_gdb - number of blocks used by the group table in group
2023 * @sb: superblock for filesystem
2024 * @group: group number to check
2026 * Return the number of blocks used by the group descriptor table
2027 * (primary or backup) in this group. In the future there may be a
2028 * different number of descriptor blocks in each group.
2030 unsigned long ext4_bg_num_gdb(struct super_block
*sb
, ext4_group_t group
)
2032 unsigned long first_meta_bg
=
2033 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_meta_bg
);
2034 unsigned long metagroup
= group
/ EXT4_DESC_PER_BLOCK(sb
);
2036 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
,EXT4_FEATURE_INCOMPAT_META_BG
) ||
2037 metagroup
< first_meta_bg
)
2038 return ext4_bg_num_gdb_nometa(sb
,group
);
2040 return ext4_bg_num_gdb_meta(sb
,group
);