xref: /linux/fs/ext2/balloc.c (revision 6e7fd890f1d6ac83805409e9c346240de2705584)
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 struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
40 					     unsigned int block_group,
41 					     struct buffer_head ** bh)
42 {
43 	unsigned long group_desc;
44 	unsigned long offset;
45 	struct ext2_group_desc * desc;
46 	struct ext2_sb_info *sbi = EXT2_SB(sb);
47 
48 	if (block_group >= sbi->s_groups_count) {
49 		WARN(1, "block_group >= groups_count - "
50 		     "block_group = %d, groups_count = %lu",
51 		     block_group, sbi->s_groups_count);
52 
53 		return NULL;
54 	}
55 
56 	group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
57 	offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
58 	if (!sbi->s_group_desc[group_desc]) {
59 		WARN(1, "Group descriptor not loaded - "
60 		     "block_group = %d, group_desc = %lu, desc = %lu",
61 		      block_group, group_desc, offset);
62 		return NULL;
63 	}
64 
65 	desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
66 	if (bh)
67 		*bh = sbi->s_group_desc[group_desc];
68 	return desc + offset;
69 }
70 
71 static int ext2_valid_block_bitmap(struct super_block *sb,
72 					struct ext2_group_desc *desc,
73 					unsigned int block_group,
74 					struct buffer_head *bh)
75 {
76 	ext2_grpblk_t offset;
77 	ext2_grpblk_t next_zero_bit;
78 	ext2_fsblk_t bitmap_blk;
79 	ext2_fsblk_t group_first_block;
80 	ext2_grpblk_t max_bit;
81 
82 	group_first_block = ext2_group_first_block_no(sb, block_group);
83 	max_bit = ext2_group_last_block_no(sb, block_group) - group_first_block;
84 
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 (offset < 0 || offset > max_bit ||
89 	    !ext2_test_bit(offset, bh->b_data))
90 		/* bad block bitmap */
91 		goto err_out;
92 
93 	/* check whether the inode bitmap block number is set */
94 	bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
95 	offset = bitmap_blk - group_first_block;
96 	if (offset < 0 || offset > max_bit ||
97 	    !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 	if (offset < 0 || offset > max_bit ||
105 	    offset + EXT2_SB(sb)->s_itb_per_group - 1 > max_bit)
106 		goto err_out;
107 	next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
108 				offset + EXT2_SB(sb)->s_itb_per_group,
109 				offset);
110 	if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
111 		/* good bitmap for inode tables */
112 		return 1;
113 
114 err_out:
115 	ext2_error(sb, __func__,
116 			"Invalid block bitmap - "
117 			"block_group = %d, block = %lu",
118 			block_group, bitmap_blk);
119 	return 0;
120 }
121 
122 /*
123  * Read the bitmap for a given block_group,and validate the
124  * bits for block/inode/inode tables are set in the bitmaps
125  *
126  * Return buffer_head on success or NULL in case of failure.
127  */
128 static struct buffer_head *
129 read_block_bitmap(struct super_block *sb, unsigned int block_group)
130 {
131 	struct ext2_group_desc * desc;
132 	struct buffer_head * bh = NULL;
133 	ext2_fsblk_t bitmap_blk;
134 	int ret;
135 
136 	desc = ext2_get_group_desc(sb, block_group, NULL);
137 	if (!desc)
138 		return NULL;
139 	bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
140 	bh = sb_getblk(sb, bitmap_blk);
141 	if (unlikely(!bh)) {
142 		ext2_error(sb, __func__,
143 			    "Cannot read block bitmap - "
144 			    "block_group = %d, block_bitmap = %u",
145 			    block_group, le32_to_cpu(desc->bg_block_bitmap));
146 		return NULL;
147 	}
148 	ret = bh_read(bh, 0);
149 	if (ret > 0)
150 		return bh;
151 	if (ret < 0) {
152 		brelse(bh);
153 		ext2_error(sb, __func__,
154 			    "Cannot read block bitmap - "
155 			    "block_group = %d, block_bitmap = %u",
156 			    block_group, le32_to_cpu(desc->bg_block_bitmap));
157 		return NULL;
158 	}
159 
160 	ext2_valid_block_bitmap(sb, desc, block_group, bh);
161 	/*
162 	 * file system mounted not to panic on error, continue with corrupt
163 	 * bitmap
164 	 */
165 	return bh;
166 }
167 
168 static void group_adjust_blocks(struct super_block *sb, int group_no,
169 	struct ext2_group_desc *desc, struct buffer_head *bh, int count)
170 {
171 	if (count) {
172 		struct ext2_sb_info *sbi = EXT2_SB(sb);
173 		unsigned free_blocks;
174 
175 		spin_lock(sb_bgl_lock(sbi, group_no));
176 		free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
177 		desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
178 		spin_unlock(sb_bgl_lock(sbi, group_no));
179 		mark_buffer_dirty(bh);
180 	}
181 }
182 
183 /*
184  * The reservation window structure operations
185  * --------------------------------------------
186  * Operations include:
187  * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
188  *
189  * We use a red-black tree to represent per-filesystem reservation
190  * windows.
191  *
192  */
193 
194 /**
195  * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
196  * @root:		root of per-filesystem reservation rb tree
197  * @verbose:		verbose mode
198  * @fn:			function which wishes to dump the reservation map
199  *
200  * If verbose is turned on, it will print the whole block reservation
201  * windows(start, end). Otherwise, it will only print out the "bad" windows,
202  * those windows that overlap with their immediate neighbors.
203  */
204 #if 1
205 static void __rsv_window_dump(struct rb_root *root, int verbose,
206 			      const char *fn)
207 {
208 	struct rb_node *n;
209 	struct ext2_reserve_window_node *rsv, *prev;
210 	int bad;
211 
212 restart:
213 	n = rb_first(root);
214 	bad = 0;
215 	prev = NULL;
216 
217 	printk("Block Allocation Reservation Windows Map (%s):\n", fn);
218 	while (n) {
219 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
220 		if (verbose)
221 			printk("reservation window 0x%p "
222 				"start: %lu, end: %lu\n",
223 				rsv, rsv->rsv_start, rsv->rsv_end);
224 		if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
225 			printk("Bad reservation %p (start >= end)\n",
226 			       rsv);
227 			bad = 1;
228 		}
229 		if (prev && prev->rsv_end >= rsv->rsv_start) {
230 			printk("Bad reservation %p (prev->end >= start)\n",
231 			       rsv);
232 			bad = 1;
233 		}
234 		if (bad) {
235 			if (!verbose) {
236 				printk("Restarting reservation walk in verbose mode\n");
237 				verbose = 1;
238 				goto restart;
239 			}
240 		}
241 		n = rb_next(n);
242 		prev = rsv;
243 	}
244 	printk("Window map complete.\n");
245 	BUG_ON(bad);
246 }
247 #define rsv_window_dump(root, verbose) \
248 	__rsv_window_dump((root), (verbose), __func__)
249 #else
250 #define rsv_window_dump(root, verbose) do {} while (0)
251 #endif
252 
253 /**
254  * goal_in_my_reservation()
255  * @rsv:		inode's reservation window
256  * @grp_goal:		given goal block relative to the allocation block group
257  * @group:		the current allocation block group
258  * @sb:			filesystem super block
259  *
260  * Test if the given goal block (group relative) is within the file's
261  * own block reservation window range.
262  *
263  * If the reservation window is outside the goal allocation group, return 0;
264  * grp_goal (given goal block) could be -1, which means no specific
265  * goal block. In this case, always return 1.
266  * If the goal block is within the reservation window, return 1;
267  * otherwise, return 0;
268  */
269 static int
270 goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
271 			unsigned int group, struct super_block * sb)
272 {
273 	ext2_fsblk_t group_first_block, group_last_block;
274 
275 	group_first_block = ext2_group_first_block_no(sb, group);
276 	group_last_block = ext2_group_last_block_no(sb, group);
277 
278 	if ((rsv->_rsv_start > group_last_block) ||
279 	    (rsv->_rsv_end < group_first_block))
280 		return 0;
281 	if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
282 		|| (grp_goal + group_first_block > rsv->_rsv_end)))
283 		return 0;
284 	return 1;
285 }
286 
287 /**
288  * search_reserve_window()
289  * @root:		root of reservation tree
290  * @goal:		target allocation block
291  *
292  * Find the reserved window which includes the goal, or the previous one
293  * if the goal is not in any window.
294  * Returns NULL if there are no windows or if all windows start after the goal.
295  */
296 static struct ext2_reserve_window_node *
297 search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
298 {
299 	struct rb_node *n = root->rb_node;
300 	struct ext2_reserve_window_node *rsv;
301 
302 	if (!n)
303 		return NULL;
304 
305 	do {
306 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
307 
308 		if (goal < rsv->rsv_start)
309 			n = n->rb_left;
310 		else if (goal > rsv->rsv_end)
311 			n = n->rb_right;
312 		else
313 			return rsv;
314 	} while (n);
315 	/*
316 	 * We've fallen off the end of the tree: the goal wasn't inside
317 	 * any particular node.  OK, the previous node must be to one
318 	 * side of the interval containing the goal.  If it's the RHS,
319 	 * we need to back up one.
320 	 */
321 	if (rsv->rsv_start > goal) {
322 		n = rb_prev(&rsv->rsv_node);
323 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
324 	}
325 	return rsv;
326 }
327 
328 /*
329  * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
330  * @sb:			super block
331  * @rsv:		reservation window to add
332  *
333  * Must be called with rsv_lock held.
334  */
335 void ext2_rsv_window_add(struct super_block *sb,
336 		    struct ext2_reserve_window_node *rsv)
337 {
338 	struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
339 	struct rb_node *node = &rsv->rsv_node;
340 	ext2_fsblk_t start = rsv->rsv_start;
341 
342 	struct rb_node ** p = &root->rb_node;
343 	struct rb_node * parent = NULL;
344 	struct ext2_reserve_window_node *this;
345 
346 	while (*p)
347 	{
348 		parent = *p;
349 		this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
350 
351 		if (start < this->rsv_start)
352 			p = &(*p)->rb_left;
353 		else if (start > this->rsv_end)
354 			p = &(*p)->rb_right;
355 		else {
356 			rsv_window_dump(root, 1);
357 			BUG();
358 		}
359 	}
360 
361 	rb_link_node(node, parent, p);
362 	rb_insert_color(node, root);
363 }
364 
365 /**
366  * rsv_window_remove() -- unlink a window from the reservation rb tree
367  * @sb:			super block
368  * @rsv:		reservation window to remove
369  *
370  * Mark the block reservation window as not allocated, and unlink it
371  * from the filesystem reservation window rb tree. Must be called with
372  * rsv_lock held.
373  */
374 static void rsv_window_remove(struct super_block *sb,
375 			      struct ext2_reserve_window_node *rsv)
376 {
377 	rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
378 	rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
379 	rsv->rsv_alloc_hit = 0;
380 	rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
381 }
382 
383 /*
384  * rsv_is_empty() -- Check if the reservation window is allocated.
385  * @rsv:		given reservation window to check
386  *
387  * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
388  */
389 static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
390 {
391 	/* a valid reservation end block could not be 0 */
392 	return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
393 }
394 
395 /**
396  * ext2_init_block_alloc_info()
397  * @inode:		file inode structure
398  *
399  * Allocate and initialize the  reservation window structure, and
400  * link the window to the ext2 inode structure at last
401  *
402  * The reservation window structure is only dynamically allocated
403  * and linked to ext2 inode the first time the open file
404  * needs a new block. So, before every ext2_new_block(s) call, for
405  * regular files, we should check whether the reservation window
406  * structure exists or not. In the latter case, this function is called.
407  * Fail to do so will result in block reservation being turned off for that
408  * open file.
409  *
410  * This function is called from ext2_get_blocks_handle(), also called
411  * when setting the reservation window size through ioctl before the file
412  * is open for write (needs block allocation).
413  *
414  * Needs truncate_mutex protection prior to calling this function.
415  */
416 void ext2_init_block_alloc_info(struct inode *inode)
417 {
418 	struct ext2_inode_info *ei = EXT2_I(inode);
419 	struct ext2_block_alloc_info *block_i;
420 	struct super_block *sb = inode->i_sb;
421 
422 	block_i = kmalloc(sizeof(*block_i), GFP_KERNEL);
423 	if (block_i) {
424 		struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
425 
426 		rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
427 		rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
428 
429 	 	/*
430 		 * if filesystem is mounted with NORESERVATION, the goal
431 		 * reservation window size is set to zero to indicate
432 		 * block reservation is off
433 		 */
434 		if (!test_opt(sb, RESERVATION))
435 			rsv->rsv_goal_size = 0;
436 		else
437 			rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
438 		rsv->rsv_alloc_hit = 0;
439 		block_i->last_alloc_logical_block = 0;
440 		block_i->last_alloc_physical_block = 0;
441 	}
442 	ei->i_block_alloc_info = block_i;
443 }
444 
445 /**
446  * ext2_discard_reservation()
447  * @inode:		inode
448  *
449  * Discard(free) block reservation window on last file close, or truncate
450  * or at last iput().
451  *
452  * It is being called in three cases:
453  * 	ext2_release_file(): last writer closes the file
454  * 	ext2_clear_inode(): last iput(), when nobody links to this file.
455  * 	ext2_truncate(): when the block indirect map is about to change.
456  */
457 void ext2_discard_reservation(struct inode *inode)
458 {
459 	struct ext2_inode_info *ei = EXT2_I(inode);
460 	struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
461 	struct ext2_reserve_window_node *rsv;
462 	spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
463 
464 	if (!block_i)
465 		return;
466 
467 	rsv = &block_i->rsv_window_node;
468 	if (!rsv_is_empty(&rsv->rsv_window)) {
469 		spin_lock(rsv_lock);
470 		if (!rsv_is_empty(&rsv->rsv_window))
471 			rsv_window_remove(inode->i_sb, rsv);
472 		spin_unlock(rsv_lock);
473 	}
474 }
475 
476 /**
477  * ext2_free_blocks() -- Free given blocks and update quota and i_blocks
478  * @inode:		inode
479  * @block:		start physical block to free
480  * @count:		number of blocks to free
481  */
482 void ext2_free_blocks(struct inode * inode, ext2_fsblk_t block,
483 		      unsigned long count)
484 {
485 	struct buffer_head *bitmap_bh = NULL;
486 	struct buffer_head * bh2;
487 	unsigned long block_group;
488 	unsigned long bit;
489 	unsigned long i;
490 	unsigned long overflow;
491 	struct super_block * sb = inode->i_sb;
492 	struct ext2_sb_info * sbi = EXT2_SB(sb);
493 	struct ext2_group_desc * desc;
494 	struct ext2_super_block * es = sbi->s_es;
495 	unsigned freed = 0, group_freed;
496 
497 	if (!ext2_data_block_valid(sbi, block, count)) {
498 		ext2_error (sb, "ext2_free_blocks",
499 			    "Freeing blocks not in datazone - "
500 			    "block = %lu, count = %lu", block, count);
501 		goto error_return;
502 	}
503 
504 	ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
505 
506 do_more:
507 	overflow = 0;
508 	block_group = (block - le32_to_cpu(es->s_first_data_block)) /
509 		      EXT2_BLOCKS_PER_GROUP(sb);
510 	bit = (block - le32_to_cpu(es->s_first_data_block)) %
511 		      EXT2_BLOCKS_PER_GROUP(sb);
512 	/*
513 	 * Check to see if we are freeing blocks across a group
514 	 * boundary.
515 	 */
516 	if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
517 		overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
518 		count -= overflow;
519 	}
520 	brelse(bitmap_bh);
521 	bitmap_bh = read_block_bitmap(sb, block_group);
522 	if (!bitmap_bh)
523 		goto error_return;
524 
525 	desc = ext2_get_group_desc (sb, block_group, &bh2);
526 	if (!desc)
527 		goto error_return;
528 
529 	if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
530 	    in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
531 	    in_range (block, le32_to_cpu(desc->bg_inode_table),
532 		      sbi->s_itb_per_group) ||
533 	    in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
534 		      sbi->s_itb_per_group)) {
535 		ext2_error (sb, "ext2_free_blocks",
536 			    "Freeing blocks in system zones - "
537 			    "Block = %lu, count = %lu",
538 			    block, count);
539 		goto error_return;
540 	}
541 
542 	for (i = 0, group_freed = 0; i < count; i++) {
543 		if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
544 						bit + i, bitmap_bh->b_data)) {
545 			ext2_error(sb, __func__,
546 				"bit already cleared for block %lu", block + i);
547 		} else {
548 			group_freed++;
549 		}
550 	}
551 
552 	mark_buffer_dirty(bitmap_bh);
553 	if (sb->s_flags & SB_SYNCHRONOUS)
554 		sync_dirty_buffer(bitmap_bh);
555 
556 	group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
557 	freed += group_freed;
558 
559 	if (overflow) {
560 		block += count;
561 		count = overflow;
562 		goto do_more;
563 	}
564 error_return:
565 	brelse(bitmap_bh);
566 	if (freed) {
567 		percpu_counter_add(&sbi->s_freeblocks_counter, freed);
568 		dquot_free_block_nodirty(inode, freed);
569 		mark_inode_dirty(inode);
570 	}
571 }
572 
573 /**
574  * bitmap_search_next_usable_block()
575  * @start:		the starting block (group relative) of the search
576  * @bh:			bufferhead contains the block group bitmap
577  * @maxblocks:		the ending block (group relative) of the reservation
578  *
579  * The bitmap search --- search forward through the actual bitmap on disk until
580  * we find a bit free.
581  */
582 static ext2_grpblk_t
583 bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
584 					ext2_grpblk_t maxblocks)
585 {
586 	ext2_grpblk_t next;
587 
588 	next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
589 	if (next >= maxblocks)
590 		return -1;
591 	return next;
592 }
593 
594 /**
595  * find_next_usable_block()
596  * @start:		the starting block (group relative) to find next
597  * 			allocatable block in bitmap.
598  * @bh:			bufferhead contains the block group bitmap
599  * @maxblocks:		the ending block (group relative) for the search
600  *
601  * Find an allocatable block in a bitmap.  We perform the "most
602  * appropriate allocation" algorithm of looking for a free block near
603  * the initial goal; then for a free byte somewhere in the bitmap;
604  * then for any free bit in the bitmap.
605  */
606 static ext2_grpblk_t
607 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
608 {
609 	ext2_grpblk_t here, next;
610 	char *p, *r;
611 
612 	if (start > 0) {
613 		/*
614 		 * The goal was occupied; search forward for a free
615 		 * block within the next XX blocks.
616 		 *
617 		 * end_goal is more or less random, but it has to be
618 		 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
619 		 * next 64-bit boundary is simple..
620 		 */
621 		ext2_grpblk_t end_goal = (start + 63) & ~63;
622 		if (end_goal > maxblocks)
623 			end_goal = maxblocks;
624 		here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
625 		if (here < end_goal)
626 			return here;
627 		ext2_debug("Bit not found near goal\n");
628 	}
629 
630 	here = start;
631 	if (here < 0)
632 		here = 0;
633 
634 	p = ((char *)bh->b_data) + (here >> 3);
635 	r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
636 	next = (r - ((char *)bh->b_data)) << 3;
637 
638 	if (next < maxblocks && next >= here)
639 		return next;
640 
641 	here = bitmap_search_next_usable_block(here, bh, maxblocks);
642 	return here;
643 }
644 
645 /**
646  * ext2_try_to_allocate()
647  * @sb:			superblock
648  * @group:		given allocation block group
649  * @bitmap_bh:		bufferhead holds the block bitmap
650  * @grp_goal:		given target block within the group
651  * @count:		target number of blocks to allocate
652  * @my_rsv:		reservation window
653  *
654  * Attempt to allocate blocks within a give range. Set the range of allocation
655  * first, then find the first free bit(s) from the bitmap (within the range),
656  * and at last, allocate the blocks by claiming the found free bit as allocated.
657  *
658  * To set the range of this allocation:
659  * 	if there is a reservation window, only try to allocate block(s)
660  * 	from the file's own reservation window;
661  * 	Otherwise, the allocation range starts from the give goal block,
662  * 	ends at the block group's last block.
663  *
664  * If we failed to allocate the desired block then we may end up crossing to a
665  * new bitmap.
666  */
667 static int
668 ext2_try_to_allocate(struct super_block *sb, int group,
669 			struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
670 			unsigned long *count,
671 			struct ext2_reserve_window *my_rsv)
672 {
673 	ext2_fsblk_t group_first_block = ext2_group_first_block_no(sb, group);
674 	ext2_fsblk_t group_last_block = ext2_group_last_block_no(sb, group);
675 	ext2_grpblk_t start, end;
676 	unsigned long num = 0;
677 
678 	start = 0;
679 	end = group_last_block - group_first_block + 1;
680 	/* we do allocation within the reservation window if we have a window */
681 	if (my_rsv) {
682 		if (my_rsv->_rsv_start >= group_first_block)
683 			start = my_rsv->_rsv_start - group_first_block;
684 		if (my_rsv->_rsv_end < group_last_block)
685 			end = my_rsv->_rsv_end - group_first_block + 1;
686 		if (grp_goal < start || grp_goal >= end)
687 			grp_goal = -1;
688 	}
689 	BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
690 
691 	if (grp_goal < 0) {
692 		grp_goal = find_next_usable_block(start, bitmap_bh, end);
693 		if (grp_goal < 0)
694 			goto fail_access;
695 		if (!my_rsv) {
696 			int i;
697 
698 			for (i = 0; i < 7 && grp_goal > start &&
699 					!ext2_test_bit(grp_goal - 1,
700 					     		bitmap_bh->b_data);
701 			     		i++, grp_goal--)
702 				;
703 		}
704 	}
705 
706 	for (; num < *count && grp_goal < end; grp_goal++) {
707 		if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
708 					grp_goal, bitmap_bh->b_data)) {
709 			if (num == 0)
710 				continue;
711 			break;
712 		}
713 		num++;
714 	}
715 
716 	if (num == 0)
717 		goto fail_access;
718 
719 	*count = num;
720 	return grp_goal - num;
721 fail_access:
722 	return -1;
723 }
724 
725 /**
726  * find_next_reservable_window - Find a reservable space within the given range.
727  * @search_head: The list to search.
728  * @my_rsv: The reservation we're currently using.
729  * @sb: The super block.
730  * @start_block: The first block we consider to start the real search from
731  * @last_block: The maximum block number that our goal reservable space
732  *	could start from.
733  *
734  * It does not allocate the reservation window: alloc_new_reservation()
735  * will do the work later.
736  *
737  * We search the given range, rather than the whole reservation double
738  * linked list, (start_block, last_block) to find a free region that is
739  * of my size and has not been reserved.
740  *
741  * @search_head is not necessarily the list head of the whole filesystem.
742  * We have both head and @start_block to assist the search for the
743  * reservable space. The list starts from head, but we will shift to
744  * the place where start_block is, then start from there, when looking
745  * for a reservable space.
746  *
747  * @last_block is normally the last block in this group. The search will end
748  * when we found the start of next possible reservable space is out
749  * of this boundary.  This could handle the cross boundary reservation
750  * window request.
751  *
752  * Return: -1 if we could not find a range of sufficient size.  If we could,
753  * return 0 and fill in @my_rsv with the range information.
754  */
755 static int find_next_reservable_window(
756 				struct ext2_reserve_window_node *search_head,
757 				struct ext2_reserve_window_node *my_rsv,
758 				struct super_block * sb,
759 				ext2_fsblk_t start_block,
760 				ext2_fsblk_t last_block)
761 {
762 	struct rb_node *next;
763 	struct ext2_reserve_window_node *rsv, *prev;
764 	ext2_fsblk_t cur;
765 	int size = my_rsv->rsv_goal_size;
766 
767 	/* TODO: make the start of the reservation window byte-aligned */
768 	/* cur = *start_block & ~7;*/
769 	cur = start_block;
770 	rsv = search_head;
771 	if (!rsv)
772 		return -1;
773 
774 	while (1) {
775 		if (cur <= rsv->rsv_end)
776 			cur = rsv->rsv_end + 1;
777 
778 		/* TODO?
779 		 * in the case we could not find a reservable space
780 		 * that is what is expected, during the re-search, we could
781 		 * remember what's the largest reservable space we could have
782 		 * and return that one.
783 		 *
784 		 * For now it will fail if we could not find the reservable
785 		 * space with expected-size (or more)...
786 		 */
787 		if (cur > last_block)
788 			return -1;		/* fail */
789 
790 		prev = rsv;
791 		next = rb_next(&rsv->rsv_node);
792 		rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
793 
794 		/*
795 		 * Reached the last reservation, we can just append to the
796 		 * previous one.
797 		 */
798 		if (!next)
799 			break;
800 
801 		if (cur + size <= rsv->rsv_start) {
802 			/*
803 			 * Found a reserveable space big enough.  We could
804 			 * have a reservation across the group boundary here
805 		 	 */
806 			break;
807 		}
808 	}
809 	/*
810 	 * we come here either :
811 	 * when we reach the end of the whole list,
812 	 * and there is empty reservable space after last entry in the list.
813 	 * append it to the end of the list.
814 	 *
815 	 * or we found one reservable space in the middle of the list,
816 	 * return the reservation window that we could append to.
817 	 * succeed.
818 	 */
819 
820 	if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
821 		rsv_window_remove(sb, my_rsv);
822 
823 	/*
824 	 * Let's book the whole available window for now.  We will check the
825 	 * disk bitmap later and then, if there are free blocks then we adjust
826 	 * the window size if it's larger than requested.
827 	 * Otherwise, we will remove this node from the tree next time
828 	 * call find_next_reservable_window.
829 	 */
830 	my_rsv->rsv_start = cur;
831 	my_rsv->rsv_end = cur + size - 1;
832 	my_rsv->rsv_alloc_hit = 0;
833 
834 	if (prev != my_rsv)
835 		ext2_rsv_window_add(sb, my_rsv);
836 
837 	return 0;
838 }
839 
840 /**
841  * alloc_new_reservation - Allocate a new reservation window.
842  * @my_rsv: The reservation we're currently using.
843  * @grp_goal: The goal block relative to the start of the group.
844  * @sb: The super block.
845  * @group: The group we are trying to allocate in.
846  * @bitmap_bh: The block group block bitmap.
847  *
848  * To make a new reservation, we search part of the filesystem reservation
849  * list (the list inside the group). We try to allocate a new
850  * reservation window near @grp_goal, or the beginning of the
851  * group, if @grp_goal is negative.
852  *
853  * We first find a reservable space after the goal, then from there,
854  * we check the bitmap for the first free block after it. If there is
855  * no free block until the end of group, then the whole group is full,
856  * we failed. Otherwise, check if the free block is inside the expected
857  * reservable space, if so, we succeed.
858  *
859  * If the first free block is outside the reservable space, then start
860  * from the first free block, we search for next available space, and
861  * go on.
862  *
863  * on succeed, a new reservation will be found and inserted into the
864  * list. It contains at least one free block, and it does not overlap
865  * with other reservation windows.
866  *
867  * Return: 0 on success, -1 if we failed to find a reservation window
868  * in this group
869  */
870 static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
871 		ext2_grpblk_t grp_goal, struct super_block *sb,
872 		unsigned int group, struct buffer_head *bitmap_bh)
873 {
874 	struct ext2_reserve_window_node *search_head;
875 	ext2_fsblk_t group_first_block, group_end_block, start_block;
876 	ext2_grpblk_t first_free_block;
877 	struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
878 	unsigned long size;
879 	int ret;
880 	spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
881 
882 	group_first_block = ext2_group_first_block_no(sb, group);
883 	group_end_block = ext2_group_last_block_no(sb, group);
884 
885 	if (grp_goal < 0)
886 		start_block = group_first_block;
887 	else
888 		start_block = grp_goal + group_first_block;
889 
890 	size = my_rsv->rsv_goal_size;
891 
892 	if (!rsv_is_empty(&my_rsv->rsv_window)) {
893 		/*
894 		 * if the old reservation is cross group boundary
895 		 * and if the goal is inside the old reservation window,
896 		 * we will come here when we just failed to allocate from
897 		 * the first part of the window. We still have another part
898 		 * that belongs to the next group. In this case, there is no
899 		 * point to discard our window and try to allocate a new one
900 		 * in this group(which will fail). we should
901 		 * keep the reservation window, just simply move on.
902 		 *
903 		 * Maybe we could shift the start block of the reservation
904 		 * window to the first block of next group.
905 		 */
906 
907 		if ((my_rsv->rsv_start <= group_end_block) &&
908 				(my_rsv->rsv_end > group_end_block) &&
909 				(start_block >= my_rsv->rsv_start))
910 			return -1;
911 
912 		if ((my_rsv->rsv_alloc_hit >
913 		     (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
914 			/*
915 			 * if the previously allocation hit ratio is
916 			 * greater than 1/2, then we double the size of
917 			 * the reservation window the next time,
918 			 * otherwise we keep the same size window
919 			 */
920 			size = size * 2;
921 			if (size > EXT2_MAX_RESERVE_BLOCKS)
922 				size = EXT2_MAX_RESERVE_BLOCKS;
923 			my_rsv->rsv_goal_size= size;
924 		}
925 	}
926 
927 	spin_lock(rsv_lock);
928 	/*
929 	 * shift the search start to the window near the goal block
930 	 */
931 	search_head = search_reserve_window(fs_rsv_root, start_block);
932 
933 	/*
934 	 * find_next_reservable_window() simply finds a reservable window
935 	 * inside the given range(start_block, group_end_block).
936 	 *
937 	 * To make sure the reservation window has a free bit inside it, we
938 	 * need to check the bitmap after we found a reservable window.
939 	 */
940 retry:
941 	ret = find_next_reservable_window(search_head, my_rsv, sb,
942 						start_block, group_end_block);
943 
944 	if (ret == -1) {
945 		if (!rsv_is_empty(&my_rsv->rsv_window))
946 			rsv_window_remove(sb, my_rsv);
947 		spin_unlock(rsv_lock);
948 		return -1;
949 	}
950 
951 	/*
952 	 * On success, find_next_reservable_window() returns the
953 	 * reservation window where there is a reservable space after it.
954 	 * Before we reserve this reservable space, we need
955 	 * to make sure there is at least a free block inside this region.
956 	 *
957 	 * Search the first free bit on the block bitmap.  Search starts from
958 	 * the start block of the reservable space we just found.
959 	 */
960 	spin_unlock(rsv_lock);
961 	first_free_block = bitmap_search_next_usable_block(
962 			my_rsv->rsv_start - group_first_block,
963 			bitmap_bh, group_end_block - group_first_block + 1);
964 
965 	if (first_free_block < 0) {
966 		/*
967 		 * no free block left on the bitmap, no point
968 		 * to reserve the space. return failed.
969 		 */
970 		spin_lock(rsv_lock);
971 		if (!rsv_is_empty(&my_rsv->rsv_window))
972 			rsv_window_remove(sb, my_rsv);
973 		spin_unlock(rsv_lock);
974 		return -1;		/* failed */
975 	}
976 
977 	start_block = first_free_block + group_first_block;
978 	/*
979 	 * check if the first free block is within the
980 	 * free space we just reserved
981 	 */
982 	if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
983 		return 0;		/* success */
984 	/*
985 	 * if the first free bit we found is out of the reservable space
986 	 * continue search for next reservable space,
987 	 * start from where the free block is,
988 	 * we also shift the list head to where we stopped last time
989 	 */
990 	search_head = my_rsv;
991 	spin_lock(rsv_lock);
992 	goto retry;
993 }
994 
995 /**
996  * try_to_extend_reservation()
997  * @my_rsv:		given reservation window
998  * @sb:			super block
999  * @size:		the delta to extend
1000  *
1001  * Attempt to expand the reservation window large enough to have
1002  * required number of free blocks
1003  *
1004  * Since ext2_try_to_allocate() will always allocate blocks within
1005  * the reservation window range, if the window size is too small,
1006  * multiple blocks allocation has to stop at the end of the reservation
1007  * window. To make this more efficient, given the total number of
1008  * blocks needed and the current size of the window, we try to
1009  * expand the reservation window size if necessary on a best-effort
1010  * basis before ext2_new_blocks() tries to allocate blocks.
1011  */
1012 static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1013 			struct super_block *sb, int size)
1014 {
1015 	struct ext2_reserve_window_node *next_rsv;
1016 	struct rb_node *next;
1017 	spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1018 
1019 	if (!spin_trylock(rsv_lock))
1020 		return;
1021 
1022 	next = rb_next(&my_rsv->rsv_node);
1023 
1024 	if (!next)
1025 		my_rsv->rsv_end += size;
1026 	else {
1027 		next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1028 
1029 		if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1030 			my_rsv->rsv_end += size;
1031 		else
1032 			my_rsv->rsv_end = next_rsv->rsv_start - 1;
1033 	}
1034 	spin_unlock(rsv_lock);
1035 }
1036 
1037 /**
1038  * ext2_try_to_allocate_with_rsv()
1039  * @sb:			superblock
1040  * @group:		given allocation block group
1041  * @bitmap_bh:		bufferhead holds the block bitmap
1042  * @grp_goal:		given target block within the group
1043  * @count:		target number of blocks to allocate
1044  * @my_rsv:		reservation window
1045  *
1046  * This is the main function used to allocate a new block and its reservation
1047  * window.
1048  *
1049  * Each time when a new block allocation is need, first try to allocate from
1050  * its own reservation.  If it does not have a reservation window, instead of
1051  * looking for a free bit on bitmap first, then look up the reservation list to
1052  * see if it is inside somebody else's reservation window, we try to allocate a
1053  * reservation window for it starting from the goal first. Then do the block
1054  * allocation within the reservation window.
1055  *
1056  * This will avoid keeping on searching the reservation list again and
1057  * again when somebody is looking for a free block (without
1058  * reservation), and there are lots of free blocks, but they are all
1059  * being reserved.
1060  *
1061  * We use a red-black tree for the per-filesystem reservation list.
1062  */
1063 static ext2_grpblk_t
1064 ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1065 			struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1066 			struct ext2_reserve_window_node * my_rsv,
1067 			unsigned long *count)
1068 {
1069 	ext2_fsblk_t group_first_block, group_last_block;
1070 	ext2_grpblk_t ret = 0;
1071 	unsigned long num = *count;
1072 
1073 	/*
1074 	 * we don't deal with reservation when
1075 	 * filesystem is mounted without reservation
1076 	 * or the file is not a regular file
1077 	 * or last attempt to allocate a block with reservation turned on failed
1078 	 */
1079 	if (my_rsv == NULL) {
1080 		return ext2_try_to_allocate(sb, group, bitmap_bh,
1081 						grp_goal, count, NULL);
1082 	}
1083 	/*
1084 	 * grp_goal is a group relative block number (if there is a goal)
1085 	 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1086 	 * first block is a filesystem wide block number
1087 	 * first block is the block number of the first block in this group
1088 	 */
1089 	group_first_block = ext2_group_first_block_no(sb, group);
1090 	group_last_block = ext2_group_last_block_no(sb, group);
1091 
1092 	/*
1093 	 * Basically we will allocate a new block from inode's reservation
1094 	 * window.
1095 	 *
1096 	 * We need to allocate a new reservation window, if:
1097 	 * a) inode does not have a reservation window; or
1098 	 * b) last attempt to allocate a block from existing reservation
1099 	 *    failed; or
1100 	 * c) we come here with a goal and with a reservation window
1101 	 *
1102 	 * We do not need to allocate a new reservation window if we come here
1103 	 * at the beginning with a goal and the goal is inside the window, or
1104 	 * we don't have a goal but already have a reservation window.
1105 	 * then we could go to allocate from the reservation window directly.
1106 	 */
1107 	while (1) {
1108 		if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1109 			!goal_in_my_reservation(&my_rsv->rsv_window,
1110 						grp_goal, group, sb)) {
1111 			if (my_rsv->rsv_goal_size < *count)
1112 				my_rsv->rsv_goal_size = *count;
1113 			ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1114 							group, bitmap_bh);
1115 			if (ret < 0)
1116 				break;			/* failed */
1117 
1118 			if (!goal_in_my_reservation(&my_rsv->rsv_window,
1119 							grp_goal, group, sb))
1120 				grp_goal = -1;
1121 		} else if (grp_goal >= 0) {
1122 			int curr = my_rsv->rsv_end -
1123 					(grp_goal + group_first_block) + 1;
1124 
1125 			if (curr < *count)
1126 				try_to_extend_reservation(my_rsv, sb,
1127 							*count - curr);
1128 		}
1129 
1130 		if ((my_rsv->rsv_start > group_last_block) ||
1131 				(my_rsv->rsv_end < group_first_block)) {
1132 			ext2_error(sb, __func__,
1133 				   "Reservation out of group %u range goal %d fsb[%lu,%lu] rsv[%lu, %lu]",
1134 				   group, grp_goal, group_first_block,
1135 				   group_last_block, my_rsv->rsv_start,
1136 				   my_rsv->rsv_end);
1137 			rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1138 			return -1;
1139 		}
1140 		ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1141 					   &num, &my_rsv->rsv_window);
1142 		if (ret >= 0) {
1143 			my_rsv->rsv_alloc_hit += num;
1144 			*count = num;
1145 			break;				/* succeed */
1146 		}
1147 		num = *count;
1148 	}
1149 	return ret;
1150 }
1151 
1152 /**
1153  * ext2_has_free_blocks()
1154  * @sbi:		in-core super block structure.
1155  *
1156  * Check if filesystem has at least 1 free block available for allocation.
1157  */
1158 static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1159 {
1160 	ext2_fsblk_t free_blocks, root_blocks;
1161 
1162 	free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1163 	root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1164 	if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1165 		!uid_eq(sbi->s_resuid, current_fsuid()) &&
1166 		(gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1167 		 !in_group_p (sbi->s_resgid))) {
1168 		return 0;
1169 	}
1170 	return 1;
1171 }
1172 
1173 /*
1174  * Returns 1 if the passed-in block region is valid; 0 if some part overlaps
1175  * with filesystem metadata blocks.
1176  */
1177 int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk,
1178 			  unsigned int count)
1179 {
1180 	if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
1181 	    (start_blk + count - 1 < start_blk) ||
1182 	    (start_blk + count - 1 >= le32_to_cpu(sbi->s_es->s_blocks_count)))
1183 		return 0;
1184 
1185 	/* Ensure we do not step over superblock */
1186 	if ((start_blk <= sbi->s_sb_block) &&
1187 	    (start_blk + count - 1 >= sbi->s_sb_block))
1188 		return 0;
1189 
1190 	return 1;
1191 }
1192 
1193 /*
1194  * ext2_new_blocks() -- core block(s) allocation function
1195  * @inode:		file inode
1196  * @goal:		given target block(filesystem wide)
1197  * @count:		target number of blocks to allocate
1198  * @errp:		error code
1199  * @flags:		allocate flags
1200  *
1201  * ext2_new_blocks uses a goal block to assist allocation.  If the goal is
1202  * free, or there is a free block within 32 blocks of the goal, that block
1203  * is allocated.  Otherwise a forward search is made for a free block; within
1204  * each block group the search first looks for an entire free byte in the block
1205  * bitmap, and then for any free bit if that fails.
1206  * This function also updates quota and i_blocks field.
1207  */
1208 ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1209 		    unsigned long *count, int *errp, unsigned int flags)
1210 {
1211 	struct buffer_head *bitmap_bh = NULL;
1212 	struct buffer_head *gdp_bh;
1213 	int group_no;
1214 	int goal_group;
1215 	ext2_grpblk_t grp_target_blk;	/* blockgroup relative goal block */
1216 	ext2_grpblk_t grp_alloc_blk;	/* blockgroup-relative allocated block*/
1217 	ext2_fsblk_t ret_block;		/* filesyetem-wide allocated block */
1218 	int bgi;			/* blockgroup iteration index */
1219 	int performed_allocation = 0;
1220 	ext2_grpblk_t free_blocks;	/* number of free blocks in a group */
1221 	struct super_block *sb;
1222 	struct ext2_group_desc *gdp;
1223 	struct ext2_super_block *es;
1224 	struct ext2_sb_info *sbi;
1225 	struct ext2_reserve_window_node *my_rsv = NULL;
1226 	struct ext2_block_alloc_info *block_i;
1227 	unsigned short windowsz = 0;
1228 	unsigned long ngroups;
1229 	unsigned long num = *count;
1230 	int ret;
1231 
1232 	*errp = -ENOSPC;
1233 	sb = inode->i_sb;
1234 
1235 	/*
1236 	 * Check quota for allocation of this block.
1237 	 */
1238 	ret = dquot_alloc_block(inode, num);
1239 	if (ret) {
1240 		*errp = ret;
1241 		return 0;
1242 	}
1243 
1244 	sbi = EXT2_SB(sb);
1245 	es = EXT2_SB(sb)->s_es;
1246 	ext2_debug("goal=%lu.\n", goal);
1247 	/*
1248 	 * Allocate a block from reservation only when the filesystem is
1249 	 * mounted with reservation(default,-o reservation), and it's a regular
1250 	 * file, and the desired window size is greater than 0 (One could use
1251 	 * ioctl command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn
1252 	 * off reservation on that particular file). Also do not use the
1253 	 * reservation window if the caller asked us not to do it.
1254 	 */
1255 	block_i = EXT2_I(inode)->i_block_alloc_info;
1256 	if (!(flags & EXT2_ALLOC_NORESERVE) && block_i) {
1257 		windowsz = block_i->rsv_window_node.rsv_goal_size;
1258 		if (windowsz > 0)
1259 			my_rsv = &block_i->rsv_window_node;
1260 	}
1261 
1262 	if (!ext2_has_free_blocks(sbi)) {
1263 		*errp = -ENOSPC;
1264 		goto out;
1265 	}
1266 
1267 	/*
1268 	 * First, test whether the goal block is free.
1269 	 */
1270 	if (goal < le32_to_cpu(es->s_first_data_block) ||
1271 	    goal >= le32_to_cpu(es->s_blocks_count))
1272 		goal = le32_to_cpu(es->s_first_data_block);
1273 	group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1274 			EXT2_BLOCKS_PER_GROUP(sb);
1275 	goal_group = group_no;
1276 retry_alloc:
1277 	gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1278 	if (!gdp)
1279 		goto io_error;
1280 
1281 	free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1282 	/*
1283 	 * if there is not enough free blocks to make a new resevation
1284 	 * turn off reservation for this allocation
1285 	 */
1286 	if (my_rsv && (free_blocks < windowsz)
1287 		&& (free_blocks > 0)
1288 		&& (rsv_is_empty(&my_rsv->rsv_window)))
1289 		my_rsv = NULL;
1290 
1291 	if (free_blocks > 0) {
1292 		grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1293 				EXT2_BLOCKS_PER_GROUP(sb));
1294 		/*
1295 		 * In case we retry allocation (due to fs reservation not
1296 		 * working out or fs corruption), the bitmap_bh is non-null
1297 		 * pointer and we have to release it before calling
1298 		 * read_block_bitmap().
1299 		 */
1300 		brelse(bitmap_bh);
1301 		bitmap_bh = read_block_bitmap(sb, group_no);
1302 		if (!bitmap_bh)
1303 			goto io_error;
1304 		grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1305 					bitmap_bh, grp_target_blk,
1306 					my_rsv, &num);
1307 		if (grp_alloc_blk >= 0)
1308 			goto allocated;
1309 	}
1310 
1311 	ngroups = EXT2_SB(sb)->s_groups_count;
1312 	smp_rmb();
1313 
1314 	/*
1315 	 * Now search the rest of the groups.  We assume that
1316 	 * group_no and gdp correctly point to the last group visited.
1317 	 */
1318 	for (bgi = 0; bgi < ngroups; bgi++) {
1319 		group_no++;
1320 		if (group_no >= ngroups)
1321 			group_no = 0;
1322 		gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1323 		if (!gdp)
1324 			goto io_error;
1325 
1326 		free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1327 		/*
1328 		 * skip this group (and avoid loading bitmap) if there
1329 		 * are no free blocks
1330 		 */
1331 		if (!free_blocks)
1332 			continue;
1333 		/*
1334 		 * skip this group if the number of
1335 		 * free blocks is less than half of the reservation
1336 		 * window size.
1337 		 */
1338 		if (my_rsv && (free_blocks <= (windowsz/2)))
1339 			continue;
1340 
1341 		brelse(bitmap_bh);
1342 		bitmap_bh = read_block_bitmap(sb, group_no);
1343 		if (!bitmap_bh)
1344 			goto io_error;
1345 		/*
1346 		 * try to allocate block(s) from this group, without a goal(-1).
1347 		 */
1348 		grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1349 					bitmap_bh, -1, my_rsv, &num);
1350 		if (grp_alloc_blk >= 0)
1351 			goto allocated;
1352 	}
1353 	/*
1354 	 * We may end up a bogus earlier ENOSPC error due to
1355 	 * filesystem is "full" of reservations, but
1356 	 * there maybe indeed free blocks available on disk
1357 	 * In this case, we just forget about the reservations
1358 	 * just do block allocation as without reservations.
1359 	 */
1360 	if (my_rsv) {
1361 		my_rsv = NULL;
1362 		windowsz = 0;
1363 		group_no = goal_group;
1364 		goto retry_alloc;
1365 	}
1366 	/* No space left on the device */
1367 	*errp = -ENOSPC;
1368 	goto out;
1369 
1370 allocated:
1371 
1372 	ext2_debug("using block group %d(%d)\n",
1373 			group_no, gdp->bg_free_blocks_count);
1374 
1375 	ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1376 
1377 	if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1378 	    in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1379 	    in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1380 		      EXT2_SB(sb)->s_itb_per_group) ||
1381 	    in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1382 		      EXT2_SB(sb)->s_itb_per_group)) {
1383 		ext2_error(sb, "ext2_new_blocks",
1384 			    "Allocating block in system zone - "
1385 			    "blocks from "E2FSBLK", length %lu",
1386 			    ret_block, num);
1387 		/*
1388 		 * ext2_try_to_allocate marked the blocks we allocated as in
1389 		 * use.  So we may want to selectively mark some of the blocks
1390 		 * as free
1391 		 */
1392 		num = *count;
1393 		goto retry_alloc;
1394 	}
1395 
1396 	performed_allocation = 1;
1397 
1398 	if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1399 		ext2_error(sb, "ext2_new_blocks",
1400 			    "block("E2FSBLK") >= blocks count(%d) - "
1401 			    "block_group = %d, es == %p ", ret_block,
1402 			le32_to_cpu(es->s_blocks_count), group_no, es);
1403 		goto out;
1404 	}
1405 
1406 	group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1407 	percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1408 
1409 	mark_buffer_dirty(bitmap_bh);
1410 	if (sb->s_flags & SB_SYNCHRONOUS)
1411 		sync_dirty_buffer(bitmap_bh);
1412 
1413 	*errp = 0;
1414 	brelse(bitmap_bh);
1415 	if (num < *count) {
1416 		dquot_free_block_nodirty(inode, *count-num);
1417 		mark_inode_dirty(inode);
1418 		*count = num;
1419 	}
1420 	return ret_block;
1421 
1422 io_error:
1423 	*errp = -EIO;
1424 out:
1425 	/*
1426 	 * Undo the block allocation
1427 	 */
1428 	if (!performed_allocation) {
1429 		dquot_free_block_nodirty(inode, *count);
1430 		mark_inode_dirty(inode);
1431 	}
1432 	brelse(bitmap_bh);
1433 	return 0;
1434 }
1435 
1436 #ifdef EXT2FS_DEBUG
1437 
1438 unsigned long ext2_count_free(struct buffer_head *map, unsigned int numchars)
1439 {
1440 	return numchars * BITS_PER_BYTE - memweight(map->b_data, numchars);
1441 }
1442 
1443 #endif  /*  EXT2FS_DEBUG  */
1444 
1445 unsigned long ext2_count_free_blocks (struct super_block * sb)
1446 {
1447 	struct ext2_group_desc * desc;
1448 	unsigned long desc_count = 0;
1449 	int i;
1450 #ifdef EXT2FS_DEBUG
1451 	unsigned long bitmap_count, x;
1452 	struct ext2_super_block *es;
1453 
1454 	es = EXT2_SB(sb)->s_es;
1455 	desc_count = 0;
1456 	bitmap_count = 0;
1457 	desc = NULL;
1458 	for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1459 		struct buffer_head *bitmap_bh;
1460 		desc = ext2_get_group_desc (sb, i, NULL);
1461 		if (!desc)
1462 			continue;
1463 		desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1464 		bitmap_bh = read_block_bitmap(sb, i);
1465 		if (!bitmap_bh)
1466 			continue;
1467 
1468 		x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1469 		printk ("group %d: stored = %d, counted = %lu\n",
1470 			i, le16_to_cpu(desc->bg_free_blocks_count), x);
1471 		bitmap_count += x;
1472 		brelse(bitmap_bh);
1473 	}
1474 	printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1475 		(long)le32_to_cpu(es->s_free_blocks_count),
1476 		desc_count, bitmap_count);
1477 	return bitmap_count;
1478 #else
1479 	for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1480 		desc = ext2_get_group_desc(sb, i, NULL);
1481 		if (!desc)
1482 			continue;
1483 		desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1484 	}
1485 	return desc_count;
1486 #endif
1487 }
1488 
1489 static inline int test_root(int a, int b)
1490 {
1491 	int num = b;
1492 
1493 	while (a > num)
1494 		num *= b;
1495 	return num == a;
1496 }
1497 
1498 static int ext2_group_sparse(int group)
1499 {
1500 	if (group <= 1)
1501 		return 1;
1502 	return (test_root(group, 3) || test_root(group, 5) ||
1503 		test_root(group, 7));
1504 }
1505 
1506 /**
1507  *	ext2_bg_has_super - number of blocks used by the superblock in group
1508  *	@sb: superblock for filesystem
1509  *	@group: group number to check
1510  *
1511  *	Return the number of blocks used by the superblock (primary or backup)
1512  *	in this group.  Currently this will be only 0 or 1.
1513  */
1514 int ext2_bg_has_super(struct super_block *sb, int group)
1515 {
1516 	if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1517 	    !ext2_group_sparse(group))
1518 		return 0;
1519 	return 1;
1520 }
1521 
1522 /**
1523  *	ext2_bg_num_gdb - number of blocks used by the group table in group
1524  *	@sb: superblock for filesystem
1525  *	@group: group number to check
1526  *
1527  *	Return the number of blocks used by the group descriptor table
1528  *	(primary or backup) in this group.  In the future there may be a
1529  *	different number of descriptor blocks in each group.
1530  */
1531 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1532 {
1533 	return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
1534 }
1535 
1536