1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext2/balloc.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
10 * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller (davem@caip.rutgers.edu), 1995
16 #include <linux/quotaops.h>
17 #include <linux/slab.h>
18 #include <linux/sched.h>
19 #include <linux/cred.h>
20 #include <linux/buffer_head.h>
21 #include <linux/capability.h>
24 * balloc.c contains the blocks allocation and deallocation routines
28 * The free blocks are managed by bitmaps. A file system contains several
29 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
30 * block for inodes, N blocks for the inode table and data blocks.
32 * The file system contains group descriptors which are located after the
33 * super block. Each descriptor contains the number of the bitmap block and
34 * the free blocks count in the block. The descriptors are loaded in memory
35 * when a file system is mounted (see ext2_fill_super).
39 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
41 struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
42 unsigned int block_group,
43 struct buffer_head ** bh)
45 unsigned long group_desc;
47 struct ext2_group_desc * desc;
48 struct ext2_sb_info *sbi = EXT2_SB(sb);
50 if (block_group >= sbi->s_groups_count) {
51 WARN(1, "block_group >= groups_count - "
52 "block_group = %d, groups_count = %lu",
53 block_group, sbi->s_groups_count);
58 group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
59 offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
60 if (!sbi->s_group_desc[group_desc]) {
61 WARN(1, "Group descriptor not loaded - "
62 "block_group = %d, group_desc = %lu, desc = %lu",
63 block_group, group_desc, offset);
67 desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
69 *bh = sbi->s_group_desc[group_desc];
73 static int ext2_valid_block_bitmap(struct super_block *sb,
74 struct ext2_group_desc *desc,
75 unsigned int block_group,
76 struct buffer_head *bh)
79 ext2_grpblk_t next_zero_bit;
80 ext2_fsblk_t bitmap_blk;
81 ext2_fsblk_t group_first_block;
83 group_first_block = ext2_group_first_block_no(sb, block_group);
85 /* check whether block bitmap block number is set */
86 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
87 offset = bitmap_blk - group_first_block;
88 if (!ext2_test_bit(offset, bh->b_data))
89 /* bad block bitmap */
92 /* check whether the inode bitmap block number is set */
93 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
94 offset = bitmap_blk - group_first_block;
95 if (!ext2_test_bit(offset, bh->b_data))
96 /* bad block bitmap */
99 /* check whether the inode table block number is set */
100 bitmap_blk = le32_to_cpu(desc->bg_inode_table);
101 offset = bitmap_blk - group_first_block;
102 next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
103 offset + EXT2_SB(sb)->s_itb_per_group,
105 if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
106 /* good bitmap for inode tables */
110 ext2_error(sb, __func__,
111 "Invalid block bitmap - "
112 "block_group = %d, block = %lu",
113 block_group, bitmap_blk);
118 * Read the bitmap for a given block_group,and validate the
119 * bits for block/inode/inode tables are set in the bitmaps
121 * Return buffer_head on success or NULL in case of failure.
123 static struct buffer_head *
124 read_block_bitmap(struct super_block *sb, unsigned int block_group)
126 struct ext2_group_desc * desc;
127 struct buffer_head * bh = NULL;
128 ext2_fsblk_t bitmap_blk;
131 desc = ext2_get_group_desc(sb, block_group, NULL);
134 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
135 bh = sb_getblk(sb, bitmap_blk);
137 ext2_error(sb, __func__,
138 "Cannot read block bitmap - "
139 "block_group = %d, block_bitmap = %u",
140 block_group, le32_to_cpu(desc->bg_block_bitmap));
143 ret = bh_read(bh, 0);
148 ext2_error(sb, __func__,
149 "Cannot read block bitmap - "
150 "block_group = %d, block_bitmap = %u",
151 block_group, le32_to_cpu(desc->bg_block_bitmap));
155 ext2_valid_block_bitmap(sb, desc, block_group, bh);
157 * file system mounted not to panic on error, continue with corrupt
163 static void group_adjust_blocks(struct super_block *sb, int group_no,
164 struct ext2_group_desc *desc, struct buffer_head *bh, int count)
167 struct ext2_sb_info *sbi = EXT2_SB(sb);
168 unsigned free_blocks;
170 spin_lock(sb_bgl_lock(sbi, group_no));
171 free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
172 desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
173 spin_unlock(sb_bgl_lock(sbi, group_no));
174 mark_buffer_dirty(bh);
179 * The reservation window structure operations
180 * --------------------------------------------
181 * Operations include:
182 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
184 * We use a red-black tree to represent per-filesystem reservation
190 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
191 * @root: root of per-filesystem reservation rb tree
192 * @verbose: verbose mode
193 * @fn: function which wishes to dump the reservation map
195 * If verbose is turned on, it will print the whole block reservation
196 * windows(start, end). Otherwise, it will only print out the "bad" windows,
197 * those windows that overlap with their immediate neighbors.
200 static void __rsv_window_dump(struct rb_root *root, int verbose,
204 struct ext2_reserve_window_node *rsv, *prev;
212 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
214 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
216 printk("reservation window 0x%p "
217 "start: %lu, end: %lu\n",
218 rsv, rsv->rsv_start, rsv->rsv_end);
219 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
220 printk("Bad reservation %p (start >= end)\n",
224 if (prev && prev->rsv_end >= rsv->rsv_start) {
225 printk("Bad reservation %p (prev->end >= start)\n",
231 printk("Restarting reservation walk in verbose mode\n");
239 printk("Window map complete.\n");
242 #define rsv_window_dump(root, verbose) \
243 __rsv_window_dump((root), (verbose), __func__)
245 #define rsv_window_dump(root, verbose) do {} while (0)
249 * goal_in_my_reservation()
250 * @rsv: inode's reservation window
251 * @grp_goal: given goal block relative to the allocation block group
252 * @group: the current allocation block group
253 * @sb: filesystem super block
255 * Test if the given goal block (group relative) is within the file's
256 * own block reservation window range.
258 * If the reservation window is outside the goal allocation group, return 0;
259 * grp_goal (given goal block) could be -1, which means no specific
260 * goal block. In this case, always return 1.
261 * If the goal block is within the reservation window, return 1;
262 * otherwise, return 0;
265 goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
266 unsigned int group, struct super_block * sb)
268 ext2_fsblk_t group_first_block, group_last_block;
270 group_first_block = ext2_group_first_block_no(sb, group);
271 group_last_block = ext2_group_last_block_no(sb, group);
273 if ((rsv->_rsv_start > group_last_block) ||
274 (rsv->_rsv_end < group_first_block))
276 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
277 || (grp_goal + group_first_block > rsv->_rsv_end)))
283 * search_reserve_window()
284 * @root: root of reservation tree
285 * @goal: target allocation block
287 * Find the reserved window which includes the goal, or the previous one
288 * if the goal is not in any window.
289 * Returns NULL if there are no windows or if all windows start after the goal.
291 static struct ext2_reserve_window_node *
292 search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
294 struct rb_node *n = root->rb_node;
295 struct ext2_reserve_window_node *rsv;
301 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
303 if (goal < rsv->rsv_start)
305 else if (goal > rsv->rsv_end)
311 * We've fallen off the end of the tree: the goal wasn't inside
312 * any particular node. OK, the previous node must be to one
313 * side of the interval containing the goal. If it's the RHS,
314 * we need to back up one.
316 if (rsv->rsv_start > goal) {
317 n = rb_prev(&rsv->rsv_node);
318 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
324 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
326 * @rsv: reservation window to add
328 * Must be called with rsv_lock held.
330 void ext2_rsv_window_add(struct super_block *sb,
331 struct ext2_reserve_window_node *rsv)
333 struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
334 struct rb_node *node = &rsv->rsv_node;
335 ext2_fsblk_t start = rsv->rsv_start;
337 struct rb_node ** p = &root->rb_node;
338 struct rb_node * parent = NULL;
339 struct ext2_reserve_window_node *this;
344 this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
346 if (start < this->rsv_start)
348 else if (start > this->rsv_end)
351 rsv_window_dump(root, 1);
356 rb_link_node(node, parent, p);
357 rb_insert_color(node, root);
361 * rsv_window_remove() -- unlink a window from the reservation rb tree
363 * @rsv: reservation window to remove
365 * Mark the block reservation window as not allocated, and unlink it
366 * from the filesystem reservation window rb tree. Must be called with
369 static void rsv_window_remove(struct super_block *sb,
370 struct ext2_reserve_window_node *rsv)
372 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
373 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
374 rsv->rsv_alloc_hit = 0;
375 rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
379 * rsv_is_empty() -- Check if the reservation window is allocated.
380 * @rsv: given reservation window to check
382 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
384 static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
386 /* a valid reservation end block could not be 0 */
387 return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
391 * ext2_init_block_alloc_info()
392 * @inode: file inode structure
394 * Allocate and initialize the reservation window structure, and
395 * link the window to the ext2 inode structure at last
397 * The reservation window structure is only dynamically allocated
398 * and linked to ext2 inode the first time the open file
399 * needs a new block. So, before every ext2_new_block(s) call, for
400 * regular files, we should check whether the reservation window
401 * structure exists or not. In the latter case, this function is called.
402 * Fail to do so will result in block reservation being turned off for that
405 * This function is called from ext2_get_blocks_handle(), also called
406 * when setting the reservation window size through ioctl before the file
407 * is open for write (needs block allocation).
409 * Needs truncate_mutex protection prior to calling this function.
411 void ext2_init_block_alloc_info(struct inode *inode)
413 struct ext2_inode_info *ei = EXT2_I(inode);
414 struct ext2_block_alloc_info *block_i;
415 struct super_block *sb = inode->i_sb;
417 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
419 struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
421 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
422 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
425 * if filesystem is mounted with NORESERVATION, the goal
426 * reservation window size is set to zero to indicate
427 * block reservation is off
429 if (!test_opt(sb, RESERVATION))
430 rsv->rsv_goal_size = 0;
432 rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
433 rsv->rsv_alloc_hit = 0;
434 block_i->last_alloc_logical_block = 0;
435 block_i->last_alloc_physical_block = 0;
437 ei->i_block_alloc_info = block_i;
441 * ext2_discard_reservation()
444 * Discard(free) block reservation window on last file close, or truncate
447 * It is being called in three cases:
448 * ext2_release_file(): last writer closes the file
449 * ext2_clear_inode(): last iput(), when nobody links to this file.
450 * ext2_truncate(): when the block indirect map is about to change.
452 void ext2_discard_reservation(struct inode *inode)
454 struct ext2_inode_info *ei = EXT2_I(inode);
455 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
456 struct ext2_reserve_window_node *rsv;
457 spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
462 rsv = &block_i->rsv_window_node;
463 if (!rsv_is_empty(&rsv->rsv_window)) {
465 if (!rsv_is_empty(&rsv->rsv_window))
466 rsv_window_remove(inode->i_sb, rsv);
467 spin_unlock(rsv_lock);
472 * ext2_free_blocks() -- Free given blocks and update quota and i_blocks
474 * @block: start physical block to free
475 * @count: number of blocks to free
477 void ext2_free_blocks(struct inode * inode, ext2_fsblk_t block,
480 struct buffer_head *bitmap_bh = NULL;
481 struct buffer_head * bh2;
482 unsigned long block_group;
485 unsigned long overflow;
486 struct super_block * sb = inode->i_sb;
487 struct ext2_sb_info * sbi = EXT2_SB(sb);
488 struct ext2_group_desc * desc;
489 struct ext2_super_block * es = sbi->s_es;
490 unsigned freed = 0, group_freed;
492 if (!ext2_data_block_valid(sbi, block, count)) {
493 ext2_error (sb, "ext2_free_blocks",
494 "Freeing blocks not in datazone - "
495 "block = %lu, count = %lu", block, count);
499 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
503 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
504 EXT2_BLOCKS_PER_GROUP(sb);
505 bit = (block - le32_to_cpu(es->s_first_data_block)) %
506 EXT2_BLOCKS_PER_GROUP(sb);
508 * Check to see if we are freeing blocks across a group
511 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
512 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
516 bitmap_bh = read_block_bitmap(sb, block_group);
520 desc = ext2_get_group_desc (sb, block_group, &bh2);
524 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
525 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
526 in_range (block, le32_to_cpu(desc->bg_inode_table),
527 sbi->s_itb_per_group) ||
528 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
529 sbi->s_itb_per_group)) {
530 ext2_error (sb, "ext2_free_blocks",
531 "Freeing blocks in system zones - "
532 "Block = %lu, count = %lu",
537 for (i = 0, group_freed = 0; i < count; i++) {
538 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
539 bit + i, bitmap_bh->b_data)) {
540 ext2_error(sb, __func__,
541 "bit already cleared for block %lu", block + i);
547 mark_buffer_dirty(bitmap_bh);
548 if (sb->s_flags & SB_SYNCHRONOUS)
549 sync_dirty_buffer(bitmap_bh);
551 group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
552 freed += group_freed;
562 percpu_counter_add(&sbi->s_freeblocks_counter, freed);
563 dquot_free_block_nodirty(inode, freed);
564 mark_inode_dirty(inode);
569 * bitmap_search_next_usable_block()
570 * @start: the starting block (group relative) of the search
571 * @bh: bufferhead contains the block group bitmap
572 * @maxblocks: the ending block (group relative) of the reservation
574 * The bitmap search --- search forward through the actual bitmap on disk until
575 * we find a bit free.
578 bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
579 ext2_grpblk_t maxblocks)
583 next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
584 if (next >= maxblocks)
590 * find_next_usable_block()
591 * @start: the starting block (group relative) to find next
592 * allocatable block in bitmap.
593 * @bh: bufferhead contains the block group bitmap
594 * @maxblocks: the ending block (group relative) for the search
596 * Find an allocatable block in a bitmap. We perform the "most
597 * appropriate allocation" algorithm of looking for a free block near
598 * the initial goal; then for a free byte somewhere in the bitmap;
599 * then for any free bit in the bitmap.
602 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
604 ext2_grpblk_t here, next;
609 * The goal was occupied; search forward for a free
610 * block within the next XX blocks.
612 * end_goal is more or less random, but it has to be
613 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
614 * next 64-bit boundary is simple..
616 ext2_grpblk_t end_goal = (start + 63) & ~63;
617 if (end_goal > maxblocks)
618 end_goal = maxblocks;
619 here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
622 ext2_debug("Bit not found near goal\n");
629 p = ((char *)bh->b_data) + (here >> 3);
630 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
631 next = (r - ((char *)bh->b_data)) << 3;
633 if (next < maxblocks && next >= here)
636 here = bitmap_search_next_usable_block(here, bh, maxblocks);
641 * ext2_try_to_allocate()
643 * @group: given allocation block group
644 * @bitmap_bh: bufferhead holds the block bitmap
645 * @grp_goal: given target block within the group
646 * @count: target number of blocks to allocate
647 * @my_rsv: reservation window
649 * Attempt to allocate blocks within a give range. Set the range of allocation
650 * first, then find the first free bit(s) from the bitmap (within the range),
651 * and at last, allocate the blocks by claiming the found free bit as allocated.
653 * To set the range of this allocation:
654 * if there is a reservation window, only try to allocate block(s)
655 * from the file's own reservation window;
656 * Otherwise, the allocation range starts from the give goal block,
657 * ends at the block group's last block.
659 * If we failed to allocate the desired block then we may end up crossing to a
663 ext2_try_to_allocate(struct super_block *sb, int group,
664 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
665 unsigned long *count,
666 struct ext2_reserve_window *my_rsv)
668 ext2_fsblk_t group_first_block = ext2_group_first_block_no(sb, group);
669 ext2_fsblk_t group_last_block = ext2_group_last_block_no(sb, group);
670 ext2_grpblk_t start, end;
671 unsigned long num = 0;
674 end = group_last_block - group_first_block + 1;
675 /* we do allocation within the reservation window if we have a window */
677 if (my_rsv->_rsv_start >= group_first_block)
678 start = my_rsv->_rsv_start - group_first_block;
679 if (my_rsv->_rsv_end < group_last_block)
680 end = my_rsv->_rsv_end - group_first_block + 1;
681 if (grp_goal < start || grp_goal >= end)
684 BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
687 grp_goal = find_next_usable_block(start, bitmap_bh, end);
693 for (i = 0; i < 7 && grp_goal > start &&
694 !ext2_test_bit(grp_goal - 1,
701 for (; num < *count && grp_goal < end; grp_goal++) {
702 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
703 grp_goal, bitmap_bh->b_data)) {
715 return grp_goal - num;
721 * find_next_reservable_window - Find a reservable space within the given range.
722 * @search_head: The list to search.
723 * @my_rsv: The reservation we're currently using.
724 * @sb: The super block.
725 * @start_block: The first block we consider to start the real search from
726 * @last_block: The maximum block number that our goal reservable space
729 * It does not allocate the reservation window: alloc_new_reservation()
730 * will do the work later.
732 * We search the given range, rather than the whole reservation double
733 * linked list, (start_block, last_block) to find a free region that is
734 * of my size and has not been reserved.
736 * @search_head is not necessarily the list head of the whole filesystem.
737 * We have both head and @start_block to assist the search for the
738 * reservable space. The list starts from head, but we will shift to
739 * the place where start_block is, then start from there, when looking
740 * for a reservable space.
742 * @last_block is normally the last block in this group. The search will end
743 * when we found the start of next possible reservable space is out
744 * of this boundary. This could handle the cross boundary reservation
747 * Return: -1 if we could not find a range of sufficient size. If we could,
748 * return 0 and fill in @my_rsv with the range information.
750 static int find_next_reservable_window(
751 struct ext2_reserve_window_node *search_head,
752 struct ext2_reserve_window_node *my_rsv,
753 struct super_block * sb,
754 ext2_fsblk_t start_block,
755 ext2_fsblk_t last_block)
757 struct rb_node *next;
758 struct ext2_reserve_window_node *rsv, *prev;
760 int size = my_rsv->rsv_goal_size;
762 /* TODO: make the start of the reservation window byte-aligned */
763 /* cur = *start_block & ~7;*/
770 if (cur <= rsv->rsv_end)
771 cur = rsv->rsv_end + 1;
774 * in the case we could not find a reservable space
775 * that is what is expected, during the re-search, we could
776 * remember what's the largest reservable space we could have
777 * and return that one.
779 * For now it will fail if we could not find the reservable
780 * space with expected-size (or more)...
782 if (cur > last_block)
783 return -1; /* fail */
786 next = rb_next(&rsv->rsv_node);
787 rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
790 * Reached the last reservation, we can just append to the
796 if (cur + size <= rsv->rsv_start) {
798 * Found a reserveable space big enough. We could
799 * have a reservation across the group boundary here
805 * we come here either :
806 * when we reach the end of the whole list,
807 * and there is empty reservable space after last entry in the list.
808 * append it to the end of the list.
810 * or we found one reservable space in the middle of the list,
811 * return the reservation window that we could append to.
815 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
816 rsv_window_remove(sb, my_rsv);
819 * Let's book the whole available window for now. We will check the
820 * disk bitmap later and then, if there are free blocks then we adjust
821 * the window size if it's larger than requested.
822 * Otherwise, we will remove this node from the tree next time
823 * call find_next_reservable_window.
825 my_rsv->rsv_start = cur;
826 my_rsv->rsv_end = cur + size - 1;
827 my_rsv->rsv_alloc_hit = 0;
830 ext2_rsv_window_add(sb, my_rsv);
836 * alloc_new_reservation - Allocate a new reservation window.
837 * @my_rsv: The reservation we're currently using.
838 * @grp_goal: The goal block relative to the start of the group.
839 * @sb: The super block.
840 * @group: The group we are trying to allocate in.
841 * @bitmap_bh: The block group block bitmap.
843 * To make a new reservation, we search part of the filesystem reservation
844 * list (the list inside the group). We try to allocate a new
845 * reservation window near @grp_goal, or the beginning of the
846 * group, if @grp_goal is negative.
848 * We first find a reservable space after the goal, then from there,
849 * we check the bitmap for the first free block after it. If there is
850 * no free block until the end of group, then the whole group is full,
851 * we failed. Otherwise, check if the free block is inside the expected
852 * reservable space, if so, we succeed.
854 * If the first free block is outside the reservable space, then start
855 * from the first free block, we search for next available space, and
858 * on succeed, a new reservation will be found and inserted into the
859 * list. It contains at least one free block, and it does not overlap
860 * with other reservation windows.
862 * Return: 0 on success, -1 if we failed to find a reservation window
865 static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
866 ext2_grpblk_t grp_goal, struct super_block *sb,
867 unsigned int group, struct buffer_head *bitmap_bh)
869 struct ext2_reserve_window_node *search_head;
870 ext2_fsblk_t group_first_block, group_end_block, start_block;
871 ext2_grpblk_t first_free_block;
872 struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
875 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
877 group_first_block = ext2_group_first_block_no(sb, group);
878 group_end_block = ext2_group_last_block_no(sb, group);
881 start_block = group_first_block;
883 start_block = grp_goal + group_first_block;
885 size = my_rsv->rsv_goal_size;
887 if (!rsv_is_empty(&my_rsv->rsv_window)) {
889 * if the old reservation is cross group boundary
890 * and if the goal is inside the old reservation window,
891 * we will come here when we just failed to allocate from
892 * the first part of the window. We still have another part
893 * that belongs to the next group. In this case, there is no
894 * point to discard our window and try to allocate a new one
895 * in this group(which will fail). we should
896 * keep the reservation window, just simply move on.
898 * Maybe we could shift the start block of the reservation
899 * window to the first block of next group.
902 if ((my_rsv->rsv_start <= group_end_block) &&
903 (my_rsv->rsv_end > group_end_block) &&
904 (start_block >= my_rsv->rsv_start))
907 if ((my_rsv->rsv_alloc_hit >
908 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
910 * if the previously allocation hit ratio is
911 * greater than 1/2, then we double the size of
912 * the reservation window the next time,
913 * otherwise we keep the same size window
916 if (size > EXT2_MAX_RESERVE_BLOCKS)
917 size = EXT2_MAX_RESERVE_BLOCKS;
918 my_rsv->rsv_goal_size= size;
924 * shift the search start to the window near the goal block
926 search_head = search_reserve_window(fs_rsv_root, start_block);
929 * find_next_reservable_window() simply finds a reservable window
930 * inside the given range(start_block, group_end_block).
932 * To make sure the reservation window has a free bit inside it, we
933 * need to check the bitmap after we found a reservable window.
936 ret = find_next_reservable_window(search_head, my_rsv, sb,
937 start_block, group_end_block);
940 if (!rsv_is_empty(&my_rsv->rsv_window))
941 rsv_window_remove(sb, my_rsv);
942 spin_unlock(rsv_lock);
947 * On success, find_next_reservable_window() returns the
948 * reservation window where there is a reservable space after it.
949 * Before we reserve this reservable space, we need
950 * to make sure there is at least a free block inside this region.
952 * Search the first free bit on the block bitmap. Search starts from
953 * the start block of the reservable space we just found.
955 spin_unlock(rsv_lock);
956 first_free_block = bitmap_search_next_usable_block(
957 my_rsv->rsv_start - group_first_block,
958 bitmap_bh, group_end_block - group_first_block + 1);
960 if (first_free_block < 0) {
962 * no free block left on the bitmap, no point
963 * to reserve the space. return failed.
966 if (!rsv_is_empty(&my_rsv->rsv_window))
967 rsv_window_remove(sb, my_rsv);
968 spin_unlock(rsv_lock);
969 return -1; /* failed */
972 start_block = first_free_block + group_first_block;
974 * check if the first free block is within the
975 * free space we just reserved
977 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
978 return 0; /* success */
980 * if the first free bit we found is out of the reservable space
981 * continue search for next reservable space,
982 * start from where the free block is,
983 * we also shift the list head to where we stopped last time
985 search_head = my_rsv;
991 * try_to_extend_reservation()
992 * @my_rsv: given reservation window
994 * @size: the delta to extend
996 * Attempt to expand the reservation window large enough to have
997 * required number of free blocks
999 * Since ext2_try_to_allocate() will always allocate blocks within
1000 * the reservation window range, if the window size is too small,
1001 * multiple blocks allocation has to stop at the end of the reservation
1002 * window. To make this more efficient, given the total number of
1003 * blocks needed and the current size of the window, we try to
1004 * expand the reservation window size if necessary on a best-effort
1005 * basis before ext2_new_blocks() tries to allocate blocks.
1007 static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1008 struct super_block *sb, int size)
1010 struct ext2_reserve_window_node *next_rsv;
1011 struct rb_node *next;
1012 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1014 if (!spin_trylock(rsv_lock))
1017 next = rb_next(&my_rsv->rsv_node);
1020 my_rsv->rsv_end += size;
1022 next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1024 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1025 my_rsv->rsv_end += size;
1027 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1029 spin_unlock(rsv_lock);
1033 * ext2_try_to_allocate_with_rsv()
1035 * @group: given allocation block group
1036 * @bitmap_bh: bufferhead holds the block bitmap
1037 * @grp_goal: given target block within the group
1038 * @count: target number of blocks to allocate
1039 * @my_rsv: reservation window
1041 * This is the main function used to allocate a new block and its reservation
1044 * Each time when a new block allocation is need, first try to allocate from
1045 * its own reservation. If it does not have a reservation window, instead of
1046 * looking for a free bit on bitmap first, then look up the reservation list to
1047 * see if it is inside somebody else's reservation window, we try to allocate a
1048 * reservation window for it starting from the goal first. Then do the block
1049 * allocation within the reservation window.
1051 * This will avoid keeping on searching the reservation list again and
1052 * again when somebody is looking for a free block (without
1053 * reservation), and there are lots of free blocks, but they are all
1056 * We use a red-black tree for the per-filesystem reservation list.
1058 static ext2_grpblk_t
1059 ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1060 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1061 struct ext2_reserve_window_node * my_rsv,
1062 unsigned long *count)
1064 ext2_fsblk_t group_first_block, group_last_block;
1065 ext2_grpblk_t ret = 0;
1066 unsigned long num = *count;
1069 * we don't deal with reservation when
1070 * filesystem is mounted without reservation
1071 * or the file is not a regular file
1072 * or last attempt to allocate a block with reservation turned on failed
1074 if (my_rsv == NULL) {
1075 return ext2_try_to_allocate(sb, group, bitmap_bh,
1076 grp_goal, count, NULL);
1079 * grp_goal is a group relative block number (if there is a goal)
1080 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1081 * first block is a filesystem wide block number
1082 * first block is the block number of the first block in this group
1084 group_first_block = ext2_group_first_block_no(sb, group);
1085 group_last_block = ext2_group_last_block_no(sb, group);
1088 * Basically we will allocate a new block from inode's reservation
1091 * We need to allocate a new reservation window, if:
1092 * a) inode does not have a reservation window; or
1093 * b) last attempt to allocate a block from existing reservation
1095 * c) we come here with a goal and with a reservation window
1097 * We do not need to allocate a new reservation window if we come here
1098 * at the beginning with a goal and the goal is inside the window, or
1099 * we don't have a goal but already have a reservation window.
1100 * then we could go to allocate from the reservation window directly.
1103 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1104 !goal_in_my_reservation(&my_rsv->rsv_window,
1105 grp_goal, group, sb)) {
1106 if (my_rsv->rsv_goal_size < *count)
1107 my_rsv->rsv_goal_size = *count;
1108 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1113 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1114 grp_goal, group, sb))
1116 } else if (grp_goal >= 0) {
1117 int curr = my_rsv->rsv_end -
1118 (grp_goal + group_first_block) + 1;
1121 try_to_extend_reservation(my_rsv, sb,
1125 if ((my_rsv->rsv_start > group_last_block) ||
1126 (my_rsv->rsv_end < group_first_block)) {
1127 ext2_error(sb, __func__,
1128 "Reservation out of group %u range goal %d fsb[%lu,%lu] rsv[%lu, %lu]",
1129 group, grp_goal, group_first_block,
1130 group_last_block, my_rsv->rsv_start,
1132 rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1135 ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1136 &num, &my_rsv->rsv_window);
1138 my_rsv->rsv_alloc_hit += num;
1140 break; /* succeed */
1148 * ext2_has_free_blocks()
1149 * @sbi: in-core super block structure.
1151 * Check if filesystem has at least 1 free block available for allocation.
1153 static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1155 ext2_fsblk_t free_blocks, root_blocks;
1157 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1158 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1159 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1160 !uid_eq(sbi->s_resuid, current_fsuid()) &&
1161 (gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1162 !in_group_p (sbi->s_resgid))) {
1169 * Returns 1 if the passed-in block region is valid; 0 if some part overlaps
1170 * with filesystem metadata blocks.
1172 int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk,
1175 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
1176 (start_blk + count - 1 < start_blk) ||
1177 (start_blk + count - 1 >= le32_to_cpu(sbi->s_es->s_blocks_count)))
1180 /* Ensure we do not step over superblock */
1181 if ((start_blk <= sbi->s_sb_block) &&
1182 (start_blk + count - 1 >= sbi->s_sb_block))
1189 * ext2_new_blocks() -- core block(s) allocation function
1190 * @inode: file inode
1191 * @goal: given target block(filesystem wide)
1192 * @count: target number of blocks to allocate
1194 * @flags: allocate flags
1196 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
1197 * free, or there is a free block within 32 blocks of the goal, that block
1198 * is allocated. Otherwise a forward search is made for a free block; within
1199 * each block group the search first looks for an entire free byte in the block
1200 * bitmap, and then for any free bit if that fails.
1201 * This function also updates quota and i_blocks field.
1203 ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1204 unsigned long *count, int *errp, unsigned int flags)
1206 struct buffer_head *bitmap_bh = NULL;
1207 struct buffer_head *gdp_bh;
1210 ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1211 ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1212 ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
1213 int bgi; /* blockgroup iteration index */
1214 int performed_allocation = 0;
1215 ext2_grpblk_t free_blocks; /* number of free blocks in a group */
1216 struct super_block *sb;
1217 struct ext2_group_desc *gdp;
1218 struct ext2_super_block *es;
1219 struct ext2_sb_info *sbi;
1220 struct ext2_reserve_window_node *my_rsv = NULL;
1221 struct ext2_block_alloc_info *block_i;
1222 unsigned short windowsz = 0;
1223 unsigned long ngroups;
1224 unsigned long num = *count;
1231 * Check quota for allocation of this block.
1233 ret = dquot_alloc_block(inode, num);
1240 es = EXT2_SB(sb)->s_es;
1241 ext2_debug("goal=%lu.\n", goal);
1243 * Allocate a block from reservation only when the filesystem is
1244 * mounted with reservation(default,-o reservation), and it's a regular
1245 * file, and the desired window size is greater than 0 (One could use
1246 * ioctl command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn
1247 * off reservation on that particular file). Also do not use the
1248 * reservation window if the caller asked us not to do it.
1250 block_i = EXT2_I(inode)->i_block_alloc_info;
1251 if (!(flags & EXT2_ALLOC_NORESERVE) && block_i) {
1252 windowsz = block_i->rsv_window_node.rsv_goal_size;
1254 my_rsv = &block_i->rsv_window_node;
1257 if (!ext2_has_free_blocks(sbi)) {
1263 * First, test whether the goal block is free.
1265 if (goal < le32_to_cpu(es->s_first_data_block) ||
1266 goal >= le32_to_cpu(es->s_blocks_count))
1267 goal = le32_to_cpu(es->s_first_data_block);
1268 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1269 EXT2_BLOCKS_PER_GROUP(sb);
1270 goal_group = group_no;
1272 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1276 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1278 * if there is not enough free blocks to make a new resevation
1279 * turn off reservation for this allocation
1281 if (my_rsv && (free_blocks < windowsz)
1282 && (free_blocks > 0)
1283 && (rsv_is_empty(&my_rsv->rsv_window)))
1286 if (free_blocks > 0) {
1287 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1288 EXT2_BLOCKS_PER_GROUP(sb));
1290 * In case we retry allocation (due to fs reservation not
1291 * working out or fs corruption), the bitmap_bh is non-null
1292 * pointer and we have to release it before calling
1293 * read_block_bitmap().
1296 bitmap_bh = read_block_bitmap(sb, group_no);
1299 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1300 bitmap_bh, grp_target_blk,
1302 if (grp_alloc_blk >= 0)
1306 ngroups = EXT2_SB(sb)->s_groups_count;
1310 * Now search the rest of the groups. We assume that
1311 * group_no and gdp correctly point to the last group visited.
1313 for (bgi = 0; bgi < ngroups; bgi++) {
1315 if (group_no >= ngroups)
1317 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1321 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1323 * skip this group (and avoid loading bitmap) if there
1324 * are no free blocks
1329 * skip this group if the number of
1330 * free blocks is less than half of the reservation
1333 if (my_rsv && (free_blocks <= (windowsz/2)))
1337 bitmap_bh = read_block_bitmap(sb, group_no);
1341 * try to allocate block(s) from this group, without a goal(-1).
1343 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1344 bitmap_bh, -1, my_rsv, &num);
1345 if (grp_alloc_blk >= 0)
1349 * We may end up a bogus earlier ENOSPC error due to
1350 * filesystem is "full" of reservations, but
1351 * there maybe indeed free blocks available on disk
1352 * In this case, we just forget about the reservations
1353 * just do block allocation as without reservations.
1358 group_no = goal_group;
1361 /* No space left on the device */
1367 ext2_debug("using block group %d(%d)\n",
1368 group_no, gdp->bg_free_blocks_count);
1370 ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1372 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1373 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1374 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1375 EXT2_SB(sb)->s_itb_per_group) ||
1376 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1377 EXT2_SB(sb)->s_itb_per_group)) {
1378 ext2_error(sb, "ext2_new_blocks",
1379 "Allocating block in system zone - "
1380 "blocks from "E2FSBLK", length %lu",
1383 * ext2_try_to_allocate marked the blocks we allocated as in
1384 * use. So we may want to selectively mark some of the blocks
1391 performed_allocation = 1;
1393 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1394 ext2_error(sb, "ext2_new_blocks",
1395 "block("E2FSBLK") >= blocks count(%d) - "
1396 "block_group = %d, es == %p ", ret_block,
1397 le32_to_cpu(es->s_blocks_count), group_no, es);
1401 group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1402 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1404 mark_buffer_dirty(bitmap_bh);
1405 if (sb->s_flags & SB_SYNCHRONOUS)
1406 sync_dirty_buffer(bitmap_bh);
1411 dquot_free_block_nodirty(inode, *count-num);
1412 mark_inode_dirty(inode);
1421 * Undo the block allocation
1423 if (!performed_allocation) {
1424 dquot_free_block_nodirty(inode, *count);
1425 mark_inode_dirty(inode);
1433 unsigned long ext2_count_free(struct buffer_head *map, unsigned int numchars)
1435 return numchars * BITS_PER_BYTE - memweight(map->b_data, numchars);
1438 #endif /* EXT2FS_DEBUG */
1440 unsigned long ext2_count_free_blocks (struct super_block * sb)
1442 struct ext2_group_desc * desc;
1443 unsigned long desc_count = 0;
1446 unsigned long bitmap_count, x;
1447 struct ext2_super_block *es;
1449 es = EXT2_SB(sb)->s_es;
1453 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1454 struct buffer_head *bitmap_bh;
1455 desc = ext2_get_group_desc (sb, i, NULL);
1458 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1459 bitmap_bh = read_block_bitmap(sb, i);
1463 x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1464 printk ("group %d: stored = %d, counted = %lu\n",
1465 i, le16_to_cpu(desc->bg_free_blocks_count), x);
1469 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1470 (long)le32_to_cpu(es->s_free_blocks_count),
1471 desc_count, bitmap_count);
1472 return bitmap_count;
1474 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1475 desc = ext2_get_group_desc(sb, i, NULL);
1478 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1484 static inline int test_root(int a, int b)
1493 static int ext2_group_sparse(int group)
1497 return (test_root(group, 3) || test_root(group, 5) ||
1498 test_root(group, 7));
1502 * ext2_bg_has_super - number of blocks used by the superblock in group
1503 * @sb: superblock for filesystem
1504 * @group: group number to check
1506 * Return the number of blocks used by the superblock (primary or backup)
1507 * in this group. Currently this will be only 0 or 1.
1509 int ext2_bg_has_super(struct super_block *sb, int group)
1511 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1512 !ext2_group_sparse(group))
1518 * ext2_bg_num_gdb - number of blocks used by the group table in group
1519 * @sb: superblock for filesystem
1520 * @group: group number to check
1522 * Return the number of blocks used by the group descriptor table
1523 * (primary or backup) in this group. In the future there may be a
1524 * different number of descriptor blocks in each group.
1526 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1528 return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;