xref: /openbmc/linux/fs/ext2/balloc.c (revision 65417d9f)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext2/balloc.c
4  *
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)
9  *
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
13  */
14 
15 #include "ext2.h"
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>
22 
23 /*
24  * balloc.c contains the blocks allocation and deallocation routines
25  */
26 
27 /*
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.
31  *
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).
36  */
37 
38 
39 #define in_range(b, first, len)	((b) >= (first) && (b) <= (first) + (len) - 1)
40 
41 struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
42 					     unsigned int block_group,
43 					     struct buffer_head ** bh)
44 {
45 	unsigned long group_desc;
46 	unsigned long offset;
47 	struct ext2_group_desc * desc;
48 	struct ext2_sb_info *sbi = EXT2_SB(sb);
49 
50 	if (block_group >= sbi->s_groups_count) {
51 		ext2_error (sb, "ext2_get_group_desc",
52 			    "block_group >= groups_count - "
53 			    "block_group = %d, groups_count = %lu",
54 			    block_group, sbi->s_groups_count);
55 
56 		return NULL;
57 	}
58 
59 	group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
60 	offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
61 	if (!sbi->s_group_desc[group_desc]) {
62 		ext2_error (sb, "ext2_get_group_desc",
63 			    "Group descriptor not loaded - "
64 			    "block_group = %d, group_desc = %lu, desc = %lu",
65 			     block_group, group_desc, offset);
66 		return NULL;
67 	}
68 
69 	desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
70 	if (bh)
71 		*bh = sbi->s_group_desc[group_desc];
72 	return desc + offset;
73 }
74 
75 static int ext2_valid_block_bitmap(struct super_block *sb,
76 					struct ext2_group_desc *desc,
77 					unsigned int block_group,
78 					struct buffer_head *bh)
79 {
80 	ext2_grpblk_t offset;
81 	ext2_grpblk_t next_zero_bit;
82 	ext2_fsblk_t bitmap_blk;
83 	ext2_fsblk_t group_first_block;
84 
85 	group_first_block = ext2_group_first_block_no(sb, block_group);
86 
87 	/* check whether block bitmap block number is set */
88 	bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
89 	offset = bitmap_blk - group_first_block;
90 	if (!ext2_test_bit(offset, bh->b_data))
91 		/* bad block bitmap */
92 		goto err_out;
93 
94 	/* check whether the inode bitmap block number is set */
95 	bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
96 	offset = bitmap_blk - group_first_block;
97 	if (!ext2_test_bit(offset, bh->b_data))
98 		/* bad block bitmap */
99 		goto err_out;
100 
101 	/* check whether the inode table block number is set */
102 	bitmap_blk = le32_to_cpu(desc->bg_inode_table);
103 	offset = bitmap_blk - group_first_block;
104 	next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
105 				offset + EXT2_SB(sb)->s_itb_per_group,
106 				offset);
107 	if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
108 		/* good bitmap for inode tables */
109 		return 1;
110 
111 err_out:
112 	ext2_error(sb, __func__,
113 			"Invalid block bitmap - "
114 			"block_group = %d, block = %lu",
115 			block_group, bitmap_blk);
116 	return 0;
117 }
118 
119 /*
120  * Read the bitmap for a given block_group,and validate the
121  * bits for block/inode/inode tables are set in the bitmaps
122  *
123  * Return buffer_head on success or NULL in case of failure.
124  */
125 static struct buffer_head *
126 read_block_bitmap(struct super_block *sb, unsigned int block_group)
127 {
128 	struct ext2_group_desc * desc;
129 	struct buffer_head * bh = NULL;
130 	ext2_fsblk_t bitmap_blk;
131 
132 	desc = ext2_get_group_desc(sb, block_group, NULL);
133 	if (!desc)
134 		return NULL;
135 	bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
136 	bh = sb_getblk(sb, bitmap_blk);
137 	if (unlikely(!bh)) {
138 		ext2_error(sb, __func__,
139 			    "Cannot read block bitmap - "
140 			    "block_group = %d, block_bitmap = %u",
141 			    block_group, le32_to_cpu(desc->bg_block_bitmap));
142 		return NULL;
143 	}
144 	if (likely(bh_uptodate_or_lock(bh)))
145 		return bh;
146 
147 	if (bh_submit_read(bh) < 0) {
148 		brelse(bh);
149 		ext2_error(sb, __func__,
150 			    "Cannot read block bitmap - "
151 			    "block_group = %d, block_bitmap = %u",
152 			    block_group, le32_to_cpu(desc->bg_block_bitmap));
153 		return NULL;
154 	}
155 
156 	ext2_valid_block_bitmap(sb, desc, block_group, bh);
157 	/*
158 	 * file system mounted not to panic on error, continue with corrupt
159 	 * bitmap
160 	 */
161 	return bh;
162 }
163 
164 static void group_adjust_blocks(struct super_block *sb, int group_no,
165 	struct ext2_group_desc *desc, struct buffer_head *bh, int count)
166 {
167 	if (count) {
168 		struct ext2_sb_info *sbi = EXT2_SB(sb);
169 		unsigned free_blocks;
170 
171 		spin_lock(sb_bgl_lock(sbi, group_no));
172 		free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
173 		desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
174 		spin_unlock(sb_bgl_lock(sbi, group_no));
175 		mark_buffer_dirty(bh);
176 	}
177 }
178 
179 /*
180  * The reservation window structure operations
181  * --------------------------------------------
182  * Operations include:
183  * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
184  *
185  * We use a red-black tree to represent per-filesystem reservation
186  * windows.
187  *
188  */
189 
190 /**
191  * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
192  * @rb_root:		root of per-filesystem reservation rb tree
193  * @verbose:		verbose mode
194  * @fn:			function which wishes to dump the reservation map
195  *
196  * If verbose is turned on, it will print the whole block reservation
197  * windows(start, end). Otherwise, it will only print out the "bad" windows,
198  * those windows that overlap with their immediate neighbors.
199  */
200 #if 1
201 static void __rsv_window_dump(struct rb_root *root, int verbose,
202 			      const char *fn)
203 {
204 	struct rb_node *n;
205 	struct ext2_reserve_window_node *rsv, *prev;
206 	int bad;
207 
208 restart:
209 	n = rb_first(root);
210 	bad = 0;
211 	prev = NULL;
212 
213 	printk("Block Allocation Reservation Windows Map (%s):\n", fn);
214 	while (n) {
215 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
216 		if (verbose)
217 			printk("reservation window 0x%p "
218 				"start: %lu, end: %lu\n",
219 				rsv, rsv->rsv_start, rsv->rsv_end);
220 		if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
221 			printk("Bad reservation %p (start >= end)\n",
222 			       rsv);
223 			bad = 1;
224 		}
225 		if (prev && prev->rsv_end >= rsv->rsv_start) {
226 			printk("Bad reservation %p (prev->end >= start)\n",
227 			       rsv);
228 			bad = 1;
229 		}
230 		if (bad) {
231 			if (!verbose) {
232 				printk("Restarting reservation walk in verbose mode\n");
233 				verbose = 1;
234 				goto restart;
235 			}
236 		}
237 		n = rb_next(n);
238 		prev = rsv;
239 	}
240 	printk("Window map complete.\n");
241 	BUG_ON(bad);
242 }
243 #define rsv_window_dump(root, verbose) \
244 	__rsv_window_dump((root), (verbose), __func__)
245 #else
246 #define rsv_window_dump(root, verbose) do {} while (0)
247 #endif
248 
249 /**
250  * goal_in_my_reservation()
251  * @rsv:		inode's reservation window
252  * @grp_goal:		given goal block relative to the allocation block group
253  * @group:		the current allocation block group
254  * @sb:			filesystem super block
255  *
256  * Test if the given goal block (group relative) is within the file's
257  * own block reservation window range.
258  *
259  * If the reservation window is outside the goal allocation group, return 0;
260  * grp_goal (given goal block) could be -1, which means no specific
261  * goal block. In this case, always return 1.
262  * If the goal block is within the reservation window, return 1;
263  * otherwise, return 0;
264  */
265 static int
266 goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
267 			unsigned int group, struct super_block * sb)
268 {
269 	ext2_fsblk_t group_first_block, group_last_block;
270 
271 	group_first_block = ext2_group_first_block_no(sb, group);
272 	group_last_block = group_first_block + EXT2_BLOCKS_PER_GROUP(sb) - 1;
273 
274 	if ((rsv->_rsv_start > group_last_block) ||
275 	    (rsv->_rsv_end < group_first_block))
276 		return 0;
277 	if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
278 		|| (grp_goal + group_first_block > rsv->_rsv_end)))
279 		return 0;
280 	return 1;
281 }
282 
283 /**
284  * search_reserve_window()
285  * @rb_root:		root of reservation tree
286  * @goal:		target allocation block
287  *
288  * Find the reserved window which includes the goal, or the previous one
289  * if the goal is not in any window.
290  * Returns NULL if there are no windows or if all windows start after the goal.
291  */
292 static struct ext2_reserve_window_node *
293 search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
294 {
295 	struct rb_node *n = root->rb_node;
296 	struct ext2_reserve_window_node *rsv;
297 
298 	if (!n)
299 		return NULL;
300 
301 	do {
302 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
303 
304 		if (goal < rsv->rsv_start)
305 			n = n->rb_left;
306 		else if (goal > rsv->rsv_end)
307 			n = n->rb_right;
308 		else
309 			return rsv;
310 	} while (n);
311 	/*
312 	 * We've fallen off the end of the tree: the goal wasn't inside
313 	 * any particular node.  OK, the previous node must be to one
314 	 * side of the interval containing the goal.  If it's the RHS,
315 	 * we need to back up one.
316 	 */
317 	if (rsv->rsv_start > goal) {
318 		n = rb_prev(&rsv->rsv_node);
319 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
320 	}
321 	return rsv;
322 }
323 
324 /*
325  * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
326  * @sb:			super block
327  * @rsv:		reservation window to add
328  *
329  * Must be called with rsv_lock held.
330  */
331 void ext2_rsv_window_add(struct super_block *sb,
332 		    struct ext2_reserve_window_node *rsv)
333 {
334 	struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
335 	struct rb_node *node = &rsv->rsv_node;
336 	ext2_fsblk_t start = rsv->rsv_start;
337 
338 	struct rb_node ** p = &root->rb_node;
339 	struct rb_node * parent = NULL;
340 	struct ext2_reserve_window_node *this;
341 
342 	while (*p)
343 	{
344 		parent = *p;
345 		this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
346 
347 		if (start < this->rsv_start)
348 			p = &(*p)->rb_left;
349 		else if (start > this->rsv_end)
350 			p = &(*p)->rb_right;
351 		else {
352 			rsv_window_dump(root, 1);
353 			BUG();
354 		}
355 	}
356 
357 	rb_link_node(node, parent, p);
358 	rb_insert_color(node, root);
359 }
360 
361 /**
362  * rsv_window_remove() -- unlink a window from the reservation rb tree
363  * @sb:			super block
364  * @rsv:		reservation window to remove
365  *
366  * Mark the block reservation window as not allocated, and unlink it
367  * from the filesystem reservation window rb tree. Must be called with
368  * rsv_lock held.
369  */
370 static void rsv_window_remove(struct super_block *sb,
371 			      struct ext2_reserve_window_node *rsv)
372 {
373 	rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
374 	rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
375 	rsv->rsv_alloc_hit = 0;
376 	rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
377 }
378 
379 /*
380  * rsv_is_empty() -- Check if the reservation window is allocated.
381  * @rsv:		given reservation window to check
382  *
383  * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
384  */
385 static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
386 {
387 	/* a valid reservation end block could not be 0 */
388 	return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
389 }
390 
391 /**
392  * ext2_init_block_alloc_info()
393  * @inode:		file inode structure
394  *
395  * Allocate and initialize the  reservation window structure, and
396  * link the window to the ext2 inode structure at last
397  *
398  * The reservation window structure is only dynamically allocated
399  * and linked to ext2 inode the first time the open file
400  * needs a new block. So, before every ext2_new_block(s) call, for
401  * regular files, we should check whether the reservation window
402  * structure exists or not. In the latter case, this function is called.
403  * Fail to do so will result in block reservation being turned off for that
404  * open file.
405  *
406  * This function is called from ext2_get_blocks_handle(), also called
407  * when setting the reservation window size through ioctl before the file
408  * is open for write (needs block allocation).
409  *
410  * Needs truncate_mutex protection prior to calling this function.
411  */
412 void ext2_init_block_alloc_info(struct inode *inode)
413 {
414 	struct ext2_inode_info *ei = EXT2_I(inode);
415 	struct ext2_block_alloc_info *block_i;
416 	struct super_block *sb = inode->i_sb;
417 
418 	block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
419 	if (block_i) {
420 		struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
421 
422 		rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
423 		rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
424 
425 	 	/*
426 		 * if filesystem is mounted with NORESERVATION, the goal
427 		 * reservation window size is set to zero to indicate
428 		 * block reservation is off
429 		 */
430 		if (!test_opt(sb, RESERVATION))
431 			rsv->rsv_goal_size = 0;
432 		else
433 			rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
434 		rsv->rsv_alloc_hit = 0;
435 		block_i->last_alloc_logical_block = 0;
436 		block_i->last_alloc_physical_block = 0;
437 	}
438 	ei->i_block_alloc_info = block_i;
439 }
440 
441 /**
442  * ext2_discard_reservation()
443  * @inode:		inode
444  *
445  * Discard(free) block reservation window on last file close, or truncate
446  * or at last iput().
447  *
448  * It is being called in three cases:
449  * 	ext2_release_file(): last writer closes the file
450  * 	ext2_clear_inode(): last iput(), when nobody links to this file.
451  * 	ext2_truncate(): when the block indirect map is about to change.
452  */
453 void ext2_discard_reservation(struct inode *inode)
454 {
455 	struct ext2_inode_info *ei = EXT2_I(inode);
456 	struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
457 	struct ext2_reserve_window_node *rsv;
458 	spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
459 
460 	if (!block_i)
461 		return;
462 
463 	rsv = &block_i->rsv_window_node;
464 	if (!rsv_is_empty(&rsv->rsv_window)) {
465 		spin_lock(rsv_lock);
466 		if (!rsv_is_empty(&rsv->rsv_window))
467 			rsv_window_remove(inode->i_sb, rsv);
468 		spin_unlock(rsv_lock);
469 	}
470 }
471 
472 /**
473  * ext2_free_blocks() -- Free given blocks and update quota and i_blocks
474  * @inode:		inode
475  * @block:		start physical block to free
476  * @count:		number of blocks to free
477  */
478 void ext2_free_blocks (struct inode * inode, unsigned long block,
479 		       unsigned long count)
480 {
481 	struct buffer_head *bitmap_bh = NULL;
482 	struct buffer_head * bh2;
483 	unsigned long block_group;
484 	unsigned long bit;
485 	unsigned long i;
486 	unsigned long overflow;
487 	struct super_block * sb = inode->i_sb;
488 	struct ext2_sb_info * sbi = EXT2_SB(sb);
489 	struct ext2_group_desc * desc;
490 	struct ext2_super_block * es = sbi->s_es;
491 	unsigned freed = 0, group_freed;
492 
493 	if (block < le32_to_cpu(es->s_first_data_block) ||
494 	    block + count < block ||
495 	    block + count > le32_to_cpu(es->s_blocks_count)) {
496 		ext2_error (sb, "ext2_free_blocks",
497 			    "Freeing blocks not in datazone - "
498 			    "block = %lu, count = %lu", block, count);
499 		goto error_return;
500 	}
501 
502 	ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
503 
504 do_more:
505 	overflow = 0;
506 	block_group = (block - le32_to_cpu(es->s_first_data_block)) /
507 		      EXT2_BLOCKS_PER_GROUP(sb);
508 	bit = (block - le32_to_cpu(es->s_first_data_block)) %
509 		      EXT2_BLOCKS_PER_GROUP(sb);
510 	/*
511 	 * Check to see if we are freeing blocks across a group
512 	 * boundary.
513 	 */
514 	if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
515 		overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
516 		count -= overflow;
517 	}
518 	brelse(bitmap_bh);
519 	bitmap_bh = read_block_bitmap(sb, block_group);
520 	if (!bitmap_bh)
521 		goto error_return;
522 
523 	desc = ext2_get_group_desc (sb, block_group, &bh2);
524 	if (!desc)
525 		goto error_return;
526 
527 	if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
528 	    in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
529 	    in_range (block, le32_to_cpu(desc->bg_inode_table),
530 		      sbi->s_itb_per_group) ||
531 	    in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
532 		      sbi->s_itb_per_group)) {
533 		ext2_error (sb, "ext2_free_blocks",
534 			    "Freeing blocks in system zones - "
535 			    "Block = %lu, count = %lu",
536 			    block, count);
537 		goto error_return;
538 	}
539 
540 	for (i = 0, group_freed = 0; i < count; i++) {
541 		if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
542 						bit + i, bitmap_bh->b_data)) {
543 			ext2_error(sb, __func__,
544 				"bit already cleared for block %lu", block + i);
545 		} else {
546 			group_freed++;
547 		}
548 	}
549 
550 	mark_buffer_dirty(bitmap_bh);
551 	if (sb->s_flags & SB_SYNCHRONOUS)
552 		sync_dirty_buffer(bitmap_bh);
553 
554 	group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
555 	freed += group_freed;
556 
557 	if (overflow) {
558 		block += count;
559 		count = overflow;
560 		goto do_more;
561 	}
562 error_return:
563 	brelse(bitmap_bh);
564 	if (freed) {
565 		percpu_counter_add(&sbi->s_freeblocks_counter, freed);
566 		dquot_free_block_nodirty(inode, freed);
567 		mark_inode_dirty(inode);
568 	}
569 }
570 
571 /**
572  * bitmap_search_next_usable_block()
573  * @start:		the starting block (group relative) of the search
574  * @bh:			bufferhead contains the block group bitmap
575  * @maxblocks:		the ending block (group relative) of the reservation
576  *
577  * The bitmap search --- search forward through the actual bitmap on disk until
578  * we find a bit free.
579  */
580 static ext2_grpblk_t
581 bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
582 					ext2_grpblk_t maxblocks)
583 {
584 	ext2_grpblk_t next;
585 
586 	next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
587 	if (next >= maxblocks)
588 		return -1;
589 	return next;
590 }
591 
592 /**
593  * find_next_usable_block()
594  * @start:		the starting block (group relative) to find next
595  * 			allocatable block in bitmap.
596  * @bh:			bufferhead contains the block group bitmap
597  * @maxblocks:		the ending block (group relative) for the search
598  *
599  * Find an allocatable block in a bitmap.  We perform the "most
600  * appropriate allocation" algorithm of looking for a free block near
601  * the initial goal; then for a free byte somewhere in the bitmap;
602  * then for any free bit in the bitmap.
603  */
604 static ext2_grpblk_t
605 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
606 {
607 	ext2_grpblk_t here, next;
608 	char *p, *r;
609 
610 	if (start > 0) {
611 		/*
612 		 * The goal was occupied; search forward for a free
613 		 * block within the next XX blocks.
614 		 *
615 		 * end_goal is more or less random, but it has to be
616 		 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
617 		 * next 64-bit boundary is simple..
618 		 */
619 		ext2_grpblk_t end_goal = (start + 63) & ~63;
620 		if (end_goal > maxblocks)
621 			end_goal = maxblocks;
622 		here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
623 		if (here < end_goal)
624 			return here;
625 		ext2_debug("Bit not found near goal\n");
626 	}
627 
628 	here = start;
629 	if (here < 0)
630 		here = 0;
631 
632 	p = ((char *)bh->b_data) + (here >> 3);
633 	r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
634 	next = (r - ((char *)bh->b_data)) << 3;
635 
636 	if (next < maxblocks && next >= here)
637 		return next;
638 
639 	here = bitmap_search_next_usable_block(here, bh, maxblocks);
640 	return here;
641 }
642 
643 /**
644  * ext2_try_to_allocate()
645  * @sb:			superblock
646  * @group:		given allocation block group
647  * @bitmap_bh:		bufferhead holds the block bitmap
648  * @grp_goal:		given target block within the group
649  * @count:		target number of blocks to allocate
650  * @my_rsv:		reservation window
651  *
652  * Attempt to allocate blocks within a give range. Set the range of allocation
653  * first, then find the first free bit(s) from the bitmap (within the range),
654  * and at last, allocate the blocks by claiming the found free bit as allocated.
655  *
656  * To set the range of this allocation:
657  * 	if there is a reservation window, only try to allocate block(s)
658  * 	from the file's own reservation window;
659  * 	Otherwise, the allocation range starts from the give goal block,
660  * 	ends at the block group's last block.
661  *
662  * If we failed to allocate the desired block then we may end up crossing to a
663  * new bitmap.
664  */
665 static int
666 ext2_try_to_allocate(struct super_block *sb, int group,
667 			struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
668 			unsigned long *count,
669 			struct ext2_reserve_window *my_rsv)
670 {
671 	ext2_fsblk_t group_first_block;
672        	ext2_grpblk_t start, end;
673 	unsigned long num = 0;
674 
675 	/* we do allocation within the reservation window if we have a window */
676 	if (my_rsv) {
677 		group_first_block = ext2_group_first_block_no(sb, group);
678 		if (my_rsv->_rsv_start >= group_first_block)
679 			start = my_rsv->_rsv_start - group_first_block;
680 		else
681 			/* reservation window cross group boundary */
682 			start = 0;
683 		end = my_rsv->_rsv_end - group_first_block + 1;
684 		if (end > EXT2_BLOCKS_PER_GROUP(sb))
685 			/* reservation window crosses group boundary */
686 			end = EXT2_BLOCKS_PER_GROUP(sb);
687 		if ((start <= grp_goal) && (grp_goal < end))
688 			start = grp_goal;
689 		else
690 			grp_goal = -1;
691 	} else {
692 		if (grp_goal > 0)
693 			start = grp_goal;
694 		else
695 			start = 0;
696 		end = EXT2_BLOCKS_PER_GROUP(sb);
697 	}
698 
699 	BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
700 
701 repeat:
702 	if (grp_goal < 0) {
703 		grp_goal = find_next_usable_block(start, bitmap_bh, end);
704 		if (grp_goal < 0)
705 			goto fail_access;
706 		if (!my_rsv) {
707 			int i;
708 
709 			for (i = 0; i < 7 && grp_goal > start &&
710 					!ext2_test_bit(grp_goal - 1,
711 					     		bitmap_bh->b_data);
712 			     		i++, grp_goal--)
713 				;
714 		}
715 	}
716 	start = grp_goal;
717 
718 	if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), grp_goal,
719 			       				bitmap_bh->b_data)) {
720 		/*
721 		 * The block was allocated by another thread, or it was
722 		 * allocated and then freed by another thread
723 		 */
724 		start++;
725 		grp_goal++;
726 		if (start >= end)
727 			goto fail_access;
728 		goto repeat;
729 	}
730 	num++;
731 	grp_goal++;
732 	while (num < *count && grp_goal < end
733 		&& !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
734 					grp_goal, bitmap_bh->b_data)) {
735 		num++;
736 		grp_goal++;
737 	}
738 	*count = num;
739 	return grp_goal - num;
740 fail_access:
741 	*count = num;
742 	return -1;
743 }
744 
745 /**
746  * 	find_next_reservable_window():
747  *		find a reservable space within the given range.
748  *		It does not allocate the reservation window for now:
749  *		alloc_new_reservation() will do the work later.
750  *
751  * 	@search_head: the head of the searching list;
752  *		This is not necessarily the list head of the whole filesystem
753  *
754  *		We have both head and start_block to assist the search
755  *		for the reservable space. The list starts from head,
756  *		but we will shift to the place where start_block is,
757  *		then start from there, when looking for a reservable space.
758  *
759  * 	@size: the target new reservation window size
760  *
761  * 	@group_first_block: the first block we consider to start
762  *			the real search from
763  *
764  * 	@last_block:
765  *		the maximum block number that our goal reservable space
766  *		could start from. This is normally the last block in this
767  *		group. The search will end when we found the start of next
768  *		possible reservable space is out of this boundary.
769  *		This could handle the cross boundary reservation window
770  *		request.
771  *
772  * 	basically we search from the given range, rather than the whole
773  * 	reservation double linked list, (start_block, last_block)
774  * 	to find a free region that is of my size and has not
775  * 	been reserved.
776  *
777  */
778 static int find_next_reservable_window(
779 				struct ext2_reserve_window_node *search_head,
780 				struct ext2_reserve_window_node *my_rsv,
781 				struct super_block * sb,
782 				ext2_fsblk_t start_block,
783 				ext2_fsblk_t last_block)
784 {
785 	struct rb_node *next;
786 	struct ext2_reserve_window_node *rsv, *prev;
787 	ext2_fsblk_t cur;
788 	int size = my_rsv->rsv_goal_size;
789 
790 	/* TODO: make the start of the reservation window byte-aligned */
791 	/* cur = *start_block & ~7;*/
792 	cur = start_block;
793 	rsv = search_head;
794 	if (!rsv)
795 		return -1;
796 
797 	while (1) {
798 		if (cur <= rsv->rsv_end)
799 			cur = rsv->rsv_end + 1;
800 
801 		/* TODO?
802 		 * in the case we could not find a reservable space
803 		 * that is what is expected, during the re-search, we could
804 		 * remember what's the largest reservable space we could have
805 		 * and return that one.
806 		 *
807 		 * For now it will fail if we could not find the reservable
808 		 * space with expected-size (or more)...
809 		 */
810 		if (cur > last_block)
811 			return -1;		/* fail */
812 
813 		prev = rsv;
814 		next = rb_next(&rsv->rsv_node);
815 		rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
816 
817 		/*
818 		 * Reached the last reservation, we can just append to the
819 		 * previous one.
820 		 */
821 		if (!next)
822 			break;
823 
824 		if (cur + size <= rsv->rsv_start) {
825 			/*
826 			 * Found a reserveable space big enough.  We could
827 			 * have a reservation across the group boundary here
828 		 	 */
829 			break;
830 		}
831 	}
832 	/*
833 	 * we come here either :
834 	 * when we reach the end of the whole list,
835 	 * and there is empty reservable space after last entry in the list.
836 	 * append it to the end of the list.
837 	 *
838 	 * or we found one reservable space in the middle of the list,
839 	 * return the reservation window that we could append to.
840 	 * succeed.
841 	 */
842 
843 	if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
844 		rsv_window_remove(sb, my_rsv);
845 
846 	/*
847 	 * Let's book the whole available window for now.  We will check the
848 	 * disk bitmap later and then, if there are free blocks then we adjust
849 	 * the window size if it's larger than requested.
850 	 * Otherwise, we will remove this node from the tree next time
851 	 * call find_next_reservable_window.
852 	 */
853 	my_rsv->rsv_start = cur;
854 	my_rsv->rsv_end = cur + size - 1;
855 	my_rsv->rsv_alloc_hit = 0;
856 
857 	if (prev != my_rsv)
858 		ext2_rsv_window_add(sb, my_rsv);
859 
860 	return 0;
861 }
862 
863 /**
864  * 	alloc_new_reservation()--allocate a new reservation window
865  *
866  *		To make a new reservation, we search part of the filesystem
867  *		reservation list (the list that inside the group). We try to
868  *		allocate a new reservation window near the allocation goal,
869  *		or the beginning of the group, if there is no goal.
870  *
871  *		We first find a reservable space after the goal, then from
872  *		there, we check the bitmap for the first free block after
873  *		it. If there is no free block until the end of group, then the
874  *		whole group is full, we failed. Otherwise, check if the free
875  *		block is inside the expected reservable space, if so, we
876  *		succeed.
877  *		If the first free block is outside the reservable space, then
878  *		start from the first free block, we search for next available
879  *		space, and go on.
880  *
881  *	on succeed, a new reservation will be found and inserted into the list
882  *	It contains at least one free block, and it does not overlap with other
883  *	reservation windows.
884  *
885  *	failed: we failed to find a reservation window in this group
886  *
887  *	@rsv: the reservation
888  *
889  *	@grp_goal: The goal (group-relative).  It is where the search for a
890  *		free reservable space should start from.
891  *		if we have a goal(goal >0 ), then start from there,
892  *		no goal(goal = -1), we start from the first block
893  *		of the group.
894  *
895  *	@sb: the super block
896  *	@group: the group we are trying to allocate in
897  *	@bitmap_bh: the block group block bitmap
898  *
899  */
900 static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
901 		ext2_grpblk_t grp_goal, struct super_block *sb,
902 		unsigned int group, struct buffer_head *bitmap_bh)
903 {
904 	struct ext2_reserve_window_node *search_head;
905 	ext2_fsblk_t group_first_block, group_end_block, start_block;
906 	ext2_grpblk_t first_free_block;
907 	struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
908 	unsigned long size;
909 	int ret;
910 	spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
911 
912 	group_first_block = ext2_group_first_block_no(sb, group);
913 	group_end_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
914 
915 	if (grp_goal < 0)
916 		start_block = group_first_block;
917 	else
918 		start_block = grp_goal + group_first_block;
919 
920 	size = my_rsv->rsv_goal_size;
921 
922 	if (!rsv_is_empty(&my_rsv->rsv_window)) {
923 		/*
924 		 * if the old reservation is cross group boundary
925 		 * and if the goal is inside the old reservation window,
926 		 * we will come here when we just failed to allocate from
927 		 * the first part of the window. We still have another part
928 		 * that belongs to the next group. In this case, there is no
929 		 * point to discard our window and try to allocate a new one
930 		 * in this group(which will fail). we should
931 		 * keep the reservation window, just simply move on.
932 		 *
933 		 * Maybe we could shift the start block of the reservation
934 		 * window to the first block of next group.
935 		 */
936 
937 		if ((my_rsv->rsv_start <= group_end_block) &&
938 				(my_rsv->rsv_end > group_end_block) &&
939 				(start_block >= my_rsv->rsv_start))
940 			return -1;
941 
942 		if ((my_rsv->rsv_alloc_hit >
943 		     (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
944 			/*
945 			 * if the previously allocation hit ratio is
946 			 * greater than 1/2, then we double the size of
947 			 * the reservation window the next time,
948 			 * otherwise we keep the same size window
949 			 */
950 			size = size * 2;
951 			if (size > EXT2_MAX_RESERVE_BLOCKS)
952 				size = EXT2_MAX_RESERVE_BLOCKS;
953 			my_rsv->rsv_goal_size= size;
954 		}
955 	}
956 
957 	spin_lock(rsv_lock);
958 	/*
959 	 * shift the search start to the window near the goal block
960 	 */
961 	search_head = search_reserve_window(fs_rsv_root, start_block);
962 
963 	/*
964 	 * find_next_reservable_window() simply finds a reservable window
965 	 * inside the given range(start_block, group_end_block).
966 	 *
967 	 * To make sure the reservation window has a free bit inside it, we
968 	 * need to check the bitmap after we found a reservable window.
969 	 */
970 retry:
971 	ret = find_next_reservable_window(search_head, my_rsv, sb,
972 						start_block, group_end_block);
973 
974 	if (ret == -1) {
975 		if (!rsv_is_empty(&my_rsv->rsv_window))
976 			rsv_window_remove(sb, my_rsv);
977 		spin_unlock(rsv_lock);
978 		return -1;
979 	}
980 
981 	/*
982 	 * On success, find_next_reservable_window() returns the
983 	 * reservation window where there is a reservable space after it.
984 	 * Before we reserve this reservable space, we need
985 	 * to make sure there is at least a free block inside this region.
986 	 *
987 	 * Search the first free bit on the block bitmap.  Search starts from
988 	 * the start block of the reservable space we just found.
989 	 */
990 	spin_unlock(rsv_lock);
991 	first_free_block = bitmap_search_next_usable_block(
992 			my_rsv->rsv_start - group_first_block,
993 			bitmap_bh, group_end_block - group_first_block + 1);
994 
995 	if (first_free_block < 0) {
996 		/*
997 		 * no free block left on the bitmap, no point
998 		 * to reserve the space. return failed.
999 		 */
1000 		spin_lock(rsv_lock);
1001 		if (!rsv_is_empty(&my_rsv->rsv_window))
1002 			rsv_window_remove(sb, my_rsv);
1003 		spin_unlock(rsv_lock);
1004 		return -1;		/* failed */
1005 	}
1006 
1007 	start_block = first_free_block + group_first_block;
1008 	/*
1009 	 * check if the first free block is within the
1010 	 * free space we just reserved
1011 	 */
1012 	if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1013 		return 0;		/* success */
1014 	/*
1015 	 * if the first free bit we found is out of the reservable space
1016 	 * continue search for next reservable space,
1017 	 * start from where the free block is,
1018 	 * we also shift the list head to where we stopped last time
1019 	 */
1020 	search_head = my_rsv;
1021 	spin_lock(rsv_lock);
1022 	goto retry;
1023 }
1024 
1025 /**
1026  * try_to_extend_reservation()
1027  * @my_rsv:		given reservation window
1028  * @sb:			super block
1029  * @size:		the delta to extend
1030  *
1031  * Attempt to expand the reservation window large enough to have
1032  * required number of free blocks
1033  *
1034  * Since ext2_try_to_allocate() will always allocate blocks within
1035  * the reservation window range, if the window size is too small,
1036  * multiple blocks allocation has to stop at the end of the reservation
1037  * window. To make this more efficient, given the total number of
1038  * blocks needed and the current size of the window, we try to
1039  * expand the reservation window size if necessary on a best-effort
1040  * basis before ext2_new_blocks() tries to allocate blocks.
1041  */
1042 static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1043 			struct super_block *sb, int size)
1044 {
1045 	struct ext2_reserve_window_node *next_rsv;
1046 	struct rb_node *next;
1047 	spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1048 
1049 	if (!spin_trylock(rsv_lock))
1050 		return;
1051 
1052 	next = rb_next(&my_rsv->rsv_node);
1053 
1054 	if (!next)
1055 		my_rsv->rsv_end += size;
1056 	else {
1057 		next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1058 
1059 		if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1060 			my_rsv->rsv_end += size;
1061 		else
1062 			my_rsv->rsv_end = next_rsv->rsv_start - 1;
1063 	}
1064 	spin_unlock(rsv_lock);
1065 }
1066 
1067 /**
1068  * ext2_try_to_allocate_with_rsv()
1069  * @sb:			superblock
1070  * @group:		given allocation block group
1071  * @bitmap_bh:		bufferhead holds the block bitmap
1072  * @grp_goal:		given target block within the group
1073  * @count:		target number of blocks to allocate
1074  * @my_rsv:		reservation window
1075  *
1076  * This is the main function used to allocate a new block and its reservation
1077  * window.
1078  *
1079  * Each time when a new block allocation is need, first try to allocate from
1080  * its own reservation.  If it does not have a reservation window, instead of
1081  * looking for a free bit on bitmap first, then look up the reservation list to
1082  * see if it is inside somebody else's reservation window, we try to allocate a
1083  * reservation window for it starting from the goal first. Then do the block
1084  * allocation within the reservation window.
1085  *
1086  * This will avoid keeping on searching the reservation list again and
1087  * again when somebody is looking for a free block (without
1088  * reservation), and there are lots of free blocks, but they are all
1089  * being reserved.
1090  *
1091  * We use a red-black tree for the per-filesystem reservation list.
1092  */
1093 static ext2_grpblk_t
1094 ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1095 			struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1096 			struct ext2_reserve_window_node * my_rsv,
1097 			unsigned long *count)
1098 {
1099 	ext2_fsblk_t group_first_block, group_last_block;
1100 	ext2_grpblk_t ret = 0;
1101 	unsigned long num = *count;
1102 
1103 	/*
1104 	 * we don't deal with reservation when
1105 	 * filesystem is mounted without reservation
1106 	 * or the file is not a regular file
1107 	 * or last attempt to allocate a block with reservation turned on failed
1108 	 */
1109 	if (my_rsv == NULL) {
1110 		return ext2_try_to_allocate(sb, group, bitmap_bh,
1111 						grp_goal, count, NULL);
1112 	}
1113 	/*
1114 	 * grp_goal is a group relative block number (if there is a goal)
1115 	 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1116 	 * first block is a filesystem wide block number
1117 	 * first block is the block number of the first block in this group
1118 	 */
1119 	group_first_block = ext2_group_first_block_no(sb, group);
1120 	group_last_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
1121 
1122 	/*
1123 	 * Basically we will allocate a new block from inode's reservation
1124 	 * window.
1125 	 *
1126 	 * We need to allocate a new reservation window, if:
1127 	 * a) inode does not have a reservation window; or
1128 	 * b) last attempt to allocate a block from existing reservation
1129 	 *    failed; or
1130 	 * c) we come here with a goal and with a reservation window
1131 	 *
1132 	 * We do not need to allocate a new reservation window if we come here
1133 	 * at the beginning with a goal and the goal is inside the window, or
1134 	 * we don't have a goal but already have a reservation window.
1135 	 * then we could go to allocate from the reservation window directly.
1136 	 */
1137 	while (1) {
1138 		if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1139 			!goal_in_my_reservation(&my_rsv->rsv_window,
1140 						grp_goal, group, sb)) {
1141 			if (my_rsv->rsv_goal_size < *count)
1142 				my_rsv->rsv_goal_size = *count;
1143 			ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1144 							group, bitmap_bh);
1145 			if (ret < 0)
1146 				break;			/* failed */
1147 
1148 			if (!goal_in_my_reservation(&my_rsv->rsv_window,
1149 							grp_goal, group, sb))
1150 				grp_goal = -1;
1151 		} else if (grp_goal >= 0) {
1152 			int curr = my_rsv->rsv_end -
1153 					(grp_goal + group_first_block) + 1;
1154 
1155 			if (curr < *count)
1156 				try_to_extend_reservation(my_rsv, sb,
1157 							*count - curr);
1158 		}
1159 
1160 		if ((my_rsv->rsv_start > group_last_block) ||
1161 				(my_rsv->rsv_end < group_first_block)) {
1162 			rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1163 			BUG();
1164 		}
1165 		ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1166 					   &num, &my_rsv->rsv_window);
1167 		if (ret >= 0) {
1168 			my_rsv->rsv_alloc_hit += num;
1169 			*count = num;
1170 			break;				/* succeed */
1171 		}
1172 		num = *count;
1173 	}
1174 	return ret;
1175 }
1176 
1177 /**
1178  * ext2_has_free_blocks()
1179  * @sbi:		in-core super block structure.
1180  *
1181  * Check if filesystem has at least 1 free block available for allocation.
1182  */
1183 static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1184 {
1185 	ext2_fsblk_t free_blocks, root_blocks;
1186 
1187 	free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1188 	root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1189 	if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1190 		!uid_eq(sbi->s_resuid, current_fsuid()) &&
1191 		(gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1192 		 !in_group_p (sbi->s_resgid))) {
1193 		return 0;
1194 	}
1195 	return 1;
1196 }
1197 
1198 /*
1199  * Returns 1 if the passed-in block region is valid; 0 if some part overlaps
1200  * with filesystem metadata blocksi.
1201  */
1202 int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk,
1203 			  unsigned int count)
1204 {
1205 	if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
1206 	    (start_blk + count < start_blk) ||
1207 	    (start_blk > le32_to_cpu(sbi->s_es->s_blocks_count)))
1208 		return 0;
1209 
1210 	/* Ensure we do not step over superblock */
1211 	if ((start_blk <= sbi->s_sb_block) &&
1212 	    (start_blk + count >= sbi->s_sb_block))
1213 		return 0;
1214 
1215 
1216 	return 1;
1217 }
1218 
1219 /*
1220  * ext2_new_blocks() -- core block(s) allocation function
1221  * @inode:		file inode
1222  * @goal:		given target block(filesystem wide)
1223  * @count:		target number of blocks to allocate
1224  * @errp:		error code
1225  *
1226  * ext2_new_blocks uses a goal block to assist allocation.  If the goal is
1227  * free, or there is a free block within 32 blocks of the goal, that block
1228  * is allocated.  Otherwise a forward search is made for a free block; within
1229  * each block group the search first looks for an entire free byte in the block
1230  * bitmap, and then for any free bit if that fails.
1231  * This function also updates quota and i_blocks field.
1232  */
1233 ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1234 		    unsigned long *count, int *errp)
1235 {
1236 	struct buffer_head *bitmap_bh = NULL;
1237 	struct buffer_head *gdp_bh;
1238 	int group_no;
1239 	int goal_group;
1240 	ext2_grpblk_t grp_target_blk;	/* blockgroup relative goal block */
1241 	ext2_grpblk_t grp_alloc_blk;	/* blockgroup-relative allocated block*/
1242 	ext2_fsblk_t ret_block;		/* filesyetem-wide allocated block */
1243 	int bgi;			/* blockgroup iteration index */
1244 	int performed_allocation = 0;
1245 	ext2_grpblk_t free_blocks;	/* number of free blocks in a group */
1246 	struct super_block *sb;
1247 	struct ext2_group_desc *gdp;
1248 	struct ext2_super_block *es;
1249 	struct ext2_sb_info *sbi;
1250 	struct ext2_reserve_window_node *my_rsv = NULL;
1251 	struct ext2_block_alloc_info *block_i;
1252 	unsigned short windowsz = 0;
1253 	unsigned long ngroups;
1254 	unsigned long num = *count;
1255 	int ret;
1256 
1257 	*errp = -ENOSPC;
1258 	sb = inode->i_sb;
1259 
1260 	/*
1261 	 * Check quota for allocation of this block.
1262 	 */
1263 	ret = dquot_alloc_block(inode, num);
1264 	if (ret) {
1265 		*errp = ret;
1266 		return 0;
1267 	}
1268 
1269 	sbi = EXT2_SB(sb);
1270 	es = EXT2_SB(sb)->s_es;
1271 	ext2_debug("goal=%lu.\n", goal);
1272 	/*
1273 	 * Allocate a block from reservation only when
1274 	 * filesystem is mounted with reservation(default,-o reservation), and
1275 	 * it's a regular file, and
1276 	 * the desired window size is greater than 0 (One could use ioctl
1277 	 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1278 	 * reservation on that particular file)
1279 	 */
1280 	block_i = EXT2_I(inode)->i_block_alloc_info;
1281 	if (block_i) {
1282 		windowsz = block_i->rsv_window_node.rsv_goal_size;
1283 		if (windowsz > 0)
1284 			my_rsv = &block_i->rsv_window_node;
1285 	}
1286 
1287 	if (!ext2_has_free_blocks(sbi)) {
1288 		*errp = -ENOSPC;
1289 		goto out;
1290 	}
1291 
1292 	/*
1293 	 * First, test whether the goal block is free.
1294 	 */
1295 	if (goal < le32_to_cpu(es->s_first_data_block) ||
1296 	    goal >= le32_to_cpu(es->s_blocks_count))
1297 		goal = le32_to_cpu(es->s_first_data_block);
1298 	group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1299 			EXT2_BLOCKS_PER_GROUP(sb);
1300 	goal_group = group_no;
1301 retry_alloc:
1302 	gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1303 	if (!gdp)
1304 		goto io_error;
1305 
1306 	free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1307 	/*
1308 	 * if there is not enough free blocks to make a new resevation
1309 	 * turn off reservation for this allocation
1310 	 */
1311 	if (my_rsv && (free_blocks < windowsz)
1312 		&& (free_blocks > 0)
1313 		&& (rsv_is_empty(&my_rsv->rsv_window)))
1314 		my_rsv = NULL;
1315 
1316 	if (free_blocks > 0) {
1317 		grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1318 				EXT2_BLOCKS_PER_GROUP(sb));
1319 		bitmap_bh = read_block_bitmap(sb, group_no);
1320 		if (!bitmap_bh)
1321 			goto io_error;
1322 		grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1323 					bitmap_bh, grp_target_blk,
1324 					my_rsv, &num);
1325 		if (grp_alloc_blk >= 0)
1326 			goto allocated;
1327 	}
1328 
1329 	ngroups = EXT2_SB(sb)->s_groups_count;
1330 	smp_rmb();
1331 
1332 	/*
1333 	 * Now search the rest of the groups.  We assume that
1334 	 * group_no and gdp correctly point to the last group visited.
1335 	 */
1336 	for (bgi = 0; bgi < ngroups; bgi++) {
1337 		group_no++;
1338 		if (group_no >= ngroups)
1339 			group_no = 0;
1340 		gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1341 		if (!gdp)
1342 			goto io_error;
1343 
1344 		free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1345 		/*
1346 		 * skip this group (and avoid loading bitmap) if there
1347 		 * are no free blocks
1348 		 */
1349 		if (!free_blocks)
1350 			continue;
1351 		/*
1352 		 * skip this group if the number of
1353 		 * free blocks is less than half of the reservation
1354 		 * window size.
1355 		 */
1356 		if (my_rsv && (free_blocks <= (windowsz/2)))
1357 			continue;
1358 
1359 		brelse(bitmap_bh);
1360 		bitmap_bh = read_block_bitmap(sb, group_no);
1361 		if (!bitmap_bh)
1362 			goto io_error;
1363 		/*
1364 		 * try to allocate block(s) from this group, without a goal(-1).
1365 		 */
1366 		grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1367 					bitmap_bh, -1, my_rsv, &num);
1368 		if (grp_alloc_blk >= 0)
1369 			goto allocated;
1370 	}
1371 	/*
1372 	 * We may end up a bogus earlier ENOSPC error due to
1373 	 * filesystem is "full" of reservations, but
1374 	 * there maybe indeed free blocks available on disk
1375 	 * In this case, we just forget about the reservations
1376 	 * just do block allocation as without reservations.
1377 	 */
1378 	if (my_rsv) {
1379 		my_rsv = NULL;
1380 		windowsz = 0;
1381 		group_no = goal_group;
1382 		goto retry_alloc;
1383 	}
1384 	/* No space left on the device */
1385 	*errp = -ENOSPC;
1386 	goto out;
1387 
1388 allocated:
1389 
1390 	ext2_debug("using block group %d(%d)\n",
1391 			group_no, gdp->bg_free_blocks_count);
1392 
1393 	ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1394 
1395 	if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1396 	    in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1397 	    in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1398 		      EXT2_SB(sb)->s_itb_per_group) ||
1399 	    in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1400 		      EXT2_SB(sb)->s_itb_per_group)) {
1401 		ext2_error(sb, "ext2_new_blocks",
1402 			    "Allocating block in system zone - "
1403 			    "blocks from "E2FSBLK", length %lu",
1404 			    ret_block, num);
1405 		/*
1406 		 * ext2_try_to_allocate marked the blocks we allocated as in
1407 		 * use.  So we may want to selectively mark some of the blocks
1408 		 * as free
1409 		 */
1410 		goto retry_alloc;
1411 	}
1412 
1413 	performed_allocation = 1;
1414 
1415 	if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1416 		ext2_error(sb, "ext2_new_blocks",
1417 			    "block("E2FSBLK") >= blocks count(%d) - "
1418 			    "block_group = %d, es == %p ", ret_block,
1419 			le32_to_cpu(es->s_blocks_count), group_no, es);
1420 		goto out;
1421 	}
1422 
1423 	group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1424 	percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1425 
1426 	mark_buffer_dirty(bitmap_bh);
1427 	if (sb->s_flags & SB_SYNCHRONOUS)
1428 		sync_dirty_buffer(bitmap_bh);
1429 
1430 	*errp = 0;
1431 	brelse(bitmap_bh);
1432 	if (num < *count) {
1433 		dquot_free_block_nodirty(inode, *count-num);
1434 		mark_inode_dirty(inode);
1435 		*count = num;
1436 	}
1437 	return ret_block;
1438 
1439 io_error:
1440 	*errp = -EIO;
1441 out:
1442 	/*
1443 	 * Undo the block allocation
1444 	 */
1445 	if (!performed_allocation) {
1446 		dquot_free_block_nodirty(inode, *count);
1447 		mark_inode_dirty(inode);
1448 	}
1449 	brelse(bitmap_bh);
1450 	return 0;
1451 }
1452 
1453 ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1454 {
1455 	unsigned long count = 1;
1456 
1457 	return ext2_new_blocks(inode, goal, &count, errp);
1458 }
1459 
1460 #ifdef EXT2FS_DEBUG
1461 
1462 unsigned long ext2_count_free(struct buffer_head *map, unsigned int numchars)
1463 {
1464 	return numchars * BITS_PER_BYTE - memweight(map->b_data, numchars);
1465 }
1466 
1467 #endif  /*  EXT2FS_DEBUG  */
1468 
1469 unsigned long ext2_count_free_blocks (struct super_block * sb)
1470 {
1471 	struct ext2_group_desc * desc;
1472 	unsigned long desc_count = 0;
1473 	int i;
1474 #ifdef EXT2FS_DEBUG
1475 	unsigned long bitmap_count, x;
1476 	struct ext2_super_block *es;
1477 
1478 	es = EXT2_SB(sb)->s_es;
1479 	desc_count = 0;
1480 	bitmap_count = 0;
1481 	desc = NULL;
1482 	for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1483 		struct buffer_head *bitmap_bh;
1484 		desc = ext2_get_group_desc (sb, i, NULL);
1485 		if (!desc)
1486 			continue;
1487 		desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1488 		bitmap_bh = read_block_bitmap(sb, i);
1489 		if (!bitmap_bh)
1490 			continue;
1491 
1492 		x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1493 		printk ("group %d: stored = %d, counted = %lu\n",
1494 			i, le16_to_cpu(desc->bg_free_blocks_count), x);
1495 		bitmap_count += x;
1496 		brelse(bitmap_bh);
1497 	}
1498 	printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1499 		(long)le32_to_cpu(es->s_free_blocks_count),
1500 		desc_count, bitmap_count);
1501 	return bitmap_count;
1502 #else
1503         for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1504                 desc = ext2_get_group_desc (sb, i, NULL);
1505                 if (!desc)
1506                         continue;
1507                 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1508 	}
1509 	return desc_count;
1510 #endif
1511 }
1512 
1513 static inline int test_root(int a, int b)
1514 {
1515 	int num = b;
1516 
1517 	while (a > num)
1518 		num *= b;
1519 	return num == a;
1520 }
1521 
1522 static int ext2_group_sparse(int group)
1523 {
1524 	if (group <= 1)
1525 		return 1;
1526 	return (test_root(group, 3) || test_root(group, 5) ||
1527 		test_root(group, 7));
1528 }
1529 
1530 /**
1531  *	ext2_bg_has_super - number of blocks used by the superblock in group
1532  *	@sb: superblock for filesystem
1533  *	@group: group number to check
1534  *
1535  *	Return the number of blocks used by the superblock (primary or backup)
1536  *	in this group.  Currently this will be only 0 or 1.
1537  */
1538 int ext2_bg_has_super(struct super_block *sb, int group)
1539 {
1540 	if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1541 	    !ext2_group_sparse(group))
1542 		return 0;
1543 	return 1;
1544 }
1545 
1546 /**
1547  *	ext2_bg_num_gdb - number of blocks used by the group table in group
1548  *	@sb: superblock for filesystem
1549  *	@group: group number to check
1550  *
1551  *	Return the number of blocks used by the group descriptor table
1552  *	(primary or backup) in this group.  In the future there may be a
1553  *	different number of descriptor blocks in each group.
1554  */
1555 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1556 {
1557 	return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
1558 }
1559 
1560