xref: /openbmc/linux/fs/ext2/balloc.c (revision d9679d00)
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 
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  * @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 = ext2_group_last_block_no(sb, group);
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  * @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 (!ext2_data_block_valid(sbi, block, count)) {
492 		ext2_error (sb, "ext2_free_blocks",
493 			    "Freeing blocks not in datazone - "
494 			    "block = %lu, count = %lu", block, count);
495 		goto error_return;
496 	}
497 
498 	ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
499 
500 do_more:
501 	overflow = 0;
502 	block_group = (block - le32_to_cpu(es->s_first_data_block)) /
503 		      EXT2_BLOCKS_PER_GROUP(sb);
504 	bit = (block - le32_to_cpu(es->s_first_data_block)) %
505 		      EXT2_BLOCKS_PER_GROUP(sb);
506 	/*
507 	 * Check to see if we are freeing blocks across a group
508 	 * boundary.
509 	 */
510 	if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
511 		overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
512 		count -= overflow;
513 	}
514 	brelse(bitmap_bh);
515 	bitmap_bh = read_block_bitmap(sb, block_group);
516 	if (!bitmap_bh)
517 		goto error_return;
518 
519 	desc = ext2_get_group_desc (sb, block_group, &bh2);
520 	if (!desc)
521 		goto error_return;
522 
523 	if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
524 	    in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
525 	    in_range (block, le32_to_cpu(desc->bg_inode_table),
526 		      sbi->s_itb_per_group) ||
527 	    in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
528 		      sbi->s_itb_per_group)) {
529 		ext2_error (sb, "ext2_free_blocks",
530 			    "Freeing blocks in system zones - "
531 			    "Block = %lu, count = %lu",
532 			    block, count);
533 		goto error_return;
534 	}
535 
536 	for (i = 0, group_freed = 0; i < count; i++) {
537 		if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
538 						bit + i, bitmap_bh->b_data)) {
539 			ext2_error(sb, __func__,
540 				"bit already cleared for block %lu", block + i);
541 		} else {
542 			group_freed++;
543 		}
544 	}
545 
546 	mark_buffer_dirty(bitmap_bh);
547 	if (sb->s_flags & SB_SYNCHRONOUS)
548 		sync_dirty_buffer(bitmap_bh);
549 
550 	group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
551 	freed += group_freed;
552 
553 	if (overflow) {
554 		block += count;
555 		count = overflow;
556 		goto do_more;
557 	}
558 error_return:
559 	brelse(bitmap_bh);
560 	if (freed) {
561 		percpu_counter_add(&sbi->s_freeblocks_counter, freed);
562 		dquot_free_block_nodirty(inode, freed);
563 		mark_inode_dirty(inode);
564 	}
565 }
566 
567 /**
568  * bitmap_search_next_usable_block()
569  * @start:		the starting block (group relative) of the search
570  * @bh:			bufferhead contains the block group bitmap
571  * @maxblocks:		the ending block (group relative) of the reservation
572  *
573  * The bitmap search --- search forward through the actual bitmap on disk until
574  * we find a bit free.
575  */
576 static ext2_grpblk_t
577 bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
578 					ext2_grpblk_t maxblocks)
579 {
580 	ext2_grpblk_t next;
581 
582 	next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
583 	if (next >= maxblocks)
584 		return -1;
585 	return next;
586 }
587 
588 /**
589  * find_next_usable_block()
590  * @start:		the starting block (group relative) to find next
591  * 			allocatable block in bitmap.
592  * @bh:			bufferhead contains the block group bitmap
593  * @maxblocks:		the ending block (group relative) for the search
594  *
595  * Find an allocatable block in a bitmap.  We perform the "most
596  * appropriate allocation" algorithm of looking for a free block near
597  * the initial goal; then for a free byte somewhere in the bitmap;
598  * then for any free bit in the bitmap.
599  */
600 static ext2_grpblk_t
601 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
602 {
603 	ext2_grpblk_t here, next;
604 	char *p, *r;
605 
606 	if (start > 0) {
607 		/*
608 		 * The goal was occupied; search forward for a free
609 		 * block within the next XX blocks.
610 		 *
611 		 * end_goal is more or less random, but it has to be
612 		 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
613 		 * next 64-bit boundary is simple..
614 		 */
615 		ext2_grpblk_t end_goal = (start + 63) & ~63;
616 		if (end_goal > maxblocks)
617 			end_goal = maxblocks;
618 		here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
619 		if (here < end_goal)
620 			return here;
621 		ext2_debug("Bit not found near goal\n");
622 	}
623 
624 	here = start;
625 	if (here < 0)
626 		here = 0;
627 
628 	p = ((char *)bh->b_data) + (here >> 3);
629 	r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
630 	next = (r - ((char *)bh->b_data)) << 3;
631 
632 	if (next < maxblocks && next >= here)
633 		return next;
634 
635 	here = bitmap_search_next_usable_block(here, bh, maxblocks);
636 	return here;
637 }
638 
639 /**
640  * ext2_try_to_allocate()
641  * @sb:			superblock
642  * @group:		given allocation block group
643  * @bitmap_bh:		bufferhead holds the block bitmap
644  * @grp_goal:		given target block within the group
645  * @count:		target number of blocks to allocate
646  * @my_rsv:		reservation window
647  *
648  * Attempt to allocate blocks within a give range. Set the range of allocation
649  * first, then find the first free bit(s) from the bitmap (within the range),
650  * and at last, allocate the blocks by claiming the found free bit as allocated.
651  *
652  * To set the range of this allocation:
653  * 	if there is a reservation window, only try to allocate block(s)
654  * 	from the file's own reservation window;
655  * 	Otherwise, the allocation range starts from the give goal block,
656  * 	ends at the block group's last block.
657  *
658  * If we failed to allocate the desired block then we may end up crossing to a
659  * new bitmap.
660  */
661 static int
662 ext2_try_to_allocate(struct super_block *sb, int group,
663 			struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
664 			unsigned long *count,
665 			struct ext2_reserve_window *my_rsv)
666 {
667 	ext2_fsblk_t group_first_block = ext2_group_first_block_no(sb, group);
668 	ext2_fsblk_t group_last_block = ext2_group_last_block_no(sb, group);
669        	ext2_grpblk_t start, end;
670 	unsigned long num = 0;
671 
672 	start = 0;
673 	end = group_last_block - group_first_block + 1;
674 	/* we do allocation within the reservation window if we have a window */
675 	if (my_rsv) {
676 		if (my_rsv->_rsv_start >= group_first_block)
677 			start = my_rsv->_rsv_start - group_first_block;
678 		if (my_rsv->_rsv_end < group_last_block)
679 			end = my_rsv->_rsv_end - group_first_block + 1;
680 		if (grp_goal < start || grp_goal >= end)
681 			grp_goal = -1;
682 	}
683 	BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
684 
685 	if (grp_goal < 0) {
686 		grp_goal = find_next_usable_block(start, bitmap_bh, end);
687 		if (grp_goal < 0)
688 			goto fail_access;
689 		if (!my_rsv) {
690 			int i;
691 
692 			for (i = 0; i < 7 && grp_goal > start &&
693 					!ext2_test_bit(grp_goal - 1,
694 					     		bitmap_bh->b_data);
695 			     		i++, grp_goal--)
696 				;
697 		}
698 	}
699 
700 	for (; num < *count && grp_goal < end; grp_goal++) {
701 		if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
702 					grp_goal, bitmap_bh->b_data)) {
703 			if (num == 0)
704 				continue;
705 			break;
706 		}
707 		num++;
708 	}
709 
710 	if (num == 0)
711 		goto fail_access;
712 
713 	*count = num;
714 	return grp_goal - num;
715 fail_access:
716 	return -1;
717 }
718 
719 /**
720  * 	find_next_reservable_window():
721  *		find a reservable space within the given range.
722  *		It does not allocate the reservation window for now:
723  *		alloc_new_reservation() will do the work later.
724  *
725  * 	@search_head: the head of the searching list;
726  *		This is not necessarily the list head of the whole filesystem
727  *
728  *		We have both head and start_block to assist the search
729  *		for the reservable space. The list starts from head,
730  *		but we will shift to the place where start_block is,
731  *		then start from there, when looking for a reservable space.
732  *
733  *	@sb: the super block.
734  *
735  * 	@start_block: the first block we consider to start the real search from
736  *
737  * 	@last_block:
738  *		the maximum block number that our goal reservable space
739  *		could start from. This is normally the last block in this
740  *		group. The search will end when we found the start of next
741  *		possible reservable space is out of this boundary.
742  *		This could handle the cross boundary reservation window
743  *		request.
744  *
745  * 	basically we search from the given range, rather than the whole
746  * 	reservation double linked list, (start_block, last_block)
747  * 	to find a free region that is of my size and has not
748  * 	been reserved.
749  *
750  */
751 static int find_next_reservable_window(
752 				struct ext2_reserve_window_node *search_head,
753 				struct ext2_reserve_window_node *my_rsv,
754 				struct super_block * sb,
755 				ext2_fsblk_t start_block,
756 				ext2_fsblk_t last_block)
757 {
758 	struct rb_node *next;
759 	struct ext2_reserve_window_node *rsv, *prev;
760 	ext2_fsblk_t cur;
761 	int size = my_rsv->rsv_goal_size;
762 
763 	/* TODO: make the start of the reservation window byte-aligned */
764 	/* cur = *start_block & ~7;*/
765 	cur = start_block;
766 	rsv = search_head;
767 	if (!rsv)
768 		return -1;
769 
770 	while (1) {
771 		if (cur <= rsv->rsv_end)
772 			cur = rsv->rsv_end + 1;
773 
774 		/* TODO?
775 		 * in the case we could not find a reservable space
776 		 * that is what is expected, during the re-search, we could
777 		 * remember what's the largest reservable space we could have
778 		 * and return that one.
779 		 *
780 		 * For now it will fail if we could not find the reservable
781 		 * space with expected-size (or more)...
782 		 */
783 		if (cur > last_block)
784 			return -1;		/* fail */
785 
786 		prev = rsv;
787 		next = rb_next(&rsv->rsv_node);
788 		rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
789 
790 		/*
791 		 * Reached the last reservation, we can just append to the
792 		 * previous one.
793 		 */
794 		if (!next)
795 			break;
796 
797 		if (cur + size <= rsv->rsv_start) {
798 			/*
799 			 * Found a reserveable space big enough.  We could
800 			 * have a reservation across the group boundary here
801 		 	 */
802 			break;
803 		}
804 	}
805 	/*
806 	 * we come here either :
807 	 * when we reach the end of the whole list,
808 	 * and there is empty reservable space after last entry in the list.
809 	 * append it to the end of the list.
810 	 *
811 	 * or we found one reservable space in the middle of the list,
812 	 * return the reservation window that we could append to.
813 	 * succeed.
814 	 */
815 
816 	if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
817 		rsv_window_remove(sb, my_rsv);
818 
819 	/*
820 	 * Let's book the whole available window for now.  We will check the
821 	 * disk bitmap later and then, if there are free blocks then we adjust
822 	 * the window size if it's larger than requested.
823 	 * Otherwise, we will remove this node from the tree next time
824 	 * call find_next_reservable_window.
825 	 */
826 	my_rsv->rsv_start = cur;
827 	my_rsv->rsv_end = cur + size - 1;
828 	my_rsv->rsv_alloc_hit = 0;
829 
830 	if (prev != my_rsv)
831 		ext2_rsv_window_add(sb, my_rsv);
832 
833 	return 0;
834 }
835 
836 /**
837  * 	alloc_new_reservation()--allocate a new reservation window
838  *
839  *		To make a new reservation, we search part of the filesystem
840  *		reservation list (the list that inside the group). We try to
841  *		allocate a new reservation window near the allocation goal,
842  *		or the beginning of the group, if there is no goal.
843  *
844  *		We first find a reservable space after the goal, then from
845  *		there, we check the bitmap for the first free block after
846  *		it. If there is no free block until the end of group, then the
847  *		whole group is full, we failed. Otherwise, check if the free
848  *		block is inside the expected reservable space, if so, we
849  *		succeed.
850  *		If the first free block is outside the reservable space, then
851  *		start from the first free block, we search for next available
852  *		space, and go on.
853  *
854  *	on succeed, a new reservation will be found and inserted into the list
855  *	It contains at least one free block, and it does not overlap with other
856  *	reservation windows.
857  *
858  *	failed: we failed to find a reservation window in this group
859  *
860  *	@my_rsv: the reservation
861  *
862  *	@grp_goal: The goal (group-relative).  It is where the search for a
863  *		free reservable space should start from.
864  *		if we have a goal(goal >0 ), then start from there,
865  *		no goal(goal = -1), we start from the first block
866  *		of the group.
867  *
868  *	@sb: the super block
869  *	@group: the group we are trying to allocate in
870  *	@bitmap_bh: the block group block bitmap
871  *
872  */
873 static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
874 		ext2_grpblk_t grp_goal, struct super_block *sb,
875 		unsigned int group, struct buffer_head *bitmap_bh)
876 {
877 	struct ext2_reserve_window_node *search_head;
878 	ext2_fsblk_t group_first_block, group_end_block, start_block;
879 	ext2_grpblk_t first_free_block;
880 	struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
881 	unsigned long size;
882 	int ret;
883 	spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
884 
885 	group_first_block = ext2_group_first_block_no(sb, group);
886 	group_end_block = ext2_group_last_block_no(sb, group);
887 
888 	if (grp_goal < 0)
889 		start_block = group_first_block;
890 	else
891 		start_block = grp_goal + group_first_block;
892 
893 	size = my_rsv->rsv_goal_size;
894 
895 	if (!rsv_is_empty(&my_rsv->rsv_window)) {
896 		/*
897 		 * if the old reservation is cross group boundary
898 		 * and if the goal is inside the old reservation window,
899 		 * we will come here when we just failed to allocate from
900 		 * the first part of the window. We still have another part
901 		 * that belongs to the next group. In this case, there is no
902 		 * point to discard our window and try to allocate a new one
903 		 * in this group(which will fail). we should
904 		 * keep the reservation window, just simply move on.
905 		 *
906 		 * Maybe we could shift the start block of the reservation
907 		 * window to the first block of next group.
908 		 */
909 
910 		if ((my_rsv->rsv_start <= group_end_block) &&
911 				(my_rsv->rsv_end > group_end_block) &&
912 				(start_block >= my_rsv->rsv_start))
913 			return -1;
914 
915 		if ((my_rsv->rsv_alloc_hit >
916 		     (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
917 			/*
918 			 * if the previously allocation hit ratio is
919 			 * greater than 1/2, then we double the size of
920 			 * the reservation window the next time,
921 			 * otherwise we keep the same size window
922 			 */
923 			size = size * 2;
924 			if (size > EXT2_MAX_RESERVE_BLOCKS)
925 				size = EXT2_MAX_RESERVE_BLOCKS;
926 			my_rsv->rsv_goal_size= size;
927 		}
928 	}
929 
930 	spin_lock(rsv_lock);
931 	/*
932 	 * shift the search start to the window near the goal block
933 	 */
934 	search_head = search_reserve_window(fs_rsv_root, start_block);
935 
936 	/*
937 	 * find_next_reservable_window() simply finds a reservable window
938 	 * inside the given range(start_block, group_end_block).
939 	 *
940 	 * To make sure the reservation window has a free bit inside it, we
941 	 * need to check the bitmap after we found a reservable window.
942 	 */
943 retry:
944 	ret = find_next_reservable_window(search_head, my_rsv, sb,
945 						start_block, group_end_block);
946 
947 	if (ret == -1) {
948 		if (!rsv_is_empty(&my_rsv->rsv_window))
949 			rsv_window_remove(sb, my_rsv);
950 		spin_unlock(rsv_lock);
951 		return -1;
952 	}
953 
954 	/*
955 	 * On success, find_next_reservable_window() returns the
956 	 * reservation window where there is a reservable space after it.
957 	 * Before we reserve this reservable space, we need
958 	 * to make sure there is at least a free block inside this region.
959 	 *
960 	 * Search the first free bit on the block bitmap.  Search starts from
961 	 * the start block of the reservable space we just found.
962 	 */
963 	spin_unlock(rsv_lock);
964 	first_free_block = bitmap_search_next_usable_block(
965 			my_rsv->rsv_start - group_first_block,
966 			bitmap_bh, group_end_block - group_first_block + 1);
967 
968 	if (first_free_block < 0) {
969 		/*
970 		 * no free block left on the bitmap, no point
971 		 * to reserve the space. return failed.
972 		 */
973 		spin_lock(rsv_lock);
974 		if (!rsv_is_empty(&my_rsv->rsv_window))
975 			rsv_window_remove(sb, my_rsv);
976 		spin_unlock(rsv_lock);
977 		return -1;		/* failed */
978 	}
979 
980 	start_block = first_free_block + group_first_block;
981 	/*
982 	 * check if the first free block is within the
983 	 * free space we just reserved
984 	 */
985 	if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
986 		return 0;		/* success */
987 	/*
988 	 * if the first free bit we found is out of the reservable space
989 	 * continue search for next reservable space,
990 	 * start from where the free block is,
991 	 * we also shift the list head to where we stopped last time
992 	 */
993 	search_head = my_rsv;
994 	spin_lock(rsv_lock);
995 	goto retry;
996 }
997 
998 /**
999  * try_to_extend_reservation()
1000  * @my_rsv:		given reservation window
1001  * @sb:			super block
1002  * @size:		the delta to extend
1003  *
1004  * Attempt to expand the reservation window large enough to have
1005  * required number of free blocks
1006  *
1007  * Since ext2_try_to_allocate() will always allocate blocks within
1008  * the reservation window range, if the window size is too small,
1009  * multiple blocks allocation has to stop at the end of the reservation
1010  * window. To make this more efficient, given the total number of
1011  * blocks needed and the current size of the window, we try to
1012  * expand the reservation window size if necessary on a best-effort
1013  * basis before ext2_new_blocks() tries to allocate blocks.
1014  */
1015 static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1016 			struct super_block *sb, int size)
1017 {
1018 	struct ext2_reserve_window_node *next_rsv;
1019 	struct rb_node *next;
1020 	spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1021 
1022 	if (!spin_trylock(rsv_lock))
1023 		return;
1024 
1025 	next = rb_next(&my_rsv->rsv_node);
1026 
1027 	if (!next)
1028 		my_rsv->rsv_end += size;
1029 	else {
1030 		next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1031 
1032 		if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1033 			my_rsv->rsv_end += size;
1034 		else
1035 			my_rsv->rsv_end = next_rsv->rsv_start - 1;
1036 	}
1037 	spin_unlock(rsv_lock);
1038 }
1039 
1040 /**
1041  * ext2_try_to_allocate_with_rsv()
1042  * @sb:			superblock
1043  * @group:		given allocation block group
1044  * @bitmap_bh:		bufferhead holds the block bitmap
1045  * @grp_goal:		given target block within the group
1046  * @count:		target number of blocks to allocate
1047  * @my_rsv:		reservation window
1048  *
1049  * This is the main function used to allocate a new block and its reservation
1050  * window.
1051  *
1052  * Each time when a new block allocation is need, first try to allocate from
1053  * its own reservation.  If it does not have a reservation window, instead of
1054  * looking for a free bit on bitmap first, then look up the reservation list to
1055  * see if it is inside somebody else's reservation window, we try to allocate a
1056  * reservation window for it starting from the goal first. Then do the block
1057  * allocation within the reservation window.
1058  *
1059  * This will avoid keeping on searching the reservation list again and
1060  * again when somebody is looking for a free block (without
1061  * reservation), and there are lots of free blocks, but they are all
1062  * being reserved.
1063  *
1064  * We use a red-black tree for the per-filesystem reservation list.
1065  */
1066 static ext2_grpblk_t
1067 ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1068 			struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1069 			struct ext2_reserve_window_node * my_rsv,
1070 			unsigned long *count)
1071 {
1072 	ext2_fsblk_t group_first_block, group_last_block;
1073 	ext2_grpblk_t ret = 0;
1074 	unsigned long num = *count;
1075 
1076 	/*
1077 	 * we don't deal with reservation when
1078 	 * filesystem is mounted without reservation
1079 	 * or the file is not a regular file
1080 	 * or last attempt to allocate a block with reservation turned on failed
1081 	 */
1082 	if (my_rsv == NULL) {
1083 		return ext2_try_to_allocate(sb, group, bitmap_bh,
1084 						grp_goal, count, NULL);
1085 	}
1086 	/*
1087 	 * grp_goal is a group relative block number (if there is a goal)
1088 	 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1089 	 * first block is a filesystem wide block number
1090 	 * first block is the block number of the first block in this group
1091 	 */
1092 	group_first_block = ext2_group_first_block_no(sb, group);
1093 	group_last_block = ext2_group_last_block_no(sb, group);
1094 
1095 	/*
1096 	 * Basically we will allocate a new block from inode's reservation
1097 	 * window.
1098 	 *
1099 	 * We need to allocate a new reservation window, if:
1100 	 * a) inode does not have a reservation window; or
1101 	 * b) last attempt to allocate a block from existing reservation
1102 	 *    failed; or
1103 	 * c) we come here with a goal and with a reservation window
1104 	 *
1105 	 * We do not need to allocate a new reservation window if we come here
1106 	 * at the beginning with a goal and the goal is inside the window, or
1107 	 * we don't have a goal but already have a reservation window.
1108 	 * then we could go to allocate from the reservation window directly.
1109 	 */
1110 	while (1) {
1111 		if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1112 			!goal_in_my_reservation(&my_rsv->rsv_window,
1113 						grp_goal, group, sb)) {
1114 			if (my_rsv->rsv_goal_size < *count)
1115 				my_rsv->rsv_goal_size = *count;
1116 			ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1117 							group, bitmap_bh);
1118 			if (ret < 0)
1119 				break;			/* failed */
1120 
1121 			if (!goal_in_my_reservation(&my_rsv->rsv_window,
1122 							grp_goal, group, sb))
1123 				grp_goal = -1;
1124 		} else if (grp_goal >= 0) {
1125 			int curr = my_rsv->rsv_end -
1126 					(grp_goal + group_first_block) + 1;
1127 
1128 			if (curr < *count)
1129 				try_to_extend_reservation(my_rsv, sb,
1130 							*count - curr);
1131 		}
1132 
1133 		if ((my_rsv->rsv_start > group_last_block) ||
1134 				(my_rsv->rsv_end < group_first_block)) {
1135 			rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1136 			BUG();
1137 		}
1138 		ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1139 					   &num, &my_rsv->rsv_window);
1140 		if (ret >= 0) {
1141 			my_rsv->rsv_alloc_hit += num;
1142 			*count = num;
1143 			break;				/* succeed */
1144 		}
1145 		num = *count;
1146 	}
1147 	return ret;
1148 }
1149 
1150 /**
1151  * ext2_has_free_blocks()
1152  * @sbi:		in-core super block structure.
1153  *
1154  * Check if filesystem has at least 1 free block available for allocation.
1155  */
1156 static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1157 {
1158 	ext2_fsblk_t free_blocks, root_blocks;
1159 
1160 	free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1161 	root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1162 	if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1163 		!uid_eq(sbi->s_resuid, current_fsuid()) &&
1164 		(gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1165 		 !in_group_p (sbi->s_resgid))) {
1166 		return 0;
1167 	}
1168 	return 1;
1169 }
1170 
1171 /*
1172  * Returns 1 if the passed-in block region is valid; 0 if some part overlaps
1173  * with filesystem metadata blocks.
1174  */
1175 int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk,
1176 			  unsigned int count)
1177 {
1178 	if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
1179 	    (start_blk + count - 1 < start_blk) ||
1180 	    (start_blk + count - 1 >= le32_to_cpu(sbi->s_es->s_blocks_count)))
1181 		return 0;
1182 
1183 	/* Ensure we do not step over superblock */
1184 	if ((start_blk <= sbi->s_sb_block) &&
1185 	    (start_blk + count - 1 >= sbi->s_sb_block))
1186 		return 0;
1187 
1188 	return 1;
1189 }
1190 
1191 /*
1192  * ext2_new_blocks() -- core block(s) allocation function
1193  * @inode:		file inode
1194  * @goal:		given target block(filesystem wide)
1195  * @count:		target number of blocks to allocate
1196  * @errp:		error code
1197  *
1198  * ext2_new_blocks uses a goal block to assist allocation.  If the goal is
1199  * free, or there is a free block within 32 blocks of the goal, that block
1200  * is allocated.  Otherwise a forward search is made for a free block; within
1201  * each block group the search first looks for an entire free byte in the block
1202  * bitmap, and then for any free bit if that fails.
1203  * This function also updates quota and i_blocks field.
1204  */
1205 ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1206 		    unsigned long *count, int *errp)
1207 {
1208 	struct buffer_head *bitmap_bh = NULL;
1209 	struct buffer_head *gdp_bh;
1210 	int group_no;
1211 	int goal_group;
1212 	ext2_grpblk_t grp_target_blk;	/* blockgroup relative goal block */
1213 	ext2_grpblk_t grp_alloc_blk;	/* blockgroup-relative allocated block*/
1214 	ext2_fsblk_t ret_block;		/* filesyetem-wide allocated block */
1215 	int bgi;			/* blockgroup iteration index */
1216 	int performed_allocation = 0;
1217 	ext2_grpblk_t free_blocks;	/* number of free blocks in a group */
1218 	struct super_block *sb;
1219 	struct ext2_group_desc *gdp;
1220 	struct ext2_super_block *es;
1221 	struct ext2_sb_info *sbi;
1222 	struct ext2_reserve_window_node *my_rsv = NULL;
1223 	struct ext2_block_alloc_info *block_i;
1224 	unsigned short windowsz = 0;
1225 	unsigned long ngroups;
1226 	unsigned long num = *count;
1227 	int ret;
1228 
1229 	*errp = -ENOSPC;
1230 	sb = inode->i_sb;
1231 
1232 	/*
1233 	 * Check quota for allocation of this block.
1234 	 */
1235 	ret = dquot_alloc_block(inode, num);
1236 	if (ret) {
1237 		*errp = ret;
1238 		return 0;
1239 	}
1240 
1241 	sbi = EXT2_SB(sb);
1242 	es = EXT2_SB(sb)->s_es;
1243 	ext2_debug("goal=%lu.\n", goal);
1244 	/*
1245 	 * Allocate a block from reservation only when
1246 	 * filesystem is mounted with reservation(default,-o reservation), and
1247 	 * it's a regular file, and
1248 	 * the desired window size is greater than 0 (One could use ioctl
1249 	 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1250 	 * reservation on that particular file)
1251 	 */
1252 	block_i = EXT2_I(inode)->i_block_alloc_info;
1253 	if (block_i) {
1254 		windowsz = block_i->rsv_window_node.rsv_goal_size;
1255 		if (windowsz > 0)
1256 			my_rsv = &block_i->rsv_window_node;
1257 	}
1258 
1259 	if (!ext2_has_free_blocks(sbi)) {
1260 		*errp = -ENOSPC;
1261 		goto out;
1262 	}
1263 
1264 	/*
1265 	 * First, test whether the goal block is free.
1266 	 */
1267 	if (goal < le32_to_cpu(es->s_first_data_block) ||
1268 	    goal >= le32_to_cpu(es->s_blocks_count))
1269 		goal = le32_to_cpu(es->s_first_data_block);
1270 	group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1271 			EXT2_BLOCKS_PER_GROUP(sb);
1272 	goal_group = group_no;
1273 retry_alloc:
1274 	gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1275 	if (!gdp)
1276 		goto io_error;
1277 
1278 	free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1279 	/*
1280 	 * if there is not enough free blocks to make a new resevation
1281 	 * turn off reservation for this allocation
1282 	 */
1283 	if (my_rsv && (free_blocks < windowsz)
1284 		&& (free_blocks > 0)
1285 		&& (rsv_is_empty(&my_rsv->rsv_window)))
1286 		my_rsv = NULL;
1287 
1288 	if (free_blocks > 0) {
1289 		grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1290 				EXT2_BLOCKS_PER_GROUP(sb));
1291 		/*
1292 		 * In case we retry allocation (due to fs reservation not
1293 		 * working out or fs corruption), the bitmap_bh is non-null
1294 		 * pointer and we have to release it before calling
1295 		 * read_block_bitmap().
1296 		 */
1297 		brelse(bitmap_bh);
1298 		bitmap_bh = read_block_bitmap(sb, group_no);
1299 		if (!bitmap_bh)
1300 			goto io_error;
1301 		grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1302 					bitmap_bh, grp_target_blk,
1303 					my_rsv, &num);
1304 		if (grp_alloc_blk >= 0)
1305 			goto allocated;
1306 	}
1307 
1308 	ngroups = EXT2_SB(sb)->s_groups_count;
1309 	smp_rmb();
1310 
1311 	/*
1312 	 * Now search the rest of the groups.  We assume that
1313 	 * group_no and gdp correctly point to the last group visited.
1314 	 */
1315 	for (bgi = 0; bgi < ngroups; bgi++) {
1316 		group_no++;
1317 		if (group_no >= ngroups)
1318 			group_no = 0;
1319 		gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1320 		if (!gdp)
1321 			goto io_error;
1322 
1323 		free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1324 		/*
1325 		 * skip this group (and avoid loading bitmap) if there
1326 		 * are no free blocks
1327 		 */
1328 		if (!free_blocks)
1329 			continue;
1330 		/*
1331 		 * skip this group if the number of
1332 		 * free blocks is less than half of the reservation
1333 		 * window size.
1334 		 */
1335 		if (my_rsv && (free_blocks <= (windowsz/2)))
1336 			continue;
1337 
1338 		brelse(bitmap_bh);
1339 		bitmap_bh = read_block_bitmap(sb, group_no);
1340 		if (!bitmap_bh)
1341 			goto io_error;
1342 		/*
1343 		 * try to allocate block(s) from this group, without a goal(-1).
1344 		 */
1345 		grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1346 					bitmap_bh, -1, my_rsv, &num);
1347 		if (grp_alloc_blk >= 0)
1348 			goto allocated;
1349 	}
1350 	/*
1351 	 * We may end up a bogus earlier ENOSPC error due to
1352 	 * filesystem is "full" of reservations, but
1353 	 * there maybe indeed free blocks available on disk
1354 	 * In this case, we just forget about the reservations
1355 	 * just do block allocation as without reservations.
1356 	 */
1357 	if (my_rsv) {
1358 		my_rsv = NULL;
1359 		windowsz = 0;
1360 		group_no = goal_group;
1361 		goto retry_alloc;
1362 	}
1363 	/* No space left on the device */
1364 	*errp = -ENOSPC;
1365 	goto out;
1366 
1367 allocated:
1368 
1369 	ext2_debug("using block group %d(%d)\n",
1370 			group_no, gdp->bg_free_blocks_count);
1371 
1372 	ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1373 
1374 	if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1375 	    in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1376 	    in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1377 		      EXT2_SB(sb)->s_itb_per_group) ||
1378 	    in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1379 		      EXT2_SB(sb)->s_itb_per_group)) {
1380 		ext2_error(sb, "ext2_new_blocks",
1381 			    "Allocating block in system zone - "
1382 			    "blocks from "E2FSBLK", length %lu",
1383 			    ret_block, num);
1384 		/*
1385 		 * ext2_try_to_allocate marked the blocks we allocated as in
1386 		 * use.  So we may want to selectively mark some of the blocks
1387 		 * as free
1388 		 */
1389 		num = *count;
1390 		goto retry_alloc;
1391 	}
1392 
1393 	performed_allocation = 1;
1394 
1395 	if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1396 		ext2_error(sb, "ext2_new_blocks",
1397 			    "block("E2FSBLK") >= blocks count(%d) - "
1398 			    "block_group = %d, es == %p ", ret_block,
1399 			le32_to_cpu(es->s_blocks_count), group_no, es);
1400 		goto out;
1401 	}
1402 
1403 	group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1404 	percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1405 
1406 	mark_buffer_dirty(bitmap_bh);
1407 	if (sb->s_flags & SB_SYNCHRONOUS)
1408 		sync_dirty_buffer(bitmap_bh);
1409 
1410 	*errp = 0;
1411 	brelse(bitmap_bh);
1412 	if (num < *count) {
1413 		dquot_free_block_nodirty(inode, *count-num);
1414 		mark_inode_dirty(inode);
1415 		*count = num;
1416 	}
1417 	return ret_block;
1418 
1419 io_error:
1420 	*errp = -EIO;
1421 out:
1422 	/*
1423 	 * Undo the block allocation
1424 	 */
1425 	if (!performed_allocation) {
1426 		dquot_free_block_nodirty(inode, *count);
1427 		mark_inode_dirty(inode);
1428 	}
1429 	brelse(bitmap_bh);
1430 	return 0;
1431 }
1432 
1433 ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1434 {
1435 	unsigned long count = 1;
1436 
1437 	return ext2_new_blocks(inode, goal, &count, errp);
1438 }
1439 
1440 #ifdef EXT2FS_DEBUG
1441 
1442 unsigned long ext2_count_free(struct buffer_head *map, unsigned int numchars)
1443 {
1444 	return numchars * BITS_PER_BYTE - memweight(map->b_data, numchars);
1445 }
1446 
1447 #endif  /*  EXT2FS_DEBUG  */
1448 
1449 unsigned long ext2_count_free_blocks (struct super_block * sb)
1450 {
1451 	struct ext2_group_desc * desc;
1452 	unsigned long desc_count = 0;
1453 	int i;
1454 #ifdef EXT2FS_DEBUG
1455 	unsigned long bitmap_count, x;
1456 	struct ext2_super_block *es;
1457 
1458 	es = EXT2_SB(sb)->s_es;
1459 	desc_count = 0;
1460 	bitmap_count = 0;
1461 	desc = NULL;
1462 	for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1463 		struct buffer_head *bitmap_bh;
1464 		desc = ext2_get_group_desc (sb, i, NULL);
1465 		if (!desc)
1466 			continue;
1467 		desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1468 		bitmap_bh = read_block_bitmap(sb, i);
1469 		if (!bitmap_bh)
1470 			continue;
1471 
1472 		x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1473 		printk ("group %d: stored = %d, counted = %lu\n",
1474 			i, le16_to_cpu(desc->bg_free_blocks_count), x);
1475 		bitmap_count += x;
1476 		brelse(bitmap_bh);
1477 	}
1478 	printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1479 		(long)le32_to_cpu(es->s_free_blocks_count),
1480 		desc_count, bitmap_count);
1481 	return bitmap_count;
1482 #else
1483         for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1484                 desc = ext2_get_group_desc (sb, i, NULL);
1485                 if (!desc)
1486                         continue;
1487                 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1488 	}
1489 	return desc_count;
1490 #endif
1491 }
1492 
1493 static inline int test_root(int a, int b)
1494 {
1495 	int num = b;
1496 
1497 	while (a > num)
1498 		num *= b;
1499 	return num == a;
1500 }
1501 
1502 static int ext2_group_sparse(int group)
1503 {
1504 	if (group <= 1)
1505 		return 1;
1506 	return (test_root(group, 3) || test_root(group, 5) ||
1507 		test_root(group, 7));
1508 }
1509 
1510 /**
1511  *	ext2_bg_has_super - number of blocks used by the superblock in group
1512  *	@sb: superblock for filesystem
1513  *	@group: group number to check
1514  *
1515  *	Return the number of blocks used by the superblock (primary or backup)
1516  *	in this group.  Currently this will be only 0 or 1.
1517  */
1518 int ext2_bg_has_super(struct super_block *sb, int group)
1519 {
1520 	if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1521 	    !ext2_group_sparse(group))
1522 		return 0;
1523 	return 1;
1524 }
1525 
1526 /**
1527  *	ext2_bg_num_gdb - number of blocks used by the group table in group
1528  *	@sb: superblock for filesystem
1529  *	@group: group number to check
1530  *
1531  *	Return the number of blocks used by the group descriptor table
1532  *	(primary or backup) in this group.  In the future there may be a
1533  *	different number of descriptor blocks in each group.
1534  */
1535 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1536 {
1537 	return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
1538 }
1539 
1540