xref: /linux/fs/ext2/balloc.c (revision 293d5b43948309434568f4dcbb36cce4c3c51bd5)
1 /*
2  *  linux/fs/ext2/balloc.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
10  *  Big-endian to little-endian byte-swapping/bitmaps by
11  *        David S. Miller (davem@caip.rutgers.edu), 1995
12  */
13 
14 #include "ext2.h"
15 #include <linux/quotaops.h>
16 #include <linux/slab.h>
17 #include <linux/sched.h>
18 #include <linux/buffer_head.h>
19 #include <linux/capability.h>
20 
21 /*
22  * balloc.c contains the blocks allocation and deallocation routines
23  */
24 
25 /*
26  * The free blocks are managed by bitmaps.  A file system contains several
27  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
28  * block for inodes, N blocks for the inode table and data blocks.
29  *
30  * The file system contains group descriptors which are located after the
31  * super block.  Each descriptor contains the number of the bitmap block and
32  * the free blocks count in the block.  The descriptors are loaded in memory
33  * when a file system is mounted (see ext2_fill_super).
34  */
35 
36 
37 #define in_range(b, first, len)	((b) >= (first) && (b) <= (first) + (len) - 1)
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 		ext2_error (sb, "ext2_get_group_desc",
50 			    "block_group >= groups_count - "
51 			    "block_group = %d, groups_count = %lu",
52 			    block_group, sbi->s_groups_count);
53 
54 		return NULL;
55 	}
56 
57 	group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
58 	offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
59 	if (!sbi->s_group_desc[group_desc]) {
60 		ext2_error (sb, "ext2_get_group_desc",
61 			    "Group descriptor not loaded - "
62 			    "block_group = %d, group_desc = %lu, desc = %lu",
63 			     block_group, group_desc, offset);
64 		return NULL;
65 	}
66 
67 	desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
68 	if (bh)
69 		*bh = sbi->s_group_desc[group_desc];
70 	return desc + offset;
71 }
72 
73 static int ext2_valid_block_bitmap(struct super_block *sb,
74 					struct ext2_group_desc *desc,
75 					unsigned int block_group,
76 					struct buffer_head *bh)
77 {
78 	ext2_grpblk_t offset;
79 	ext2_grpblk_t next_zero_bit;
80 	ext2_fsblk_t bitmap_blk;
81 	ext2_fsblk_t group_first_block;
82 
83 	group_first_block = ext2_group_first_block_no(sb, block_group);
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 (!ext2_test_bit(offset, bh->b_data))
89 		/* bad block bitmap */
90 		goto err_out;
91 
92 	/* check whether the inode bitmap block number is set */
93 	bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
94 	offset = bitmap_blk - group_first_block;
95 	if (!ext2_test_bit(offset, bh->b_data))
96 		/* bad block bitmap */
97 		goto err_out;
98 
99 	/* check whether the inode table block number is set */
100 	bitmap_blk = le32_to_cpu(desc->bg_inode_table);
101 	offset = bitmap_blk - group_first_block;
102 	next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
103 				offset + EXT2_SB(sb)->s_itb_per_group,
104 				offset);
105 	if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
106 		/* good bitmap for inode tables */
107 		return 1;
108 
109 err_out:
110 	ext2_error(sb, __func__,
111 			"Invalid block bitmap - "
112 			"block_group = %d, block = %lu",
113 			block_group, bitmap_blk);
114 	return 0;
115 }
116 
117 /*
118  * Read the bitmap for a given block_group,and validate the
119  * bits for block/inode/inode tables are set in the bitmaps
120  *
121  * Return buffer_head on success or NULL in case of failure.
122  */
123 static struct buffer_head *
124 read_block_bitmap(struct super_block *sb, unsigned int block_group)
125 {
126 	struct ext2_group_desc * desc;
127 	struct buffer_head * bh = NULL;
128 	ext2_fsblk_t bitmap_blk;
129 
130 	desc = ext2_get_group_desc(sb, block_group, NULL);
131 	if (!desc)
132 		return NULL;
133 	bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
134 	bh = sb_getblk(sb, bitmap_blk);
135 	if (unlikely(!bh)) {
136 		ext2_error(sb, __func__,
137 			    "Cannot read block bitmap - "
138 			    "block_group = %d, block_bitmap = %u",
139 			    block_group, le32_to_cpu(desc->bg_block_bitmap));
140 		return NULL;
141 	}
142 	if (likely(bh_uptodate_or_lock(bh)))
143 		return bh;
144 
145 	if (bh_submit_read(bh) < 0) {
146 		brelse(bh);
147 		ext2_error(sb, __func__,
148 			    "Cannot read block bitmap - "
149 			    "block_group = %d, block_bitmap = %u",
150 			    block_group, le32_to_cpu(desc->bg_block_bitmap));
151 		return NULL;
152 	}
153 
154 	ext2_valid_block_bitmap(sb, desc, block_group, bh);
155 	/*
156 	 * file system mounted not to panic on error, continue with corrupt
157 	 * bitmap
158 	 */
159 	return bh;
160 }
161 
162 static void group_adjust_blocks(struct super_block *sb, int group_no,
163 	struct ext2_group_desc *desc, struct buffer_head *bh, int count)
164 {
165 	if (count) {
166 		struct ext2_sb_info *sbi = EXT2_SB(sb);
167 		unsigned free_blocks;
168 
169 		spin_lock(sb_bgl_lock(sbi, group_no));
170 		free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
171 		desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
172 		spin_unlock(sb_bgl_lock(sbi, group_no));
173 		mark_buffer_dirty(bh);
174 	}
175 }
176 
177 /*
178  * The reservation window structure operations
179  * --------------------------------------------
180  * Operations include:
181  * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
182  *
183  * We use a red-black tree to represent per-filesystem reservation
184  * windows.
185  *
186  */
187 
188 /**
189  * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
190  * @rb_root:		root of per-filesystem reservation rb tree
191  * @verbose:		verbose mode
192  * @fn:			function which wishes to dump the reservation map
193  *
194  * If verbose is turned on, it will print the whole block reservation
195  * windows(start, end). Otherwise, it will only print out the "bad" windows,
196  * those windows that overlap with their immediate neighbors.
197  */
198 #if 1
199 static void __rsv_window_dump(struct rb_root *root, int verbose,
200 			      const char *fn)
201 {
202 	struct rb_node *n;
203 	struct ext2_reserve_window_node *rsv, *prev;
204 	int bad;
205 
206 restart:
207 	n = rb_first(root);
208 	bad = 0;
209 	prev = NULL;
210 
211 	printk("Block Allocation Reservation Windows Map (%s):\n", fn);
212 	while (n) {
213 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
214 		if (verbose)
215 			printk("reservation window 0x%p "
216 				"start: %lu, end: %lu\n",
217 				rsv, rsv->rsv_start, rsv->rsv_end);
218 		if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
219 			printk("Bad reservation %p (start >= end)\n",
220 			       rsv);
221 			bad = 1;
222 		}
223 		if (prev && prev->rsv_end >= rsv->rsv_start) {
224 			printk("Bad reservation %p (prev->end >= start)\n",
225 			       rsv);
226 			bad = 1;
227 		}
228 		if (bad) {
229 			if (!verbose) {
230 				printk("Restarting reservation walk in verbose mode\n");
231 				verbose = 1;
232 				goto restart;
233 			}
234 		}
235 		n = rb_next(n);
236 		prev = rsv;
237 	}
238 	printk("Window map complete.\n");
239 	BUG_ON(bad);
240 }
241 #define rsv_window_dump(root, verbose) \
242 	__rsv_window_dump((root), (verbose), __func__)
243 #else
244 #define rsv_window_dump(root, verbose) do {} while (0)
245 #endif
246 
247 /**
248  * goal_in_my_reservation()
249  * @rsv:		inode's reservation window
250  * @grp_goal:		given goal block relative to the allocation block group
251  * @group:		the current allocation block group
252  * @sb:			filesystem super block
253  *
254  * Test if the given goal block (group relative) is within the file's
255  * own block reservation window range.
256  *
257  * If the reservation window is outside the goal allocation group, return 0;
258  * grp_goal (given goal block) could be -1, which means no specific
259  * goal block. In this case, always return 1.
260  * If the goal block is within the reservation window, return 1;
261  * otherwise, return 0;
262  */
263 static int
264 goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
265 			unsigned int group, struct super_block * sb)
266 {
267 	ext2_fsblk_t group_first_block, group_last_block;
268 
269 	group_first_block = ext2_group_first_block_no(sb, group);
270 	group_last_block = group_first_block + EXT2_BLOCKS_PER_GROUP(sb) - 1;
271 
272 	if ((rsv->_rsv_start > group_last_block) ||
273 	    (rsv->_rsv_end < group_first_block))
274 		return 0;
275 	if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
276 		|| (grp_goal + group_first_block > rsv->_rsv_end)))
277 		return 0;
278 	return 1;
279 }
280 
281 /**
282  * search_reserve_window()
283  * @rb_root:		root of reservation tree
284  * @goal:		target allocation block
285  *
286  * Find the reserved window which includes the goal, or the previous one
287  * if the goal is not in any window.
288  * Returns NULL if there are no windows or if all windows start after the goal.
289  */
290 static struct ext2_reserve_window_node *
291 search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
292 {
293 	struct rb_node *n = root->rb_node;
294 	struct ext2_reserve_window_node *rsv;
295 
296 	if (!n)
297 		return NULL;
298 
299 	do {
300 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
301 
302 		if (goal < rsv->rsv_start)
303 			n = n->rb_left;
304 		else if (goal > rsv->rsv_end)
305 			n = n->rb_right;
306 		else
307 			return rsv;
308 	} while (n);
309 	/*
310 	 * We've fallen off the end of the tree: the goal wasn't inside
311 	 * any particular node.  OK, the previous node must be to one
312 	 * side of the interval containing the goal.  If it's the RHS,
313 	 * we need to back up one.
314 	 */
315 	if (rsv->rsv_start > goal) {
316 		n = rb_prev(&rsv->rsv_node);
317 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
318 	}
319 	return rsv;
320 }
321 
322 /*
323  * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
324  * @sb:			super block
325  * @rsv:		reservation window to add
326  *
327  * Must be called with rsv_lock held.
328  */
329 void ext2_rsv_window_add(struct super_block *sb,
330 		    struct ext2_reserve_window_node *rsv)
331 {
332 	struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
333 	struct rb_node *node = &rsv->rsv_node;
334 	ext2_fsblk_t start = rsv->rsv_start;
335 
336 	struct rb_node ** p = &root->rb_node;
337 	struct rb_node * parent = NULL;
338 	struct ext2_reserve_window_node *this;
339 
340 	while (*p)
341 	{
342 		parent = *p;
343 		this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
344 
345 		if (start < this->rsv_start)
346 			p = &(*p)->rb_left;
347 		else if (start > this->rsv_end)
348 			p = &(*p)->rb_right;
349 		else {
350 			rsv_window_dump(root, 1);
351 			BUG();
352 		}
353 	}
354 
355 	rb_link_node(node, parent, p);
356 	rb_insert_color(node, root);
357 }
358 
359 /**
360  * rsv_window_remove() -- unlink a window from the reservation rb tree
361  * @sb:			super block
362  * @rsv:		reservation window to remove
363  *
364  * Mark the block reservation window as not allocated, and unlink it
365  * from the filesystem reservation window rb tree. Must be called with
366  * rsv_lock held.
367  */
368 static void rsv_window_remove(struct super_block *sb,
369 			      struct ext2_reserve_window_node *rsv)
370 {
371 	rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
372 	rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
373 	rsv->rsv_alloc_hit = 0;
374 	rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
375 }
376 
377 /*
378  * rsv_is_empty() -- Check if the reservation window is allocated.
379  * @rsv:		given reservation window to check
380  *
381  * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
382  */
383 static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
384 {
385 	/* a valid reservation end block could not be 0 */
386 	return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
387 }
388 
389 /**
390  * ext2_init_block_alloc_info()
391  * @inode:		file inode structure
392  *
393  * Allocate and initialize the  reservation window structure, and
394  * link the window to the ext2 inode structure at last
395  *
396  * The reservation window structure is only dynamically allocated
397  * and linked to ext2 inode the first time the open file
398  * needs a new block. So, before every ext2_new_block(s) call, for
399  * regular files, we should check whether the reservation window
400  * structure exists or not. In the latter case, this function is called.
401  * Fail to do so will result in block reservation being turned off for that
402  * open file.
403  *
404  * This function is called from ext2_get_blocks_handle(), also called
405  * when setting the reservation window size through ioctl before the file
406  * is open for write (needs block allocation).
407  *
408  * Needs truncate_mutex protection prior to calling this function.
409  */
410 void ext2_init_block_alloc_info(struct inode *inode)
411 {
412 	struct ext2_inode_info *ei = EXT2_I(inode);
413 	struct ext2_block_alloc_info *block_i;
414 	struct super_block *sb = inode->i_sb;
415 
416 	block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
417 	if (block_i) {
418 		struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
419 
420 		rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
421 		rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
422 
423 	 	/*
424 		 * if filesystem is mounted with NORESERVATION, the goal
425 		 * reservation window size is set to zero to indicate
426 		 * block reservation is off
427 		 */
428 		if (!test_opt(sb, RESERVATION))
429 			rsv->rsv_goal_size = 0;
430 		else
431 			rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
432 		rsv->rsv_alloc_hit = 0;
433 		block_i->last_alloc_logical_block = 0;
434 		block_i->last_alloc_physical_block = 0;
435 	}
436 	ei->i_block_alloc_info = block_i;
437 }
438 
439 /**
440  * ext2_discard_reservation()
441  * @inode:		inode
442  *
443  * Discard(free) block reservation window on last file close, or truncate
444  * or at last iput().
445  *
446  * It is being called in three cases:
447  * 	ext2_release_file(): last writer closes the file
448  * 	ext2_clear_inode(): last iput(), when nobody links to this file.
449  * 	ext2_truncate(): when the block indirect map is about to change.
450  */
451 void ext2_discard_reservation(struct inode *inode)
452 {
453 	struct ext2_inode_info *ei = EXT2_I(inode);
454 	struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
455 	struct ext2_reserve_window_node *rsv;
456 	spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
457 
458 	if (!block_i)
459 		return;
460 
461 	rsv = &block_i->rsv_window_node;
462 	if (!rsv_is_empty(&rsv->rsv_window)) {
463 		spin_lock(rsv_lock);
464 		if (!rsv_is_empty(&rsv->rsv_window))
465 			rsv_window_remove(inode->i_sb, rsv);
466 		spin_unlock(rsv_lock);
467 	}
468 }
469 
470 /**
471  * ext2_free_blocks() -- Free given blocks and update quota and i_blocks
472  * @inode:		inode
473  * @block:		start physical block to free
474  * @count:		number of blocks to free
475  */
476 void ext2_free_blocks (struct inode * inode, unsigned long block,
477 		       unsigned long count)
478 {
479 	struct buffer_head *bitmap_bh = NULL;
480 	struct buffer_head * bh2;
481 	unsigned long block_group;
482 	unsigned long bit;
483 	unsigned long i;
484 	unsigned long overflow;
485 	struct super_block * sb = inode->i_sb;
486 	struct ext2_sb_info * sbi = EXT2_SB(sb);
487 	struct ext2_group_desc * desc;
488 	struct ext2_super_block * es = sbi->s_es;
489 	unsigned freed = 0, group_freed;
490 
491 	if (block < le32_to_cpu(es->s_first_data_block) ||
492 	    block + count < block ||
493 	    block + count > le32_to_cpu(es->s_blocks_count)) {
494 		ext2_error (sb, "ext2_free_blocks",
495 			    "Freeing blocks not in datazone - "
496 			    "block = %lu, count = %lu", block, count);
497 		goto error_return;
498 	}
499 
500 	ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
501 
502 do_more:
503 	overflow = 0;
504 	block_group = (block - le32_to_cpu(es->s_first_data_block)) /
505 		      EXT2_BLOCKS_PER_GROUP(sb);
506 	bit = (block - le32_to_cpu(es->s_first_data_block)) %
507 		      EXT2_BLOCKS_PER_GROUP(sb);
508 	/*
509 	 * Check to see if we are freeing blocks across a group
510 	 * boundary.
511 	 */
512 	if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
513 		overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
514 		count -= overflow;
515 	}
516 	brelse(bitmap_bh);
517 	bitmap_bh = read_block_bitmap(sb, block_group);
518 	if (!bitmap_bh)
519 		goto error_return;
520 
521 	desc = ext2_get_group_desc (sb, block_group, &bh2);
522 	if (!desc)
523 		goto error_return;
524 
525 	if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
526 	    in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
527 	    in_range (block, le32_to_cpu(desc->bg_inode_table),
528 		      sbi->s_itb_per_group) ||
529 	    in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
530 		      sbi->s_itb_per_group)) {
531 		ext2_error (sb, "ext2_free_blocks",
532 			    "Freeing blocks in system zones - "
533 			    "Block = %lu, count = %lu",
534 			    block, count);
535 		goto error_return;
536 	}
537 
538 	for (i = 0, group_freed = 0; i < count; i++) {
539 		if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
540 						bit + i, bitmap_bh->b_data)) {
541 			ext2_error(sb, __func__,
542 				"bit already cleared for block %lu", block + i);
543 		} else {
544 			group_freed++;
545 		}
546 	}
547 
548 	mark_buffer_dirty(bitmap_bh);
549 	if (sb->s_flags & MS_SYNCHRONOUS)
550 		sync_dirty_buffer(bitmap_bh);
551 
552 	group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
553 	freed += group_freed;
554 
555 	if (overflow) {
556 		block += count;
557 		count = overflow;
558 		goto do_more;
559 	}
560 error_return:
561 	brelse(bitmap_bh);
562 	if (freed) {
563 		percpu_counter_add(&sbi->s_freeblocks_counter, freed);
564 		dquot_free_block_nodirty(inode, freed);
565 		mark_inode_dirty(inode);
566 	}
567 }
568 
569 /**
570  * bitmap_search_next_usable_block()
571  * @start:		the starting block (group relative) of the search
572  * @bh:			bufferhead contains the block group bitmap
573  * @maxblocks:		the ending block (group relative) of the reservation
574  *
575  * The bitmap search --- search forward through the actual bitmap on disk until
576  * we find a bit free.
577  */
578 static ext2_grpblk_t
579 bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
580 					ext2_grpblk_t maxblocks)
581 {
582 	ext2_grpblk_t next;
583 
584 	next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
585 	if (next >= maxblocks)
586 		return -1;
587 	return next;
588 }
589 
590 /**
591  * find_next_usable_block()
592  * @start:		the starting block (group relative) to find next
593  * 			allocatable block in bitmap.
594  * @bh:			bufferhead contains the block group bitmap
595  * @maxblocks:		the ending block (group relative) for the search
596  *
597  * Find an allocatable block in a bitmap.  We perform the "most
598  * appropriate allocation" algorithm of looking for a free block near
599  * the initial goal; then for a free byte somewhere in the bitmap;
600  * then for any free bit in the bitmap.
601  */
602 static ext2_grpblk_t
603 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
604 {
605 	ext2_grpblk_t here, next;
606 	char *p, *r;
607 
608 	if (start > 0) {
609 		/*
610 		 * The goal was occupied; search forward for a free
611 		 * block within the next XX blocks.
612 		 *
613 		 * end_goal is more or less random, but it has to be
614 		 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
615 		 * next 64-bit boundary is simple..
616 		 */
617 		ext2_grpblk_t end_goal = (start + 63) & ~63;
618 		if (end_goal > maxblocks)
619 			end_goal = maxblocks;
620 		here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
621 		if (here < end_goal)
622 			return here;
623 		ext2_debug("Bit not found near goal\n");
624 	}
625 
626 	here = start;
627 	if (here < 0)
628 		here = 0;
629 
630 	p = ((char *)bh->b_data) + (here >> 3);
631 	r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
632 	next = (r - ((char *)bh->b_data)) << 3;
633 
634 	if (next < maxblocks && next >= here)
635 		return next;
636 
637 	here = bitmap_search_next_usable_block(here, bh, maxblocks);
638 	return here;
639 }
640 
641 /**
642  * ext2_try_to_allocate()
643  * @sb:			superblock
644  * @group:		given allocation block group
645  * @bitmap_bh:		bufferhead holds the block bitmap
646  * @grp_goal:		given target block within the group
647  * @count:		target number of blocks to allocate
648  * @my_rsv:		reservation window
649  *
650  * Attempt to allocate blocks within a give range. Set the range of allocation
651  * first, then find the first free bit(s) from the bitmap (within the range),
652  * and at last, allocate the blocks by claiming the found free bit as allocated.
653  *
654  * To set the range of this allocation:
655  * 	if there is a reservation window, only try to allocate block(s)
656  * 	from the file's own reservation window;
657  * 	Otherwise, the allocation range starts from the give goal block,
658  * 	ends at the block group's last block.
659  *
660  * If we failed to allocate the desired block then we may end up crossing to a
661  * new bitmap.
662  */
663 static int
664 ext2_try_to_allocate(struct super_block *sb, int group,
665 			struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
666 			unsigned long *count,
667 			struct ext2_reserve_window *my_rsv)
668 {
669 	ext2_fsblk_t group_first_block;
670        	ext2_grpblk_t start, end;
671 	unsigned long num = 0;
672 
673 	/* we do allocation within the reservation window if we have a window */
674 	if (my_rsv) {
675 		group_first_block = ext2_group_first_block_no(sb, group);
676 		if (my_rsv->_rsv_start >= group_first_block)
677 			start = my_rsv->_rsv_start - group_first_block;
678 		else
679 			/* reservation window cross group boundary */
680 			start = 0;
681 		end = my_rsv->_rsv_end - group_first_block + 1;
682 		if (end > EXT2_BLOCKS_PER_GROUP(sb))
683 			/* reservation window crosses group boundary */
684 			end = EXT2_BLOCKS_PER_GROUP(sb);
685 		if ((start <= grp_goal) && (grp_goal < end))
686 			start = grp_goal;
687 		else
688 			grp_goal = -1;
689 	} else {
690 		if (grp_goal > 0)
691 			start = grp_goal;
692 		else
693 			start = 0;
694 		end = EXT2_BLOCKS_PER_GROUP(sb);
695 	}
696 
697 	BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
698 
699 repeat:
700 	if (grp_goal < 0) {
701 		grp_goal = find_next_usable_block(start, bitmap_bh, end);
702 		if (grp_goal < 0)
703 			goto fail_access;
704 		if (!my_rsv) {
705 			int i;
706 
707 			for (i = 0; i < 7 && grp_goal > start &&
708 					!ext2_test_bit(grp_goal - 1,
709 					     		bitmap_bh->b_data);
710 			     		i++, grp_goal--)
711 				;
712 		}
713 	}
714 	start = grp_goal;
715 
716 	if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), grp_goal,
717 			       				bitmap_bh->b_data)) {
718 		/*
719 		 * The block was allocated by another thread, or it was
720 		 * allocated and then freed by another thread
721 		 */
722 		start++;
723 		grp_goal++;
724 		if (start >= end)
725 			goto fail_access;
726 		goto repeat;
727 	}
728 	num++;
729 	grp_goal++;
730 	while (num < *count && grp_goal < end
731 		&& !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
732 					grp_goal, bitmap_bh->b_data)) {
733 		num++;
734 		grp_goal++;
735 	}
736 	*count = num;
737 	return grp_goal - num;
738 fail_access:
739 	*count = num;
740 	return -1;
741 }
742 
743 /**
744  * 	find_next_reservable_window():
745  *		find a reservable space within the given range.
746  *		It does not allocate the reservation window for now:
747  *		alloc_new_reservation() will do the work later.
748  *
749  * 	@search_head: the head of the searching list;
750  *		This is not necessarily the list head of the whole filesystem
751  *
752  *		We have both head and start_block to assist the search
753  *		for the reservable space. The list starts from head,
754  *		but we will shift to the place where start_block is,
755  *		then start from there, when looking for a reservable space.
756  *
757  * 	@size: the target new reservation window size
758  *
759  * 	@group_first_block: the first block we consider to start
760  *			the real search from
761  *
762  * 	@last_block:
763  *		the maximum block number that our goal reservable space
764  *		could start from. This is normally the last block in this
765  *		group. The search will end when we found the start of next
766  *		possible reservable space is out of this boundary.
767  *		This could handle the cross boundary reservation window
768  *		request.
769  *
770  * 	basically we search from the given range, rather than the whole
771  * 	reservation double linked list, (start_block, last_block)
772  * 	to find a free region that is of my size and has not
773  * 	been reserved.
774  *
775  */
776 static int find_next_reservable_window(
777 				struct ext2_reserve_window_node *search_head,
778 				struct ext2_reserve_window_node *my_rsv,
779 				struct super_block * sb,
780 				ext2_fsblk_t start_block,
781 				ext2_fsblk_t last_block)
782 {
783 	struct rb_node *next;
784 	struct ext2_reserve_window_node *rsv, *prev;
785 	ext2_fsblk_t cur;
786 	int size = my_rsv->rsv_goal_size;
787 
788 	/* TODO: make the start of the reservation window byte-aligned */
789 	/* cur = *start_block & ~7;*/
790 	cur = start_block;
791 	rsv = search_head;
792 	if (!rsv)
793 		return -1;
794 
795 	while (1) {
796 		if (cur <= rsv->rsv_end)
797 			cur = rsv->rsv_end + 1;
798 
799 		/* TODO?
800 		 * in the case we could not find a reservable space
801 		 * that is what is expected, during the re-search, we could
802 		 * remember what's the largest reservable space we could have
803 		 * and return that one.
804 		 *
805 		 * For now it will fail if we could not find the reservable
806 		 * space with expected-size (or more)...
807 		 */
808 		if (cur > last_block)
809 			return -1;		/* fail */
810 
811 		prev = rsv;
812 		next = rb_next(&rsv->rsv_node);
813 		rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
814 
815 		/*
816 		 * Reached the last reservation, we can just append to the
817 		 * previous one.
818 		 */
819 		if (!next)
820 			break;
821 
822 		if (cur + size <= rsv->rsv_start) {
823 			/*
824 			 * Found a reserveable space big enough.  We could
825 			 * have a reservation across the group boundary here
826 		 	 */
827 			break;
828 		}
829 	}
830 	/*
831 	 * we come here either :
832 	 * when we reach the end of the whole list,
833 	 * and there is empty reservable space after last entry in the list.
834 	 * append it to the end of the list.
835 	 *
836 	 * or we found one reservable space in the middle of the list,
837 	 * return the reservation window that we could append to.
838 	 * succeed.
839 	 */
840 
841 	if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
842 		rsv_window_remove(sb, my_rsv);
843 
844 	/*
845 	 * Let's book the whole available window for now.  We will check the
846 	 * disk bitmap later and then, if there are free blocks then we adjust
847 	 * the window size if it's larger than requested.
848 	 * Otherwise, we will remove this node from the tree next time
849 	 * call find_next_reservable_window.
850 	 */
851 	my_rsv->rsv_start = cur;
852 	my_rsv->rsv_end = cur + size - 1;
853 	my_rsv->rsv_alloc_hit = 0;
854 
855 	if (prev != my_rsv)
856 		ext2_rsv_window_add(sb, my_rsv);
857 
858 	return 0;
859 }
860 
861 /**
862  * 	alloc_new_reservation()--allocate a new reservation window
863  *
864  *		To make a new reservation, we search part of the filesystem
865  *		reservation list (the list that inside the group). We try to
866  *		allocate a new reservation window near the allocation goal,
867  *		or the beginning of the group, if there is no goal.
868  *
869  *		We first find a reservable space after the goal, then from
870  *		there, we check the bitmap for the first free block after
871  *		it. If there is no free block until the end of group, then the
872  *		whole group is full, we failed. Otherwise, check if the free
873  *		block is inside the expected reservable space, if so, we
874  *		succeed.
875  *		If the first free block is outside the reservable space, then
876  *		start from the first free block, we search for next available
877  *		space, and go on.
878  *
879  *	on succeed, a new reservation will be found and inserted into the list
880  *	It contains at least one free block, and it does not overlap with other
881  *	reservation windows.
882  *
883  *	failed: we failed to find a reservation window in this group
884  *
885  *	@rsv: the reservation
886  *
887  *	@grp_goal: The goal (group-relative).  It is where the search for a
888  *		free reservable space should start from.
889  *		if we have a goal(goal >0 ), then start from there,
890  *		no goal(goal = -1), we start from the first block
891  *		of the group.
892  *
893  *	@sb: the super block
894  *	@group: the group we are trying to allocate in
895  *	@bitmap_bh: the block group block bitmap
896  *
897  */
898 static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
899 		ext2_grpblk_t grp_goal, struct super_block *sb,
900 		unsigned int group, struct buffer_head *bitmap_bh)
901 {
902 	struct ext2_reserve_window_node *search_head;
903 	ext2_fsblk_t group_first_block, group_end_block, start_block;
904 	ext2_grpblk_t first_free_block;
905 	struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
906 	unsigned long size;
907 	int ret;
908 	spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
909 
910 	group_first_block = ext2_group_first_block_no(sb, group);
911 	group_end_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
912 
913 	if (grp_goal < 0)
914 		start_block = group_first_block;
915 	else
916 		start_block = grp_goal + group_first_block;
917 
918 	size = my_rsv->rsv_goal_size;
919 
920 	if (!rsv_is_empty(&my_rsv->rsv_window)) {
921 		/*
922 		 * if the old reservation is cross group boundary
923 		 * and if the goal is inside the old reservation window,
924 		 * we will come here when we just failed to allocate from
925 		 * the first part of the window. We still have another part
926 		 * that belongs to the next group. In this case, there is no
927 		 * point to discard our window and try to allocate a new one
928 		 * in this group(which will fail). we should
929 		 * keep the reservation window, just simply move on.
930 		 *
931 		 * Maybe we could shift the start block of the reservation
932 		 * window to the first block of next group.
933 		 */
934 
935 		if ((my_rsv->rsv_start <= group_end_block) &&
936 				(my_rsv->rsv_end > group_end_block) &&
937 				(start_block >= my_rsv->rsv_start))
938 			return -1;
939 
940 		if ((my_rsv->rsv_alloc_hit >
941 		     (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
942 			/*
943 			 * if the previously allocation hit ratio is
944 			 * greater than 1/2, then we double the size of
945 			 * the reservation window the next time,
946 			 * otherwise we keep the same size window
947 			 */
948 			size = size * 2;
949 			if (size > EXT2_MAX_RESERVE_BLOCKS)
950 				size = EXT2_MAX_RESERVE_BLOCKS;
951 			my_rsv->rsv_goal_size= size;
952 		}
953 	}
954 
955 	spin_lock(rsv_lock);
956 	/*
957 	 * shift the search start to the window near the goal block
958 	 */
959 	search_head = search_reserve_window(fs_rsv_root, start_block);
960 
961 	/*
962 	 * find_next_reservable_window() simply finds a reservable window
963 	 * inside the given range(start_block, group_end_block).
964 	 *
965 	 * To make sure the reservation window has a free bit inside it, we
966 	 * need to check the bitmap after we found a reservable window.
967 	 */
968 retry:
969 	ret = find_next_reservable_window(search_head, my_rsv, sb,
970 						start_block, group_end_block);
971 
972 	if (ret == -1) {
973 		if (!rsv_is_empty(&my_rsv->rsv_window))
974 			rsv_window_remove(sb, my_rsv);
975 		spin_unlock(rsv_lock);
976 		return -1;
977 	}
978 
979 	/*
980 	 * On success, find_next_reservable_window() returns the
981 	 * reservation window where there is a reservable space after it.
982 	 * Before we reserve this reservable space, we need
983 	 * to make sure there is at least a free block inside this region.
984 	 *
985 	 * Search the first free bit on the block bitmap.  Search starts from
986 	 * the start block of the reservable space we just found.
987 	 */
988 	spin_unlock(rsv_lock);
989 	first_free_block = bitmap_search_next_usable_block(
990 			my_rsv->rsv_start - group_first_block,
991 			bitmap_bh, group_end_block - group_first_block + 1);
992 
993 	if (first_free_block < 0) {
994 		/*
995 		 * no free block left on the bitmap, no point
996 		 * to reserve the space. return failed.
997 		 */
998 		spin_lock(rsv_lock);
999 		if (!rsv_is_empty(&my_rsv->rsv_window))
1000 			rsv_window_remove(sb, my_rsv);
1001 		spin_unlock(rsv_lock);
1002 		return -1;		/* failed */
1003 	}
1004 
1005 	start_block = first_free_block + group_first_block;
1006 	/*
1007 	 * check if the first free block is within the
1008 	 * free space we just reserved
1009 	 */
1010 	if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1011 		return 0;		/* success */
1012 	/*
1013 	 * if the first free bit we found is out of the reservable space
1014 	 * continue search for next reservable space,
1015 	 * start from where the free block is,
1016 	 * we also shift the list head to where we stopped last time
1017 	 */
1018 	search_head = my_rsv;
1019 	spin_lock(rsv_lock);
1020 	goto retry;
1021 }
1022 
1023 /**
1024  * try_to_extend_reservation()
1025  * @my_rsv:		given reservation window
1026  * @sb:			super block
1027  * @size:		the delta to extend
1028  *
1029  * Attempt to expand the reservation window large enough to have
1030  * required number of free blocks
1031  *
1032  * Since ext2_try_to_allocate() will always allocate blocks within
1033  * the reservation window range, if the window size is too small,
1034  * multiple blocks allocation has to stop at the end of the reservation
1035  * window. To make this more efficient, given the total number of
1036  * blocks needed and the current size of the window, we try to
1037  * expand the reservation window size if necessary on a best-effort
1038  * basis before ext2_new_blocks() tries to allocate blocks.
1039  */
1040 static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1041 			struct super_block *sb, int size)
1042 {
1043 	struct ext2_reserve_window_node *next_rsv;
1044 	struct rb_node *next;
1045 	spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1046 
1047 	if (!spin_trylock(rsv_lock))
1048 		return;
1049 
1050 	next = rb_next(&my_rsv->rsv_node);
1051 
1052 	if (!next)
1053 		my_rsv->rsv_end += size;
1054 	else {
1055 		next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1056 
1057 		if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1058 			my_rsv->rsv_end += size;
1059 		else
1060 			my_rsv->rsv_end = next_rsv->rsv_start - 1;
1061 	}
1062 	spin_unlock(rsv_lock);
1063 }
1064 
1065 /**
1066  * ext2_try_to_allocate_with_rsv()
1067  * @sb:			superblock
1068  * @group:		given allocation block group
1069  * @bitmap_bh:		bufferhead holds the block bitmap
1070  * @grp_goal:		given target block within the group
1071  * @count:		target number of blocks to allocate
1072  * @my_rsv:		reservation window
1073  *
1074  * This is the main function used to allocate a new block and its reservation
1075  * window.
1076  *
1077  * Each time when a new block allocation is need, first try to allocate from
1078  * its own reservation.  If it does not have a reservation window, instead of
1079  * looking for a free bit on bitmap first, then look up the reservation list to
1080  * see if it is inside somebody else's reservation window, we try to allocate a
1081  * reservation window for it starting from the goal first. Then do the block
1082  * allocation within the reservation window.
1083  *
1084  * This will avoid keeping on searching the reservation list again and
1085  * again when somebody is looking for a free block (without
1086  * reservation), and there are lots of free blocks, but they are all
1087  * being reserved.
1088  *
1089  * We use a red-black tree for the per-filesystem reservation list.
1090  */
1091 static ext2_grpblk_t
1092 ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1093 			struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1094 			struct ext2_reserve_window_node * my_rsv,
1095 			unsigned long *count)
1096 {
1097 	ext2_fsblk_t group_first_block, group_last_block;
1098 	ext2_grpblk_t ret = 0;
1099 	unsigned long num = *count;
1100 
1101 	/*
1102 	 * we don't deal with reservation when
1103 	 * filesystem is mounted without reservation
1104 	 * or the file is not a regular file
1105 	 * or last attempt to allocate a block with reservation turned on failed
1106 	 */
1107 	if (my_rsv == NULL) {
1108 		return ext2_try_to_allocate(sb, group, bitmap_bh,
1109 						grp_goal, count, NULL);
1110 	}
1111 	/*
1112 	 * grp_goal is a group relative block number (if there is a goal)
1113 	 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1114 	 * first block is a filesystem wide block number
1115 	 * first block is the block number of the first block in this group
1116 	 */
1117 	group_first_block = ext2_group_first_block_no(sb, group);
1118 	group_last_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
1119 
1120 	/*
1121 	 * Basically we will allocate a new block from inode's reservation
1122 	 * window.
1123 	 *
1124 	 * We need to allocate a new reservation window, if:
1125 	 * a) inode does not have a reservation window; or
1126 	 * b) last attempt to allocate a block from existing reservation
1127 	 *    failed; or
1128 	 * c) we come here with a goal and with a reservation window
1129 	 *
1130 	 * We do not need to allocate a new reservation window if we come here
1131 	 * at the beginning with a goal and the goal is inside the window, or
1132 	 * we don't have a goal but already have a reservation window.
1133 	 * then we could go to allocate from the reservation window directly.
1134 	 */
1135 	while (1) {
1136 		if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1137 			!goal_in_my_reservation(&my_rsv->rsv_window,
1138 						grp_goal, group, sb)) {
1139 			if (my_rsv->rsv_goal_size < *count)
1140 				my_rsv->rsv_goal_size = *count;
1141 			ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1142 							group, bitmap_bh);
1143 			if (ret < 0)
1144 				break;			/* failed */
1145 
1146 			if (!goal_in_my_reservation(&my_rsv->rsv_window,
1147 							grp_goal, group, sb))
1148 				grp_goal = -1;
1149 		} else if (grp_goal >= 0) {
1150 			int curr = my_rsv->rsv_end -
1151 					(grp_goal + group_first_block) + 1;
1152 
1153 			if (curr < *count)
1154 				try_to_extend_reservation(my_rsv, sb,
1155 							*count - curr);
1156 		}
1157 
1158 		if ((my_rsv->rsv_start > group_last_block) ||
1159 				(my_rsv->rsv_end < group_first_block)) {
1160 			rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1161 			BUG();
1162 		}
1163 		ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1164 					   &num, &my_rsv->rsv_window);
1165 		if (ret >= 0) {
1166 			my_rsv->rsv_alloc_hit += num;
1167 			*count = num;
1168 			break;				/* succeed */
1169 		}
1170 		num = *count;
1171 	}
1172 	return ret;
1173 }
1174 
1175 /**
1176  * ext2_has_free_blocks()
1177  * @sbi:		in-core super block structure.
1178  *
1179  * Check if filesystem has at least 1 free block available for allocation.
1180  */
1181 static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1182 {
1183 	ext2_fsblk_t free_blocks, root_blocks;
1184 
1185 	free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1186 	root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1187 	if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1188 		!uid_eq(sbi->s_resuid, current_fsuid()) &&
1189 		(gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1190 		 !in_group_p (sbi->s_resgid))) {
1191 		return 0;
1192 	}
1193 	return 1;
1194 }
1195 
1196 /*
1197  * Returns 1 if the passed-in block region is valid; 0 if some part overlaps
1198  * with filesystem metadata blocksi.
1199  */
1200 int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk,
1201 			  unsigned int count)
1202 {
1203 	if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
1204 	    (start_blk + count < start_blk) ||
1205 	    (start_blk > le32_to_cpu(sbi->s_es->s_blocks_count)))
1206 		return 0;
1207 
1208 	/* Ensure we do not step over superblock */
1209 	if ((start_blk <= sbi->s_sb_block) &&
1210 	    (start_blk + count >= sbi->s_sb_block))
1211 		return 0;
1212 
1213 
1214 	return 1;
1215 }
1216 
1217 /*
1218  * ext2_new_blocks() -- core block(s) allocation function
1219  * @inode:		file inode
1220  * @goal:		given target block(filesystem wide)
1221  * @count:		target number of blocks to allocate
1222  * @errp:		error code
1223  *
1224  * ext2_new_blocks uses a goal block to assist allocation.  If the goal is
1225  * free, or there is a free block within 32 blocks of the goal, that block
1226  * is allocated.  Otherwise a forward search is made for a free block; within
1227  * each block group the search first looks for an entire free byte in the block
1228  * bitmap, and then for any free bit if that fails.
1229  * This function also updates quota and i_blocks field.
1230  */
1231 ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1232 		    unsigned long *count, int *errp)
1233 {
1234 	struct buffer_head *bitmap_bh = NULL;
1235 	struct buffer_head *gdp_bh;
1236 	int group_no;
1237 	int goal_group;
1238 	ext2_grpblk_t grp_target_blk;	/* blockgroup relative goal block */
1239 	ext2_grpblk_t grp_alloc_blk;	/* blockgroup-relative allocated block*/
1240 	ext2_fsblk_t ret_block;		/* filesyetem-wide allocated block */
1241 	int bgi;			/* blockgroup iteration index */
1242 	int performed_allocation = 0;
1243 	ext2_grpblk_t free_blocks;	/* number of free blocks in a group */
1244 	struct super_block *sb;
1245 	struct ext2_group_desc *gdp;
1246 	struct ext2_super_block *es;
1247 	struct ext2_sb_info *sbi;
1248 	struct ext2_reserve_window_node *my_rsv = NULL;
1249 	struct ext2_block_alloc_info *block_i;
1250 	unsigned short windowsz = 0;
1251 	unsigned long ngroups;
1252 	unsigned long num = *count;
1253 	int ret;
1254 
1255 	*errp = -ENOSPC;
1256 	sb = inode->i_sb;
1257 
1258 	/*
1259 	 * Check quota for allocation of this block.
1260 	 */
1261 	ret = dquot_alloc_block(inode, num);
1262 	if (ret) {
1263 		*errp = ret;
1264 		return 0;
1265 	}
1266 
1267 	sbi = EXT2_SB(sb);
1268 	es = EXT2_SB(sb)->s_es;
1269 	ext2_debug("goal=%lu.\n", goal);
1270 	/*
1271 	 * Allocate a block from reservation only when
1272 	 * filesystem is mounted with reservation(default,-o reservation), and
1273 	 * it's a regular file, and
1274 	 * the desired window size is greater than 0 (One could use ioctl
1275 	 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1276 	 * reservation on that particular file)
1277 	 */
1278 	block_i = EXT2_I(inode)->i_block_alloc_info;
1279 	if (block_i) {
1280 		windowsz = block_i->rsv_window_node.rsv_goal_size;
1281 		if (windowsz > 0)
1282 			my_rsv = &block_i->rsv_window_node;
1283 	}
1284 
1285 	if (!ext2_has_free_blocks(sbi)) {
1286 		*errp = -ENOSPC;
1287 		goto out;
1288 	}
1289 
1290 	/*
1291 	 * First, test whether the goal block is free.
1292 	 */
1293 	if (goal < le32_to_cpu(es->s_first_data_block) ||
1294 	    goal >= le32_to_cpu(es->s_blocks_count))
1295 		goal = le32_to_cpu(es->s_first_data_block);
1296 	group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1297 			EXT2_BLOCKS_PER_GROUP(sb);
1298 	goal_group = group_no;
1299 retry_alloc:
1300 	gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1301 	if (!gdp)
1302 		goto io_error;
1303 
1304 	free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1305 	/*
1306 	 * if there is not enough free blocks to make a new resevation
1307 	 * turn off reservation for this allocation
1308 	 */
1309 	if (my_rsv && (free_blocks < windowsz)
1310 		&& (free_blocks > 0)
1311 		&& (rsv_is_empty(&my_rsv->rsv_window)))
1312 		my_rsv = NULL;
1313 
1314 	if (free_blocks > 0) {
1315 		grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1316 				EXT2_BLOCKS_PER_GROUP(sb));
1317 		bitmap_bh = read_block_bitmap(sb, group_no);
1318 		if (!bitmap_bh)
1319 			goto io_error;
1320 		grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1321 					bitmap_bh, grp_target_blk,
1322 					my_rsv, &num);
1323 		if (grp_alloc_blk >= 0)
1324 			goto allocated;
1325 	}
1326 
1327 	ngroups = EXT2_SB(sb)->s_groups_count;
1328 	smp_rmb();
1329 
1330 	/*
1331 	 * Now search the rest of the groups.  We assume that
1332 	 * group_no and gdp correctly point to the last group visited.
1333 	 */
1334 	for (bgi = 0; bgi < ngroups; bgi++) {
1335 		group_no++;
1336 		if (group_no >= ngroups)
1337 			group_no = 0;
1338 		gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1339 		if (!gdp)
1340 			goto io_error;
1341 
1342 		free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1343 		/*
1344 		 * skip this group (and avoid loading bitmap) if there
1345 		 * are no free blocks
1346 		 */
1347 		if (!free_blocks)
1348 			continue;
1349 		/*
1350 		 * skip this group if the number of
1351 		 * free blocks is less than half of the reservation
1352 		 * window size.
1353 		 */
1354 		if (my_rsv && (free_blocks <= (windowsz/2)))
1355 			continue;
1356 
1357 		brelse(bitmap_bh);
1358 		bitmap_bh = read_block_bitmap(sb, group_no);
1359 		if (!bitmap_bh)
1360 			goto io_error;
1361 		/*
1362 		 * try to allocate block(s) from this group, without a goal(-1).
1363 		 */
1364 		grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1365 					bitmap_bh, -1, my_rsv, &num);
1366 		if (grp_alloc_blk >= 0)
1367 			goto allocated;
1368 	}
1369 	/*
1370 	 * We may end up a bogus earlier ENOSPC error due to
1371 	 * filesystem is "full" of reservations, but
1372 	 * there maybe indeed free blocks available on disk
1373 	 * In this case, we just forget about the reservations
1374 	 * just do block allocation as without reservations.
1375 	 */
1376 	if (my_rsv) {
1377 		my_rsv = NULL;
1378 		windowsz = 0;
1379 		group_no = goal_group;
1380 		goto retry_alloc;
1381 	}
1382 	/* No space left on the device */
1383 	*errp = -ENOSPC;
1384 	goto out;
1385 
1386 allocated:
1387 
1388 	ext2_debug("using block group %d(%d)\n",
1389 			group_no, gdp->bg_free_blocks_count);
1390 
1391 	ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1392 
1393 	if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1394 	    in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1395 	    in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1396 		      EXT2_SB(sb)->s_itb_per_group) ||
1397 	    in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1398 		      EXT2_SB(sb)->s_itb_per_group)) {
1399 		ext2_error(sb, "ext2_new_blocks",
1400 			    "Allocating block in system zone - "
1401 			    "blocks from "E2FSBLK", length %lu",
1402 			    ret_block, num);
1403 		/*
1404 		 * ext2_try_to_allocate marked the blocks we allocated as in
1405 		 * use.  So we may want to selectively mark some of the blocks
1406 		 * as free
1407 		 */
1408 		goto retry_alloc;
1409 	}
1410 
1411 	performed_allocation = 1;
1412 
1413 	if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1414 		ext2_error(sb, "ext2_new_blocks",
1415 			    "block("E2FSBLK") >= blocks count(%d) - "
1416 			    "block_group = %d, es == %p ", ret_block,
1417 			le32_to_cpu(es->s_blocks_count), group_no, es);
1418 		goto out;
1419 	}
1420 
1421 	group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1422 	percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1423 
1424 	mark_buffer_dirty(bitmap_bh);
1425 	if (sb->s_flags & MS_SYNCHRONOUS)
1426 		sync_dirty_buffer(bitmap_bh);
1427 
1428 	*errp = 0;
1429 	brelse(bitmap_bh);
1430 	if (num < *count) {
1431 		dquot_free_block_nodirty(inode, *count-num);
1432 		mark_inode_dirty(inode);
1433 		*count = num;
1434 	}
1435 	return ret_block;
1436 
1437 io_error:
1438 	*errp = -EIO;
1439 out:
1440 	/*
1441 	 * Undo the block allocation
1442 	 */
1443 	if (!performed_allocation) {
1444 		dquot_free_block_nodirty(inode, *count);
1445 		mark_inode_dirty(inode);
1446 	}
1447 	brelse(bitmap_bh);
1448 	return 0;
1449 }
1450 
1451 ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1452 {
1453 	unsigned long count = 1;
1454 
1455 	return ext2_new_blocks(inode, goal, &count, errp);
1456 }
1457 
1458 #ifdef EXT2FS_DEBUG
1459 
1460 unsigned long ext2_count_free(struct buffer_head *map, unsigned int numchars)
1461 {
1462 	return numchars * BITS_PER_BYTE - memweight(map->b_data, numchars);
1463 }
1464 
1465 #endif  /*  EXT2FS_DEBUG  */
1466 
1467 unsigned long ext2_count_free_blocks (struct super_block * sb)
1468 {
1469 	struct ext2_group_desc * desc;
1470 	unsigned long desc_count = 0;
1471 	int i;
1472 #ifdef EXT2FS_DEBUG
1473 	unsigned long bitmap_count, x;
1474 	struct ext2_super_block *es;
1475 
1476 	es = EXT2_SB(sb)->s_es;
1477 	desc_count = 0;
1478 	bitmap_count = 0;
1479 	desc = NULL;
1480 	for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1481 		struct buffer_head *bitmap_bh;
1482 		desc = ext2_get_group_desc (sb, i, NULL);
1483 		if (!desc)
1484 			continue;
1485 		desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1486 		bitmap_bh = read_block_bitmap(sb, i);
1487 		if (!bitmap_bh)
1488 			continue;
1489 
1490 		x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1491 		printk ("group %d: stored = %d, counted = %lu\n",
1492 			i, le16_to_cpu(desc->bg_free_blocks_count), x);
1493 		bitmap_count += x;
1494 		brelse(bitmap_bh);
1495 	}
1496 	printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1497 		(long)le32_to_cpu(es->s_free_blocks_count),
1498 		desc_count, bitmap_count);
1499 	return bitmap_count;
1500 #else
1501         for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1502                 desc = ext2_get_group_desc (sb, i, NULL);
1503                 if (!desc)
1504                         continue;
1505                 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1506 	}
1507 	return desc_count;
1508 #endif
1509 }
1510 
1511 static inline int test_root(int a, int b)
1512 {
1513 	int num = b;
1514 
1515 	while (a > num)
1516 		num *= b;
1517 	return num == a;
1518 }
1519 
1520 static int ext2_group_sparse(int group)
1521 {
1522 	if (group <= 1)
1523 		return 1;
1524 	return (test_root(group, 3) || test_root(group, 5) ||
1525 		test_root(group, 7));
1526 }
1527 
1528 /**
1529  *	ext2_bg_has_super - number of blocks used by the superblock in group
1530  *	@sb: superblock for filesystem
1531  *	@group: group number to check
1532  *
1533  *	Return the number of blocks used by the superblock (primary or backup)
1534  *	in this group.  Currently this will be only 0 or 1.
1535  */
1536 int ext2_bg_has_super(struct super_block *sb, int group)
1537 {
1538 	if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1539 	    !ext2_group_sparse(group))
1540 		return 0;
1541 	return 1;
1542 }
1543 
1544 /**
1545  *	ext2_bg_num_gdb - number of blocks used by the group table in group
1546  *	@sb: superblock for filesystem
1547  *	@group: group number to check
1548  *
1549  *	Return the number of blocks used by the group descriptor table
1550  *	(primary or backup) in this group.  In the future there may be a
1551  *	different number of descriptor blocks in each group.
1552  */
1553 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1554 {
1555 	return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
1556 }
1557 
1558