xref: /openbmc/linux/fs/ext2/balloc.c (revision 724ba675)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext2/balloc.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
11  *  Big-endian to little-endian byte-swapping/bitmaps by
12  *        David S. Miller (davem@caip.rutgers.edu), 1995
13  */
14 
15 #include "ext2.h"
16 #include <linux/quotaops.h>
17 #include <linux/slab.h>
18 #include <linux/sched.h>
19 #include <linux/cred.h>
20 #include <linux/buffer_head.h>
21 #include <linux/capability.h>
22 
23 /*
24  * balloc.c contains the blocks allocation and deallocation routines
25  */
26 
27 /*
28  * The free blocks are managed by bitmaps.  A file system contains several
29  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
30  * block for inodes, N blocks for the inode table and data blocks.
31  *
32  * The file system contains group descriptors which are located after the
33  * super block.  Each descriptor contains the number of the bitmap block and
34  * the free blocks count in the block.  The descriptors are loaded in memory
35  * when a file system is mounted (see ext2_fill_super).
36  */
37 
38 
39 #define in_range(b, first, len)	((b) >= (first) && (b) <= (first) + (len) - 1)
40 
41 struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
42 					     unsigned int block_group,
43 					     struct buffer_head ** bh)
44 {
45 	unsigned long group_desc;
46 	unsigned long offset;
47 	struct ext2_group_desc * desc;
48 	struct ext2_sb_info *sbi = EXT2_SB(sb);
49 
50 	if (block_group >= sbi->s_groups_count) {
51 		WARN(1, "block_group >= groups_count - "
52 		     "block_group = %d, groups_count = %lu",
53 		     block_group, sbi->s_groups_count);
54 
55 		return NULL;
56 	}
57 
58 	group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
59 	offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
60 	if (!sbi->s_group_desc[group_desc]) {
61 		WARN(1, "Group descriptor not loaded - "
62 		     "block_group = %d, group_desc = %lu, desc = %lu",
63 		      block_group, group_desc, offset);
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 	int ret;
130 
131 	desc = ext2_get_group_desc(sb, block_group, NULL);
132 	if (!desc)
133 		return NULL;
134 	bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
135 	bh = sb_getblk(sb, bitmap_blk);
136 	if (unlikely(!bh)) {
137 		ext2_error(sb, __func__,
138 			    "Cannot read block bitmap - "
139 			    "block_group = %d, block_bitmap = %u",
140 			    block_group, le32_to_cpu(desc->bg_block_bitmap));
141 		return NULL;
142 	}
143 	ret = bh_read(bh, 0);
144 	if (ret > 0)
145 		return bh;
146 	if (ret < 0) {
147 		brelse(bh);
148 		ext2_error(sb, __func__,
149 			    "Cannot read block bitmap - "
150 			    "block_group = %d, block_bitmap = %u",
151 			    block_group, le32_to_cpu(desc->bg_block_bitmap));
152 		return NULL;
153 	}
154 
155 	ext2_valid_block_bitmap(sb, desc, block_group, bh);
156 	/*
157 	 * file system mounted not to panic on error, continue with corrupt
158 	 * bitmap
159 	 */
160 	return bh;
161 }
162 
163 static void group_adjust_blocks(struct super_block *sb, int group_no,
164 	struct ext2_group_desc *desc, struct buffer_head *bh, int count)
165 {
166 	if (count) {
167 		struct ext2_sb_info *sbi = EXT2_SB(sb);
168 		unsigned free_blocks;
169 
170 		spin_lock(sb_bgl_lock(sbi, group_no));
171 		free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
172 		desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
173 		spin_unlock(sb_bgl_lock(sbi, group_no));
174 		mark_buffer_dirty(bh);
175 	}
176 }
177 
178 /*
179  * The reservation window structure operations
180  * --------------------------------------------
181  * Operations include:
182  * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
183  *
184  * We use a red-black tree to represent per-filesystem reservation
185  * windows.
186  *
187  */
188 
189 /**
190  * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
191  * @root:		root of per-filesystem reservation rb tree
192  * @verbose:		verbose mode
193  * @fn:			function which wishes to dump the reservation map
194  *
195  * If verbose is turned on, it will print the whole block reservation
196  * windows(start, end). Otherwise, it will only print out the "bad" windows,
197  * those windows that overlap with their immediate neighbors.
198  */
199 #if 1
200 static void __rsv_window_dump(struct rb_root *root, int verbose,
201 			      const char *fn)
202 {
203 	struct rb_node *n;
204 	struct ext2_reserve_window_node *rsv, *prev;
205 	int bad;
206 
207 restart:
208 	n = rb_first(root);
209 	bad = 0;
210 	prev = NULL;
211 
212 	printk("Block Allocation Reservation Windows Map (%s):\n", fn);
213 	while (n) {
214 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
215 		if (verbose)
216 			printk("reservation window 0x%p "
217 				"start: %lu, end: %lu\n",
218 				rsv, rsv->rsv_start, rsv->rsv_end);
219 		if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
220 			printk("Bad reservation %p (start >= end)\n",
221 			       rsv);
222 			bad = 1;
223 		}
224 		if (prev && prev->rsv_end >= rsv->rsv_start) {
225 			printk("Bad reservation %p (prev->end >= start)\n",
226 			       rsv);
227 			bad = 1;
228 		}
229 		if (bad) {
230 			if (!verbose) {
231 				printk("Restarting reservation walk in verbose mode\n");
232 				verbose = 1;
233 				goto restart;
234 			}
235 		}
236 		n = rb_next(n);
237 		prev = rsv;
238 	}
239 	printk("Window map complete.\n");
240 	BUG_ON(bad);
241 }
242 #define rsv_window_dump(root, verbose) \
243 	__rsv_window_dump((root), (verbose), __func__)
244 #else
245 #define rsv_window_dump(root, verbose) do {} while (0)
246 #endif
247 
248 /**
249  * goal_in_my_reservation()
250  * @rsv:		inode's reservation window
251  * @grp_goal:		given goal block relative to the allocation block group
252  * @group:		the current allocation block group
253  * @sb:			filesystem super block
254  *
255  * Test if the given goal block (group relative) is within the file's
256  * own block reservation window range.
257  *
258  * If the reservation window is outside the goal allocation group, return 0;
259  * grp_goal (given goal block) could be -1, which means no specific
260  * goal block. In this case, always return 1.
261  * If the goal block is within the reservation window, return 1;
262  * otherwise, return 0;
263  */
264 static int
265 goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
266 			unsigned int group, struct super_block * sb)
267 {
268 	ext2_fsblk_t group_first_block, group_last_block;
269 
270 	group_first_block = ext2_group_first_block_no(sb, group);
271 	group_last_block = ext2_group_last_block_no(sb, group);
272 
273 	if ((rsv->_rsv_start > group_last_block) ||
274 	    (rsv->_rsv_end < group_first_block))
275 		return 0;
276 	if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
277 		|| (grp_goal + group_first_block > rsv->_rsv_end)))
278 		return 0;
279 	return 1;
280 }
281 
282 /**
283  * search_reserve_window()
284  * @root:		root of reservation tree
285  * @goal:		target allocation block
286  *
287  * Find the reserved window which includes the goal, or the previous one
288  * if the goal is not in any window.
289  * Returns NULL if there are no windows or if all windows start after the goal.
290  */
291 static struct ext2_reserve_window_node *
292 search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
293 {
294 	struct rb_node *n = root->rb_node;
295 	struct ext2_reserve_window_node *rsv;
296 
297 	if (!n)
298 		return NULL;
299 
300 	do {
301 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
302 
303 		if (goal < rsv->rsv_start)
304 			n = n->rb_left;
305 		else if (goal > rsv->rsv_end)
306 			n = n->rb_right;
307 		else
308 			return rsv;
309 	} while (n);
310 	/*
311 	 * We've fallen off the end of the tree: the goal wasn't inside
312 	 * any particular node.  OK, the previous node must be to one
313 	 * side of the interval containing the goal.  If it's the RHS,
314 	 * we need to back up one.
315 	 */
316 	if (rsv->rsv_start > goal) {
317 		n = rb_prev(&rsv->rsv_node);
318 		rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
319 	}
320 	return rsv;
321 }
322 
323 /*
324  * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
325  * @sb:			super block
326  * @rsv:		reservation window to add
327  *
328  * Must be called with rsv_lock held.
329  */
330 void ext2_rsv_window_add(struct super_block *sb,
331 		    struct ext2_reserve_window_node *rsv)
332 {
333 	struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
334 	struct rb_node *node = &rsv->rsv_node;
335 	ext2_fsblk_t start = rsv->rsv_start;
336 
337 	struct rb_node ** p = &root->rb_node;
338 	struct rb_node * parent = NULL;
339 	struct ext2_reserve_window_node *this;
340 
341 	while (*p)
342 	{
343 		parent = *p;
344 		this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
345 
346 		if (start < this->rsv_start)
347 			p = &(*p)->rb_left;
348 		else if (start > this->rsv_end)
349 			p = &(*p)->rb_right;
350 		else {
351 			rsv_window_dump(root, 1);
352 			BUG();
353 		}
354 	}
355 
356 	rb_link_node(node, parent, p);
357 	rb_insert_color(node, root);
358 }
359 
360 /**
361  * rsv_window_remove() -- unlink a window from the reservation rb tree
362  * @sb:			super block
363  * @rsv:		reservation window to remove
364  *
365  * Mark the block reservation window as not allocated, and unlink it
366  * from the filesystem reservation window rb tree. Must be called with
367  * rsv_lock held.
368  */
369 static void rsv_window_remove(struct super_block *sb,
370 			      struct ext2_reserve_window_node *rsv)
371 {
372 	rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
373 	rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
374 	rsv->rsv_alloc_hit = 0;
375 	rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
376 }
377 
378 /*
379  * rsv_is_empty() -- Check if the reservation window is allocated.
380  * @rsv:		given reservation window to check
381  *
382  * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
383  */
384 static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
385 {
386 	/* a valid reservation end block could not be 0 */
387 	return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
388 }
389 
390 /**
391  * ext2_init_block_alloc_info()
392  * @inode:		file inode structure
393  *
394  * Allocate and initialize the  reservation window structure, and
395  * link the window to the ext2 inode structure at last
396  *
397  * The reservation window structure is only dynamically allocated
398  * and linked to ext2 inode the first time the open file
399  * needs a new block. So, before every ext2_new_block(s) call, for
400  * regular files, we should check whether the reservation window
401  * structure exists or not. In the latter case, this function is called.
402  * Fail to do so will result in block reservation being turned off for that
403  * open file.
404  *
405  * This function is called from ext2_get_blocks_handle(), also called
406  * when setting the reservation window size through ioctl before the file
407  * is open for write (needs block allocation).
408  *
409  * Needs truncate_mutex protection prior to calling this function.
410  */
411 void ext2_init_block_alloc_info(struct inode *inode)
412 {
413 	struct ext2_inode_info *ei = EXT2_I(inode);
414 	struct ext2_block_alloc_info *block_i;
415 	struct super_block *sb = inode->i_sb;
416 
417 	block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
418 	if (block_i) {
419 		struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
420 
421 		rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
422 		rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
423 
424 	 	/*
425 		 * if filesystem is mounted with NORESERVATION, the goal
426 		 * reservation window size is set to zero to indicate
427 		 * block reservation is off
428 		 */
429 		if (!test_opt(sb, RESERVATION))
430 			rsv->rsv_goal_size = 0;
431 		else
432 			rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
433 		rsv->rsv_alloc_hit = 0;
434 		block_i->last_alloc_logical_block = 0;
435 		block_i->last_alloc_physical_block = 0;
436 	}
437 	ei->i_block_alloc_info = block_i;
438 }
439 
440 /**
441  * ext2_discard_reservation()
442  * @inode:		inode
443  *
444  * Discard(free) block reservation window on last file close, or truncate
445  * or at last iput().
446  *
447  * It is being called in three cases:
448  * 	ext2_release_file(): last writer closes the file
449  * 	ext2_clear_inode(): last iput(), when nobody links to this file.
450  * 	ext2_truncate(): when the block indirect map is about to change.
451  */
452 void ext2_discard_reservation(struct inode *inode)
453 {
454 	struct ext2_inode_info *ei = EXT2_I(inode);
455 	struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
456 	struct ext2_reserve_window_node *rsv;
457 	spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
458 
459 	if (!block_i)
460 		return;
461 
462 	rsv = &block_i->rsv_window_node;
463 	if (!rsv_is_empty(&rsv->rsv_window)) {
464 		spin_lock(rsv_lock);
465 		if (!rsv_is_empty(&rsv->rsv_window))
466 			rsv_window_remove(inode->i_sb, rsv);
467 		spin_unlock(rsv_lock);
468 	}
469 }
470 
471 /**
472  * ext2_free_blocks() -- Free given blocks and update quota and i_blocks
473  * @inode:		inode
474  * @block:		start physical block to free
475  * @count:		number of blocks to free
476  */
477 void ext2_free_blocks (struct inode * inode, unsigned long block,
478 		       unsigned long count)
479 {
480 	struct buffer_head *bitmap_bh = NULL;
481 	struct buffer_head * bh2;
482 	unsigned long block_group;
483 	unsigned long bit;
484 	unsigned long i;
485 	unsigned long overflow;
486 	struct super_block * sb = inode->i_sb;
487 	struct ext2_sb_info * sbi = EXT2_SB(sb);
488 	struct ext2_group_desc * desc;
489 	struct ext2_super_block * es = sbi->s_es;
490 	unsigned freed = 0, group_freed;
491 
492 	if (!ext2_data_block_valid(sbi, block, count)) {
493 		ext2_error (sb, "ext2_free_blocks",
494 			    "Freeing blocks not in datazone - "
495 			    "block = %lu, count = %lu", block, count);
496 		goto error_return;
497 	}
498 
499 	ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
500 
501 do_more:
502 	overflow = 0;
503 	block_group = (block - le32_to_cpu(es->s_first_data_block)) /
504 		      EXT2_BLOCKS_PER_GROUP(sb);
505 	bit = (block - le32_to_cpu(es->s_first_data_block)) %
506 		      EXT2_BLOCKS_PER_GROUP(sb);
507 	/*
508 	 * Check to see if we are freeing blocks across a group
509 	 * boundary.
510 	 */
511 	if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
512 		overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
513 		count -= overflow;
514 	}
515 	brelse(bitmap_bh);
516 	bitmap_bh = read_block_bitmap(sb, block_group);
517 	if (!bitmap_bh)
518 		goto error_return;
519 
520 	desc = ext2_get_group_desc (sb, block_group, &bh2);
521 	if (!desc)
522 		goto error_return;
523 
524 	if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
525 	    in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
526 	    in_range (block, le32_to_cpu(desc->bg_inode_table),
527 		      sbi->s_itb_per_group) ||
528 	    in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
529 		      sbi->s_itb_per_group)) {
530 		ext2_error (sb, "ext2_free_blocks",
531 			    "Freeing blocks in system zones - "
532 			    "Block = %lu, count = %lu",
533 			    block, count);
534 		goto error_return;
535 	}
536 
537 	for (i = 0, group_freed = 0; i < count; i++) {
538 		if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
539 						bit + i, bitmap_bh->b_data)) {
540 			ext2_error(sb, __func__,
541 				"bit already cleared for block %lu", block + i);
542 		} else {
543 			group_freed++;
544 		}
545 	}
546 
547 	mark_buffer_dirty(bitmap_bh);
548 	if (sb->s_flags & SB_SYNCHRONOUS)
549 		sync_dirty_buffer(bitmap_bh);
550 
551 	group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
552 	freed += group_freed;
553 
554 	if (overflow) {
555 		block += count;
556 		count = overflow;
557 		goto do_more;
558 	}
559 error_return:
560 	brelse(bitmap_bh);
561 	if (freed) {
562 		percpu_counter_add(&sbi->s_freeblocks_counter, freed);
563 		dquot_free_block_nodirty(inode, freed);
564 		mark_inode_dirty(inode);
565 	}
566 }
567 
568 /**
569  * bitmap_search_next_usable_block()
570  * @start:		the starting block (group relative) of the search
571  * @bh:			bufferhead contains the block group bitmap
572  * @maxblocks:		the ending block (group relative) of the reservation
573  *
574  * The bitmap search --- search forward through the actual bitmap on disk until
575  * we find a bit free.
576  */
577 static ext2_grpblk_t
578 bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
579 					ext2_grpblk_t maxblocks)
580 {
581 	ext2_grpblk_t next;
582 
583 	next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
584 	if (next >= maxblocks)
585 		return -1;
586 	return next;
587 }
588 
589 /**
590  * find_next_usable_block()
591  * @start:		the starting block (group relative) to find next
592  * 			allocatable block in bitmap.
593  * @bh:			bufferhead contains the block group bitmap
594  * @maxblocks:		the ending block (group relative) for the search
595  *
596  * Find an allocatable block in a bitmap.  We perform the "most
597  * appropriate allocation" algorithm of looking for a free block near
598  * the initial goal; then for a free byte somewhere in the bitmap;
599  * then for any free bit in the bitmap.
600  */
601 static ext2_grpblk_t
602 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
603 {
604 	ext2_grpblk_t here, next;
605 	char *p, *r;
606 
607 	if (start > 0) {
608 		/*
609 		 * The goal was occupied; search forward for a free
610 		 * block within the next XX blocks.
611 		 *
612 		 * end_goal is more or less random, but it has to be
613 		 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
614 		 * next 64-bit boundary is simple..
615 		 */
616 		ext2_grpblk_t end_goal = (start + 63) & ~63;
617 		if (end_goal > maxblocks)
618 			end_goal = maxblocks;
619 		here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
620 		if (here < end_goal)
621 			return here;
622 		ext2_debug("Bit not found near goal\n");
623 	}
624 
625 	here = start;
626 	if (here < 0)
627 		here = 0;
628 
629 	p = ((char *)bh->b_data) + (here >> 3);
630 	r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
631 	next = (r - ((char *)bh->b_data)) << 3;
632 
633 	if (next < maxblocks && next >= here)
634 		return next;
635 
636 	here = bitmap_search_next_usable_block(here, bh, maxblocks);
637 	return here;
638 }
639 
640 /**
641  * ext2_try_to_allocate()
642  * @sb:			superblock
643  * @group:		given allocation block group
644  * @bitmap_bh:		bufferhead holds the block bitmap
645  * @grp_goal:		given target block within the group
646  * @count:		target number of blocks to allocate
647  * @my_rsv:		reservation window
648  *
649  * Attempt to allocate blocks within a give range. Set the range of allocation
650  * first, then find the first free bit(s) from the bitmap (within the range),
651  * and at last, allocate the blocks by claiming the found free bit as allocated.
652  *
653  * To set the range of this allocation:
654  * 	if there is a reservation window, only try to allocate block(s)
655  * 	from the file's own reservation window;
656  * 	Otherwise, the allocation range starts from the give goal block,
657  * 	ends at the block group's last block.
658  *
659  * If we failed to allocate the desired block then we may end up crossing to a
660  * new bitmap.
661  */
662 static int
663 ext2_try_to_allocate(struct super_block *sb, int group,
664 			struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
665 			unsigned long *count,
666 			struct ext2_reserve_window *my_rsv)
667 {
668 	ext2_fsblk_t group_first_block = ext2_group_first_block_no(sb, group);
669 	ext2_fsblk_t group_last_block = ext2_group_last_block_no(sb, group);
670 	ext2_grpblk_t start, end;
671 	unsigned long num = 0;
672 
673 	start = 0;
674 	end = group_last_block - group_first_block + 1;
675 	/* we do allocation within the reservation window if we have a window */
676 	if (my_rsv) {
677 		if (my_rsv->_rsv_start >= group_first_block)
678 			start = my_rsv->_rsv_start - group_first_block;
679 		if (my_rsv->_rsv_end < group_last_block)
680 			end = my_rsv->_rsv_end - group_first_block + 1;
681 		if (grp_goal < start || grp_goal >= end)
682 			grp_goal = -1;
683 	}
684 	BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
685 
686 	if (grp_goal < 0) {
687 		grp_goal = find_next_usable_block(start, bitmap_bh, end);
688 		if (grp_goal < 0)
689 			goto fail_access;
690 		if (!my_rsv) {
691 			int i;
692 
693 			for (i = 0; i < 7 && grp_goal > start &&
694 					!ext2_test_bit(grp_goal - 1,
695 					     		bitmap_bh->b_data);
696 			     		i++, grp_goal--)
697 				;
698 		}
699 	}
700 
701 	for (; num < *count && grp_goal < end; grp_goal++) {
702 		if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
703 					grp_goal, bitmap_bh->b_data)) {
704 			if (num == 0)
705 				continue;
706 			break;
707 		}
708 		num++;
709 	}
710 
711 	if (num == 0)
712 		goto fail_access;
713 
714 	*count = num;
715 	return grp_goal - num;
716 fail_access:
717 	return -1;
718 }
719 
720 /**
721  * 	find_next_reservable_window():
722  *		find a reservable space within the given range.
723  *		It does not allocate the reservation window for now:
724  *		alloc_new_reservation() will do the work later.
725  *
726  * 	@search_head: the head of the searching list;
727  *		This is not necessarily the list head of the whole filesystem
728  *
729  *		We have both head and start_block to assist the search
730  *		for the reservable space. The list starts from head,
731  *		but we will shift to the place where start_block is,
732  *		then start from there, when looking for a reservable space.
733  *
734  *	@sb: the super block.
735  *
736  * 	@start_block: the first block we consider to start the real search from
737  *
738  * 	@last_block:
739  *		the maximum block number that our goal reservable space
740  *		could start from. This is normally the last block in this
741  *		group. The search will end when we found the start of next
742  *		possible reservable space is out of this boundary.
743  *		This could handle the cross boundary reservation window
744  *		request.
745  *
746  * 	basically we search from the given range, rather than the whole
747  * 	reservation double linked list, (start_block, last_block)
748  * 	to find a free region that is of my size and has not
749  * 	been reserved.
750  *
751  */
752 static int find_next_reservable_window(
753 				struct ext2_reserve_window_node *search_head,
754 				struct ext2_reserve_window_node *my_rsv,
755 				struct super_block * sb,
756 				ext2_fsblk_t start_block,
757 				ext2_fsblk_t last_block)
758 {
759 	struct rb_node *next;
760 	struct ext2_reserve_window_node *rsv, *prev;
761 	ext2_fsblk_t cur;
762 	int size = my_rsv->rsv_goal_size;
763 
764 	/* TODO: make the start of the reservation window byte-aligned */
765 	/* cur = *start_block & ~7;*/
766 	cur = start_block;
767 	rsv = search_head;
768 	if (!rsv)
769 		return -1;
770 
771 	while (1) {
772 		if (cur <= rsv->rsv_end)
773 			cur = rsv->rsv_end + 1;
774 
775 		/* TODO?
776 		 * in the case we could not find a reservable space
777 		 * that is what is expected, during the re-search, we could
778 		 * remember what's the largest reservable space we could have
779 		 * and return that one.
780 		 *
781 		 * For now it will fail if we could not find the reservable
782 		 * space with expected-size (or more)...
783 		 */
784 		if (cur > last_block)
785 			return -1;		/* fail */
786 
787 		prev = rsv;
788 		next = rb_next(&rsv->rsv_node);
789 		rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
790 
791 		/*
792 		 * Reached the last reservation, we can just append to the
793 		 * previous one.
794 		 */
795 		if (!next)
796 			break;
797 
798 		if (cur + size <= rsv->rsv_start) {
799 			/*
800 			 * Found a reserveable space big enough.  We could
801 			 * have a reservation across the group boundary here
802 		 	 */
803 			break;
804 		}
805 	}
806 	/*
807 	 * we come here either :
808 	 * when we reach the end of the whole list,
809 	 * and there is empty reservable space after last entry in the list.
810 	 * append it to the end of the list.
811 	 *
812 	 * or we found one reservable space in the middle of the list,
813 	 * return the reservation window that we could append to.
814 	 * succeed.
815 	 */
816 
817 	if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
818 		rsv_window_remove(sb, my_rsv);
819 
820 	/*
821 	 * Let's book the whole available window for now.  We will check the
822 	 * disk bitmap later and then, if there are free blocks then we adjust
823 	 * the window size if it's larger than requested.
824 	 * Otherwise, we will remove this node from the tree next time
825 	 * call find_next_reservable_window.
826 	 */
827 	my_rsv->rsv_start = cur;
828 	my_rsv->rsv_end = cur + size - 1;
829 	my_rsv->rsv_alloc_hit = 0;
830 
831 	if (prev != my_rsv)
832 		ext2_rsv_window_add(sb, my_rsv);
833 
834 	return 0;
835 }
836 
837 /**
838  * 	alloc_new_reservation()--allocate a new reservation window
839  *
840  *		To make a new reservation, we search part of the filesystem
841  *		reservation list (the list that inside the group). We try to
842  *		allocate a new reservation window near the allocation goal,
843  *		or the beginning of the group, if there is no goal.
844  *
845  *		We first find a reservable space after the goal, then from
846  *		there, we check the bitmap for the first free block after
847  *		it. If there is no free block until the end of group, then the
848  *		whole group is full, we failed. Otherwise, check if the free
849  *		block is inside the expected reservable space, if so, we
850  *		succeed.
851  *		If the first free block is outside the reservable space, then
852  *		start from the first free block, we search for next available
853  *		space, and go on.
854  *
855  *	on succeed, a new reservation will be found and inserted into the list
856  *	It contains at least one free block, and it does not overlap with other
857  *	reservation windows.
858  *
859  *	failed: we failed to find a reservation window in this group
860  *
861  *	@my_rsv: the reservation
862  *
863  *	@grp_goal: The goal (group-relative).  It is where the search for a
864  *		free reservable space should start from.
865  *		if we have a goal(goal >0 ), then start from there,
866  *		no goal(goal = -1), we start from the first block
867  *		of the group.
868  *
869  *	@sb: the super block
870  *	@group: the group we are trying to allocate in
871  *	@bitmap_bh: the block group block bitmap
872  *
873  */
874 static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
875 		ext2_grpblk_t grp_goal, struct super_block *sb,
876 		unsigned int group, struct buffer_head *bitmap_bh)
877 {
878 	struct ext2_reserve_window_node *search_head;
879 	ext2_fsblk_t group_first_block, group_end_block, start_block;
880 	ext2_grpblk_t first_free_block;
881 	struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
882 	unsigned long size;
883 	int ret;
884 	spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
885 
886 	group_first_block = ext2_group_first_block_no(sb, group);
887 	group_end_block = ext2_group_last_block_no(sb, group);
888 
889 	if (grp_goal < 0)
890 		start_block = group_first_block;
891 	else
892 		start_block = grp_goal + group_first_block;
893 
894 	size = my_rsv->rsv_goal_size;
895 
896 	if (!rsv_is_empty(&my_rsv->rsv_window)) {
897 		/*
898 		 * if the old reservation is cross group boundary
899 		 * and if the goal is inside the old reservation window,
900 		 * we will come here when we just failed to allocate from
901 		 * the first part of the window. We still have another part
902 		 * that belongs to the next group. In this case, there is no
903 		 * point to discard our window and try to allocate a new one
904 		 * in this group(which will fail). we should
905 		 * keep the reservation window, just simply move on.
906 		 *
907 		 * Maybe we could shift the start block of the reservation
908 		 * window to the first block of next group.
909 		 */
910 
911 		if ((my_rsv->rsv_start <= group_end_block) &&
912 				(my_rsv->rsv_end > group_end_block) &&
913 				(start_block >= my_rsv->rsv_start))
914 			return -1;
915 
916 		if ((my_rsv->rsv_alloc_hit >
917 		     (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
918 			/*
919 			 * if the previously allocation hit ratio is
920 			 * greater than 1/2, then we double the size of
921 			 * the reservation window the next time,
922 			 * otherwise we keep the same size window
923 			 */
924 			size = size * 2;
925 			if (size > EXT2_MAX_RESERVE_BLOCKS)
926 				size = EXT2_MAX_RESERVE_BLOCKS;
927 			my_rsv->rsv_goal_size= size;
928 		}
929 	}
930 
931 	spin_lock(rsv_lock);
932 	/*
933 	 * shift the search start to the window near the goal block
934 	 */
935 	search_head = search_reserve_window(fs_rsv_root, start_block);
936 
937 	/*
938 	 * find_next_reservable_window() simply finds a reservable window
939 	 * inside the given range(start_block, group_end_block).
940 	 *
941 	 * To make sure the reservation window has a free bit inside it, we
942 	 * need to check the bitmap after we found a reservable window.
943 	 */
944 retry:
945 	ret = find_next_reservable_window(search_head, my_rsv, sb,
946 						start_block, group_end_block);
947 
948 	if (ret == -1) {
949 		if (!rsv_is_empty(&my_rsv->rsv_window))
950 			rsv_window_remove(sb, my_rsv);
951 		spin_unlock(rsv_lock);
952 		return -1;
953 	}
954 
955 	/*
956 	 * On success, find_next_reservable_window() returns the
957 	 * reservation window where there is a reservable space after it.
958 	 * Before we reserve this reservable space, we need
959 	 * to make sure there is at least a free block inside this region.
960 	 *
961 	 * Search the first free bit on the block bitmap.  Search starts from
962 	 * the start block of the reservable space we just found.
963 	 */
964 	spin_unlock(rsv_lock);
965 	first_free_block = bitmap_search_next_usable_block(
966 			my_rsv->rsv_start - group_first_block,
967 			bitmap_bh, group_end_block - group_first_block + 1);
968 
969 	if (first_free_block < 0) {
970 		/*
971 		 * no free block left on the bitmap, no point
972 		 * to reserve the space. return failed.
973 		 */
974 		spin_lock(rsv_lock);
975 		if (!rsv_is_empty(&my_rsv->rsv_window))
976 			rsv_window_remove(sb, my_rsv);
977 		spin_unlock(rsv_lock);
978 		return -1;		/* failed */
979 	}
980 
981 	start_block = first_free_block + group_first_block;
982 	/*
983 	 * check if the first free block is within the
984 	 * free space we just reserved
985 	 */
986 	if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
987 		return 0;		/* success */
988 	/*
989 	 * if the first free bit we found is out of the reservable space
990 	 * continue search for next reservable space,
991 	 * start from where the free block is,
992 	 * we also shift the list head to where we stopped last time
993 	 */
994 	search_head = my_rsv;
995 	spin_lock(rsv_lock);
996 	goto retry;
997 }
998 
999 /**
1000  * try_to_extend_reservation()
1001  * @my_rsv:		given reservation window
1002  * @sb:			super block
1003  * @size:		the delta to extend
1004  *
1005  * Attempt to expand the reservation window large enough to have
1006  * required number of free blocks
1007  *
1008  * Since ext2_try_to_allocate() will always allocate blocks within
1009  * the reservation window range, if the window size is too small,
1010  * multiple blocks allocation has to stop at the end of the reservation
1011  * window. To make this more efficient, given the total number of
1012  * blocks needed and the current size of the window, we try to
1013  * expand the reservation window size if necessary on a best-effort
1014  * basis before ext2_new_blocks() tries to allocate blocks.
1015  */
1016 static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1017 			struct super_block *sb, int size)
1018 {
1019 	struct ext2_reserve_window_node *next_rsv;
1020 	struct rb_node *next;
1021 	spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1022 
1023 	if (!spin_trylock(rsv_lock))
1024 		return;
1025 
1026 	next = rb_next(&my_rsv->rsv_node);
1027 
1028 	if (!next)
1029 		my_rsv->rsv_end += size;
1030 	else {
1031 		next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1032 
1033 		if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1034 			my_rsv->rsv_end += size;
1035 		else
1036 			my_rsv->rsv_end = next_rsv->rsv_start - 1;
1037 	}
1038 	spin_unlock(rsv_lock);
1039 }
1040 
1041 /**
1042  * ext2_try_to_allocate_with_rsv()
1043  * @sb:			superblock
1044  * @group:		given allocation block group
1045  * @bitmap_bh:		bufferhead holds the block bitmap
1046  * @grp_goal:		given target block within the group
1047  * @count:		target number of blocks to allocate
1048  * @my_rsv:		reservation window
1049  *
1050  * This is the main function used to allocate a new block and its reservation
1051  * window.
1052  *
1053  * Each time when a new block allocation is need, first try to allocate from
1054  * its own reservation.  If it does not have a reservation window, instead of
1055  * looking for a free bit on bitmap first, then look up the reservation list to
1056  * see if it is inside somebody else's reservation window, we try to allocate a
1057  * reservation window for it starting from the goal first. Then do the block
1058  * allocation within the reservation window.
1059  *
1060  * This will avoid keeping on searching the reservation list again and
1061  * again when somebody is looking for a free block (without
1062  * reservation), and there are lots of free blocks, but they are all
1063  * being reserved.
1064  *
1065  * We use a red-black tree for the per-filesystem reservation list.
1066  */
1067 static ext2_grpblk_t
1068 ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1069 			struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1070 			struct ext2_reserve_window_node * my_rsv,
1071 			unsigned long *count)
1072 {
1073 	ext2_fsblk_t group_first_block, group_last_block;
1074 	ext2_grpblk_t ret = 0;
1075 	unsigned long num = *count;
1076 
1077 	/*
1078 	 * we don't deal with reservation when
1079 	 * filesystem is mounted without reservation
1080 	 * or the file is not a regular file
1081 	 * or last attempt to allocate a block with reservation turned on failed
1082 	 */
1083 	if (my_rsv == NULL) {
1084 		return ext2_try_to_allocate(sb, group, bitmap_bh,
1085 						grp_goal, count, NULL);
1086 	}
1087 	/*
1088 	 * grp_goal is a group relative block number (if there is a goal)
1089 	 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1090 	 * first block is a filesystem wide block number
1091 	 * first block is the block number of the first block in this group
1092 	 */
1093 	group_first_block = ext2_group_first_block_no(sb, group);
1094 	group_last_block = ext2_group_last_block_no(sb, group);
1095 
1096 	/*
1097 	 * Basically we will allocate a new block from inode's reservation
1098 	 * window.
1099 	 *
1100 	 * We need to allocate a new reservation window, if:
1101 	 * a) inode does not have a reservation window; or
1102 	 * b) last attempt to allocate a block from existing reservation
1103 	 *    failed; or
1104 	 * c) we come here with a goal and with a reservation window
1105 	 *
1106 	 * We do not need to allocate a new reservation window if we come here
1107 	 * at the beginning with a goal and the goal is inside the window, or
1108 	 * we don't have a goal but already have a reservation window.
1109 	 * then we could go to allocate from the reservation window directly.
1110 	 */
1111 	while (1) {
1112 		if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1113 			!goal_in_my_reservation(&my_rsv->rsv_window,
1114 						grp_goal, group, sb)) {
1115 			if (my_rsv->rsv_goal_size < *count)
1116 				my_rsv->rsv_goal_size = *count;
1117 			ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1118 							group, bitmap_bh);
1119 			if (ret < 0)
1120 				break;			/* failed */
1121 
1122 			if (!goal_in_my_reservation(&my_rsv->rsv_window,
1123 							grp_goal, group, sb))
1124 				grp_goal = -1;
1125 		} else if (grp_goal >= 0) {
1126 			int curr = my_rsv->rsv_end -
1127 					(grp_goal + group_first_block) + 1;
1128 
1129 			if (curr < *count)
1130 				try_to_extend_reservation(my_rsv, sb,
1131 							*count - curr);
1132 		}
1133 
1134 		if ((my_rsv->rsv_start > group_last_block) ||
1135 				(my_rsv->rsv_end < group_first_block)) {
1136 			rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1137 			BUG();
1138 		}
1139 		ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1140 					   &num, &my_rsv->rsv_window);
1141 		if (ret >= 0) {
1142 			my_rsv->rsv_alloc_hit += num;
1143 			*count = num;
1144 			break;				/* succeed */
1145 		}
1146 		num = *count;
1147 	}
1148 	return ret;
1149 }
1150 
1151 /**
1152  * ext2_has_free_blocks()
1153  * @sbi:		in-core super block structure.
1154  *
1155  * Check if filesystem has at least 1 free block available for allocation.
1156  */
1157 static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1158 {
1159 	ext2_fsblk_t free_blocks, root_blocks;
1160 
1161 	free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1162 	root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1163 	if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1164 		!uid_eq(sbi->s_resuid, current_fsuid()) &&
1165 		(gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1166 		 !in_group_p (sbi->s_resgid))) {
1167 		return 0;
1168 	}
1169 	return 1;
1170 }
1171 
1172 /*
1173  * Returns 1 if the passed-in block region is valid; 0 if some part overlaps
1174  * with filesystem metadata blocks.
1175  */
1176 int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk,
1177 			  unsigned int count)
1178 {
1179 	if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
1180 	    (start_blk + count - 1 < start_blk) ||
1181 	    (start_blk + count - 1 >= le32_to_cpu(sbi->s_es->s_blocks_count)))
1182 		return 0;
1183 
1184 	/* Ensure we do not step over superblock */
1185 	if ((start_blk <= sbi->s_sb_block) &&
1186 	    (start_blk + count - 1 >= sbi->s_sb_block))
1187 		return 0;
1188 
1189 	return 1;
1190 }
1191 
1192 /*
1193  * ext2_new_blocks() -- core block(s) allocation function
1194  * @inode:		file inode
1195  * @goal:		given target block(filesystem wide)
1196  * @count:		target number of blocks to allocate
1197  * @errp:		error code
1198  *
1199  * ext2_new_blocks uses a goal block to assist allocation.  If the goal is
1200  * free, or there is a free block within 32 blocks of the goal, that block
1201  * is allocated.  Otherwise a forward search is made for a free block; within
1202  * each block group the search first looks for an entire free byte in the block
1203  * bitmap, and then for any free bit if that fails.
1204  * This function also updates quota and i_blocks field.
1205  */
1206 ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1207 		    unsigned long *count, int *errp)
1208 {
1209 	struct buffer_head *bitmap_bh = NULL;
1210 	struct buffer_head *gdp_bh;
1211 	int group_no;
1212 	int goal_group;
1213 	ext2_grpblk_t grp_target_blk;	/* blockgroup relative goal block */
1214 	ext2_grpblk_t grp_alloc_blk;	/* blockgroup-relative allocated block*/
1215 	ext2_fsblk_t ret_block;		/* filesyetem-wide allocated block */
1216 	int bgi;			/* blockgroup iteration index */
1217 	int performed_allocation = 0;
1218 	ext2_grpblk_t free_blocks;	/* number of free blocks in a group */
1219 	struct super_block *sb;
1220 	struct ext2_group_desc *gdp;
1221 	struct ext2_super_block *es;
1222 	struct ext2_sb_info *sbi;
1223 	struct ext2_reserve_window_node *my_rsv = NULL;
1224 	struct ext2_block_alloc_info *block_i;
1225 	unsigned short windowsz = 0;
1226 	unsigned long ngroups;
1227 	unsigned long num = *count;
1228 	int ret;
1229 
1230 	*errp = -ENOSPC;
1231 	sb = inode->i_sb;
1232 
1233 	/*
1234 	 * Check quota for allocation of this block.
1235 	 */
1236 	ret = dquot_alloc_block(inode, num);
1237 	if (ret) {
1238 		*errp = ret;
1239 		return 0;
1240 	}
1241 
1242 	sbi = EXT2_SB(sb);
1243 	es = EXT2_SB(sb)->s_es;
1244 	ext2_debug("goal=%lu.\n", goal);
1245 	/*
1246 	 * Allocate a block from reservation only when
1247 	 * filesystem is mounted with reservation(default,-o reservation), and
1248 	 * it's a regular file, and
1249 	 * the desired window size is greater than 0 (One could use ioctl
1250 	 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1251 	 * reservation on that particular file)
1252 	 */
1253 	block_i = EXT2_I(inode)->i_block_alloc_info;
1254 	if (block_i) {
1255 		windowsz = block_i->rsv_window_node.rsv_goal_size;
1256 		if (windowsz > 0)
1257 			my_rsv = &block_i->rsv_window_node;
1258 	}
1259 
1260 	if (!ext2_has_free_blocks(sbi)) {
1261 		*errp = -ENOSPC;
1262 		goto out;
1263 	}
1264 
1265 	/*
1266 	 * First, test whether the goal block is free.
1267 	 */
1268 	if (goal < le32_to_cpu(es->s_first_data_block) ||
1269 	    goal >= le32_to_cpu(es->s_blocks_count))
1270 		goal = le32_to_cpu(es->s_first_data_block);
1271 	group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1272 			EXT2_BLOCKS_PER_GROUP(sb);
1273 	goal_group = group_no;
1274 retry_alloc:
1275 	gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1276 	if (!gdp)
1277 		goto io_error;
1278 
1279 	free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1280 	/*
1281 	 * if there is not enough free blocks to make a new resevation
1282 	 * turn off reservation for this allocation
1283 	 */
1284 	if (my_rsv && (free_blocks < windowsz)
1285 		&& (free_blocks > 0)
1286 		&& (rsv_is_empty(&my_rsv->rsv_window)))
1287 		my_rsv = NULL;
1288 
1289 	if (free_blocks > 0) {
1290 		grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1291 				EXT2_BLOCKS_PER_GROUP(sb));
1292 		/*
1293 		 * In case we retry allocation (due to fs reservation not
1294 		 * working out or fs corruption), the bitmap_bh is non-null
1295 		 * pointer and we have to release it before calling
1296 		 * read_block_bitmap().
1297 		 */
1298 		brelse(bitmap_bh);
1299 		bitmap_bh = read_block_bitmap(sb, group_no);
1300 		if (!bitmap_bh)
1301 			goto io_error;
1302 		grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1303 					bitmap_bh, grp_target_blk,
1304 					my_rsv, &num);
1305 		if (grp_alloc_blk >= 0)
1306 			goto allocated;
1307 	}
1308 
1309 	ngroups = EXT2_SB(sb)->s_groups_count;
1310 	smp_rmb();
1311 
1312 	/*
1313 	 * Now search the rest of the groups.  We assume that
1314 	 * group_no and gdp correctly point to the last group visited.
1315 	 */
1316 	for (bgi = 0; bgi < ngroups; bgi++) {
1317 		group_no++;
1318 		if (group_no >= ngroups)
1319 			group_no = 0;
1320 		gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1321 		if (!gdp)
1322 			goto io_error;
1323 
1324 		free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1325 		/*
1326 		 * skip this group (and avoid loading bitmap) if there
1327 		 * are no free blocks
1328 		 */
1329 		if (!free_blocks)
1330 			continue;
1331 		/*
1332 		 * skip this group if the number of
1333 		 * free blocks is less than half of the reservation
1334 		 * window size.
1335 		 */
1336 		if (my_rsv && (free_blocks <= (windowsz/2)))
1337 			continue;
1338 
1339 		brelse(bitmap_bh);
1340 		bitmap_bh = read_block_bitmap(sb, group_no);
1341 		if (!bitmap_bh)
1342 			goto io_error;
1343 		/*
1344 		 * try to allocate block(s) from this group, without a goal(-1).
1345 		 */
1346 		grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1347 					bitmap_bh, -1, my_rsv, &num);
1348 		if (grp_alloc_blk >= 0)
1349 			goto allocated;
1350 	}
1351 	/*
1352 	 * We may end up a bogus earlier ENOSPC error due to
1353 	 * filesystem is "full" of reservations, but
1354 	 * there maybe indeed free blocks available on disk
1355 	 * In this case, we just forget about the reservations
1356 	 * just do block allocation as without reservations.
1357 	 */
1358 	if (my_rsv) {
1359 		my_rsv = NULL;
1360 		windowsz = 0;
1361 		group_no = goal_group;
1362 		goto retry_alloc;
1363 	}
1364 	/* No space left on the device */
1365 	*errp = -ENOSPC;
1366 	goto out;
1367 
1368 allocated:
1369 
1370 	ext2_debug("using block group %d(%d)\n",
1371 			group_no, gdp->bg_free_blocks_count);
1372 
1373 	ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1374 
1375 	if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1376 	    in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1377 	    in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1378 		      EXT2_SB(sb)->s_itb_per_group) ||
1379 	    in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1380 		      EXT2_SB(sb)->s_itb_per_group)) {
1381 		ext2_error(sb, "ext2_new_blocks",
1382 			    "Allocating block in system zone - "
1383 			    "blocks from "E2FSBLK", length %lu",
1384 			    ret_block, num);
1385 		/*
1386 		 * ext2_try_to_allocate marked the blocks we allocated as in
1387 		 * use.  So we may want to selectively mark some of the blocks
1388 		 * as free
1389 		 */
1390 		num = *count;
1391 		goto retry_alloc;
1392 	}
1393 
1394 	performed_allocation = 1;
1395 
1396 	if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1397 		ext2_error(sb, "ext2_new_blocks",
1398 			    "block("E2FSBLK") >= blocks count(%d) - "
1399 			    "block_group = %d, es == %p ", ret_block,
1400 			le32_to_cpu(es->s_blocks_count), group_no, es);
1401 		goto out;
1402 	}
1403 
1404 	group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1405 	percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1406 
1407 	mark_buffer_dirty(bitmap_bh);
1408 	if (sb->s_flags & SB_SYNCHRONOUS)
1409 		sync_dirty_buffer(bitmap_bh);
1410 
1411 	*errp = 0;
1412 	brelse(bitmap_bh);
1413 	if (num < *count) {
1414 		dquot_free_block_nodirty(inode, *count-num);
1415 		mark_inode_dirty(inode);
1416 		*count = num;
1417 	}
1418 	return ret_block;
1419 
1420 io_error:
1421 	*errp = -EIO;
1422 out:
1423 	/*
1424 	 * Undo the block allocation
1425 	 */
1426 	if (!performed_allocation) {
1427 		dquot_free_block_nodirty(inode, *count);
1428 		mark_inode_dirty(inode);
1429 	}
1430 	brelse(bitmap_bh);
1431 	return 0;
1432 }
1433 
1434 ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1435 {
1436 	unsigned long count = 1;
1437 
1438 	return ext2_new_blocks(inode, goal, &count, errp);
1439 }
1440 
1441 #ifdef EXT2FS_DEBUG
1442 
1443 unsigned long ext2_count_free(struct buffer_head *map, unsigned int numchars)
1444 {
1445 	return numchars * BITS_PER_BYTE - memweight(map->b_data, numchars);
1446 }
1447 
1448 #endif  /*  EXT2FS_DEBUG  */
1449 
1450 unsigned long ext2_count_free_blocks (struct super_block * sb)
1451 {
1452 	struct ext2_group_desc * desc;
1453 	unsigned long desc_count = 0;
1454 	int i;
1455 #ifdef EXT2FS_DEBUG
1456 	unsigned long bitmap_count, x;
1457 	struct ext2_super_block *es;
1458 
1459 	es = EXT2_SB(sb)->s_es;
1460 	desc_count = 0;
1461 	bitmap_count = 0;
1462 	desc = NULL;
1463 	for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1464 		struct buffer_head *bitmap_bh;
1465 		desc = ext2_get_group_desc (sb, i, NULL);
1466 		if (!desc)
1467 			continue;
1468 		desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1469 		bitmap_bh = read_block_bitmap(sb, i);
1470 		if (!bitmap_bh)
1471 			continue;
1472 
1473 		x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1474 		printk ("group %d: stored = %d, counted = %lu\n",
1475 			i, le16_to_cpu(desc->bg_free_blocks_count), x);
1476 		bitmap_count += x;
1477 		brelse(bitmap_bh);
1478 	}
1479 	printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1480 		(long)le32_to_cpu(es->s_free_blocks_count),
1481 		desc_count, bitmap_count);
1482 	return bitmap_count;
1483 #else
1484 	for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1485 		desc = ext2_get_group_desc(sb, i, NULL);
1486 		if (!desc)
1487 			continue;
1488 		desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1489 	}
1490 	return desc_count;
1491 #endif
1492 }
1493 
1494 static inline int test_root(int a, int b)
1495 {
1496 	int num = b;
1497 
1498 	while (a > num)
1499 		num *= b;
1500 	return num == a;
1501 }
1502 
1503 static int ext2_group_sparse(int group)
1504 {
1505 	if (group <= 1)
1506 		return 1;
1507 	return (test_root(group, 3) || test_root(group, 5) ||
1508 		test_root(group, 7));
1509 }
1510 
1511 /**
1512  *	ext2_bg_has_super - number of blocks used by the superblock in group
1513  *	@sb: superblock for filesystem
1514  *	@group: group number to check
1515  *
1516  *	Return the number of blocks used by the superblock (primary or backup)
1517  *	in this group.  Currently this will be only 0 or 1.
1518  */
1519 int ext2_bg_has_super(struct super_block *sb, int group)
1520 {
1521 	if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1522 	    !ext2_group_sparse(group))
1523 		return 0;
1524 	return 1;
1525 }
1526 
1527 /**
1528  *	ext2_bg_num_gdb - number of blocks used by the group table in group
1529  *	@sb: superblock for filesystem
1530  *	@group: group number to check
1531  *
1532  *	Return the number of blocks used by the group descriptor table
1533  *	(primary or backup) in this group.  In the future there may be a
1534  *	different number of descriptor blocks in each group.
1535  */
1536 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1537 {
1538 	return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
1539 }
1540 
1541