projects/linux·blob master · refs

1539 lines · 45.1 KiB

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 ([email protected])
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * Enhanced block allocation by Stephen Tweedie ([email protected]), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller ([email protected]), 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
39struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
40 unsigned int block_group,
41 struct buffer_head ** bh)
42{
43 unsigned long group_desc;
44 unsigned long offset;
45 struct ext2_group_desc * desc;
46 struct ext2_sb_info *sbi = EXT2_SB(sb);
47
48 if (block_group >= sbi->s_groups_count) {
49 WARN(1, "block_group >= groups_count - "
50 "block_group = %d, groups_count = %lu",
51 block_group, sbi->s_groups_count);
52
53 return NULL;
54 }
55
56 group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
57 offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
58 if (!sbi->s_group_desc[group_desc]) {
59 WARN(1, "Group descriptor not loaded - "
60 "block_group = %d, group_desc = %lu, desc = %lu",
61 block_group, group_desc, offset);
62 return NULL;
63 }
64
65 desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
66 if (bh)
67 *bh = sbi->s_group_desc[group_desc];
68 return desc + offset;
69}
70
71static int ext2_valid_block_bitmap(struct super_block *sb,
72 struct ext2_group_desc *desc,
73 unsigned int block_group,
74 struct buffer_head *bh)
75{
76 ext2_grpblk_t offset;
77 ext2_grpblk_t next_zero_bit;
78 ext2_fsblk_t bitmap_blk;
79 ext2_fsblk_t group_first_block;
80 ext2_grpblk_t max_bit;
81
82 group_first_block = ext2_group_first_block_no(sb, block_group);
83 max_bit = ext2_group_last_block_no(sb, block_group) - group_first_block;
84
85 /* check whether block bitmap block number is set */
86 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
87 offset = bitmap_blk - group_first_block;
88 if (offset < 0 || offset > max_bit ||
89 !ext2_test_bit(offset, bh->b_data))
90 /* bad block bitmap */
91 goto err_out;
92
93 /* check whether the inode bitmap block number is set */
94 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
95 offset = bitmap_blk - group_first_block;
96 if (offset < 0 || offset > max_bit ||
97 !ext2_test_bit(offset, bh->b_data))
98 /* bad block bitmap */
99 goto err_out;
100
101 /* check whether the inode table block number is set */
102 bitmap_blk = le32_to_cpu(desc->bg_inode_table);
103 offset = bitmap_blk - group_first_block;
104 if (offset < 0 || offset > max_bit ||
105 offset + EXT2_SB(sb)->s_itb_per_group - 1 > max_bit)
106 goto err_out;
107 next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
108 offset + EXT2_SB(sb)->s_itb_per_group,
109 offset);
110 if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
111 /* good bitmap for inode tables */
112 return 1;
113
114err_out:
115 ext2_error(sb, __func__,
116 "Invalid block bitmap - "
117 "block_group = %d, block = %lu",
118 block_group, bitmap_blk);
119 return 0;
120}
121
122/*
123 * Read the bitmap for a given block_group,and validate the
124 * bits for block/inode/inode tables are set in the bitmaps
125 *
126 * Return buffer_head on success or NULL in case of failure.
127 */
128static struct buffer_head *
129read_block_bitmap(struct super_block *sb, unsigned int block_group)
130{
131 struct ext2_group_desc * desc;
132 struct buffer_head * bh = NULL;
133 ext2_fsblk_t bitmap_blk;
134 int ret;
135
136 desc = ext2_get_group_desc(sb, block_group, NULL);
137 if (!desc)
138 return NULL;
139 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
140 bh = sb_getblk(sb, bitmap_blk);
141 if (unlikely(!bh)) {
142 ext2_error(sb, __func__,
143 "Cannot read block bitmap - "
144 "block_group = %d, block_bitmap = %u",
145 block_group, le32_to_cpu(desc->bg_block_bitmap));
146 return NULL;
147 }
148 ret = bh_read(bh, 0);
149 if (ret > 0)
150 return bh;
151 if (ret < 0) {
152 brelse(bh);
153 ext2_error(sb, __func__,
154 "Cannot read block bitmap - "
155 "block_group = %d, block_bitmap = %u",
156 block_group, le32_to_cpu(desc->bg_block_bitmap));
157 return NULL;
158 }
159
160 ext2_valid_block_bitmap(sb, desc, block_group, bh);
161 /*
162 * file system mounted not to panic on error, continue with corrupt
163 * bitmap
164 */
165 return bh;
166}
167
168static void group_adjust_blocks(struct super_block *sb, int group_no,
169 struct ext2_group_desc *desc, struct buffer_head *bh, int count)
170{
171 if (count) {
172 struct ext2_sb_info *sbi = EXT2_SB(sb);
173 unsigned free_blocks;
174
175 spin_lock(sb_bgl_lock(sbi, group_no));
176 free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
177 desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
178 spin_unlock(sb_bgl_lock(sbi, group_no));
179 mark_buffer_dirty(bh);
180 }
181}
182
183/*
184 * The reservation window structure operations
185 * --------------------------------------------
186 * Operations include:
187 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
188 *
189 * We use a red-black tree to represent per-filesystem reservation
190 * windows.
191 *
192 */
193
194/**
195 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
196 * @root: root of per-filesystem reservation rb tree
197 * @verbose: verbose mode
198 * @fn: function which wishes to dump the reservation map
199 *
200 * If verbose is turned on, it will print the whole block reservation
201 * windows(start, end). Otherwise, it will only print out the "bad" windows,
202 * those windows that overlap with their immediate neighbors.
203 */
204#ifdef EXT2FS_DEBUG
205static void __rsv_window_dump(struct rb_root *root, int verbose,
206 const char *fn)
207{
208 struct rb_node *n;
209 struct ext2_reserve_window_node *rsv, *prev;
210 int bad;
211
212restart:
213 n = rb_first(root);
214 bad = 0;
215 prev = NULL;
216
217 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
218 while (n) {
219 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
220 if (verbose)
221 printk("reservation window 0x%p "
222 "start: %lu, end: %lu\n",
223 rsv, rsv->rsv_start, rsv->rsv_end);
224 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
225 printk("Bad reservation %p (start >= end)\n",
226 rsv);
227 bad = 1;
228 }
229 if (prev && prev->rsv_end >= rsv->rsv_start) {
230 printk("Bad reservation %p (prev->end >= start)\n",
231 rsv);
232 bad = 1;
233 }
234 if (bad) {
235 if (!verbose) {
236 printk("Restarting reservation walk in verbose mode\n");
237 verbose = 1;
238 goto restart;
239 }
240 }
241 n = rb_next(n);
242 prev = rsv;
243 }
244 printk("Window map complete.\n");
245 BUG_ON(bad);
246}
247#define rsv_window_dump(root, verbose) \
248 __rsv_window_dump((root), (verbose), __func__)
249#else
250#define rsv_window_dump(root, verbose) do {} while (0)
251#endif /* EXT2FS_DEBUG */
252
253/**
254 * goal_in_my_reservation()
255 * @rsv: inode's reservation window
256 * @grp_goal: given goal block relative to the allocation block group
257 * @group: the current allocation block group
258 * @sb: filesystem super block
259 *
260 * Test if the given goal block (group relative) is within the file's
261 * own block reservation window range.
262 *
263 * If the reservation window is outside the goal allocation group, return 0;
264 * grp_goal (given goal block) could be -1, which means no specific
265 * goal block. In this case, always return 1.
266 * If the goal block is within the reservation window, return 1;
267 * otherwise, return 0;
268 */
269static int
270goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
271 unsigned int group, struct super_block * sb)
272{
273 ext2_fsblk_t group_first_block, group_last_block;
274
275 group_first_block = ext2_group_first_block_no(sb, group);
276 group_last_block = ext2_group_last_block_no(sb, group);
277
278 if ((rsv->_rsv_start > group_last_block) ||
279 (rsv->_rsv_end < group_first_block))
280 return 0;
281 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
282 || (grp_goal + group_first_block > rsv->_rsv_end)))
283 return 0;
284 return 1;
285}
286
287/**
288 * search_reserve_window()
289 * @root: root of reservation tree
290 * @goal: target allocation block
291 *
292 * Find the reserved window which includes the goal, or the previous one
293 * if the goal is not in any window.
294 * Returns NULL if there are no windows or if all windows start after the goal.
295 */
296static struct ext2_reserve_window_node *
297search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
298{
299 struct rb_node *n = root->rb_node;
300 struct ext2_reserve_window_node *rsv;
301
302 if (!n)
303 return NULL;
304
305 do {
306 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
307
308 if (goal < rsv->rsv_start)
309 n = n->rb_left;
310 else if (goal > rsv->rsv_end)
311 n = n->rb_right;
312 else
313 return rsv;
314 } while (n);
315 /*
316 * We've fallen off the end of the tree: the goal wasn't inside
317 * any particular node. OK, the previous node must be to one
318 * side of the interval containing the goal. If it's the RHS,
319 * we need to back up one.
320 */
321 if (rsv->rsv_start > goal) {
322 n = rb_prev(&rsv->rsv_node);
323 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
324 }
325 return rsv;
326}
327
328/*
329 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
330 * @sb: super block
331 * @rsv: reservation window to add
332 *
333 * Must be called with rsv_lock held.
334 */
335void ext2_rsv_window_add(struct super_block *sb,
336 struct ext2_reserve_window_node *rsv)
337 __must_hold(&EXT2_SB(sb)->s_rsv_window_lock)
338{
339 struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
340 struct rb_node *node = &rsv->rsv_node;
341 ext2_fsblk_t start = rsv->rsv_start;
342
343 struct rb_node ** p = &root->rb_node;
344 struct rb_node * parent = NULL;
345 struct ext2_reserve_window_node *this;
346
347 while (*p)
348 {
349 parent = *p;
350 this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
351
352 if (start < this->rsv_start)
353 p = &(*p)->rb_left;
354 else if (start > this->rsv_end)
355 p = &(*p)->rb_right;
356 else {
357 rsv_window_dump(root, 1);
358 BUG();
359 }
360 }
361
362 rb_link_node(node, parent, p);
363 rb_insert_color(node, root);
364}
365
366/**
367 * rsv_window_remove() -- unlink a window from the reservation rb tree
368 * @sb: super block
369 * @rsv: reservation window to remove
370 *
371 * Mark the block reservation window as not allocated, and unlink it
372 * from the filesystem reservation window rb tree. Must be called with
373 * rsv_lock held.
374 */
375static void rsv_window_remove(struct super_block *sb,
376 struct ext2_reserve_window_node *rsv)
377 __must_hold(&EXT2_SB(sb)->s_rsv_window_lock)
378{
379 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
380 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
381 rsv->rsv_alloc_hit = 0;
382 rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
383}
384
385/*
386 * rsv_is_empty() -- Check if the reservation window is allocated.
387 * @rsv: given reservation window to check
388 *
389 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
390 */
391static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
392{
393 /* a valid reservation end block could not be 0 */
394 return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
395}
396
397/**
398 * ext2_init_block_alloc_info()
399 * @inode: file inode structure
400 *
401 * Allocate and initialize the reservation window structure, and
402 * link the window to the ext2 inode structure at last
403 *
404 * The reservation window structure is only dynamically allocated
405 * and linked to ext2 inode the first time the open file
406 * needs a new block. So, before every ext2_new_block(s) call, for
407 * regular files, we should check whether the reservation window
408 * structure exists or not. In the latter case, this function is called.
409 * Fail to do so will result in block reservation being turned off for that
410 * open file.
411 *
412 * This function is called from ext2_get_blocks_handle(), also called
413 * when setting the reservation window size through ioctl before the file
414 * is open for write (needs block allocation).
415 *
416 * Needs truncate_mutex protection prior to calling this function.
417 */
418void ext2_init_block_alloc_info(struct inode *inode)
419 __must_hold(&EXT2_I(inode)->truncate_mutex)
420{
421 struct ext2_inode_info *ei = EXT2_I(inode);
422 struct ext2_block_alloc_info *block_i;
423 struct super_block *sb = inode->i_sb;
424
425 block_i = kmalloc_obj(*block_i);
426 if (block_i) {
427 struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
428
429 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
430 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
431
432 /*
433 * if filesystem is mounted with NORESERVATION, the goal
434 * reservation window size is set to zero to indicate
435 * block reservation is off
436 */
437 if (!test_opt(sb, RESERVATION))
438 rsv->rsv_goal_size = 0;
439 else
440 rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
441 rsv->rsv_alloc_hit = 0;
442 block_i->last_alloc_logical_block = 0;
443 block_i->last_alloc_physical_block = 0;
444 }
445 ei->i_block_alloc_info = block_i;
446}
447
448/**
449 * ext2_discard_reservation()
450 * @inode: inode
451 *
452 * Discard(free) block reservation window on last file close, or truncate
453 * or at last iput().
454 *
455 * It is being called in three cases:
456 * ext2_release_file(): last writer closes the file
457 * ext2_clear_inode(): last iput(), when nobody links to this file.
458 * ext2_truncate(): when the block indirect map is about to change.
459 */
460void ext2_discard_reservation(struct inode *inode)
461{
462 struct ext2_inode_info *ei = EXT2_I(inode);
463 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
464 struct ext2_reserve_window_node *rsv;
465 spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
466
467 if (!block_i)
468 return;
469
470 rsv = &block_i->rsv_window_node;
471 if (!rsv_is_empty(&rsv->rsv_window)) {
472 spin_lock(rsv_lock);
473 if (!rsv_is_empty(&rsv->rsv_window))
474 rsv_window_remove(inode->i_sb, rsv);
475 spin_unlock(rsv_lock);
476 }
477}
478
479/**
480 * ext2_free_blocks() -- Free given blocks and update quota and i_blocks
481 * @inode: inode
482 * @block: start physical block to free
483 * @count: number of blocks to free
484 */
485void ext2_free_blocks(struct inode * inode, ext2_fsblk_t block,
486 unsigned long count)
487{
488 struct buffer_head *bitmap_bh = NULL;
489 struct buffer_head * bh2;
490 unsigned long block_group;
491 unsigned long bit;
492 unsigned long i;
493 unsigned long overflow;
494 struct super_block * sb = inode->i_sb;
495 struct ext2_sb_info * sbi = EXT2_SB(sb);
496 struct ext2_group_desc * desc;
497 struct ext2_super_block * es = sbi->s_es;
498 unsigned freed = 0, group_freed;
499
500 if (!ext2_data_block_valid(sbi, block, count)) {
501 ext2_error (sb, "ext2_free_blocks",
502 "Freeing blocks not in datazone - "
503 "block = %lu, count = %lu", block, count);
504 goto error_return;
505 }
506
507 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
508
509do_more:
510 overflow = 0;
511 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
512 EXT2_BLOCKS_PER_GROUP(sb);
513 bit = (block - le32_to_cpu(es->s_first_data_block)) %
514 EXT2_BLOCKS_PER_GROUP(sb);
515 /*
516 * Check to see if we are freeing blocks across a group
517 * boundary.
518 */
519 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
520 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
521 count -= overflow;
522 }
523 brelse(bitmap_bh);
524 bitmap_bh = read_block_bitmap(sb, block_group);
525 if (!bitmap_bh)
526 goto error_return;
527
528 desc = ext2_get_group_desc (sb, block_group, &bh2);
529 if (!desc)
530 goto error_return;
531
532 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
533 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
534 in_range (block, le32_to_cpu(desc->bg_inode_table),
535 sbi->s_itb_per_group) ||
536 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
537 sbi->s_itb_per_group)) {
538 ext2_error (sb, "ext2_free_blocks",
539 "Freeing blocks in system zones - "
540 "Block = %lu, count = %lu",
541 block, count);
542 goto error_return;
543 }
544
545 for (i = 0, group_freed = 0; i < count; i++) {
546 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
547 bit + i, bitmap_bh->b_data)) {
548 ext2_error(sb, __func__,
549 "bit already cleared for block %lu", block + i);
550 } else {
551 group_freed++;
552 }
553 }
554
555 mark_buffer_dirty(bitmap_bh);
556 if (sb->s_flags & SB_SYNCHRONOUS)
557 sync_dirty_buffer(bitmap_bh);
558
559 group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
560 freed += group_freed;
561
562 if (overflow) {
563 block += count;
564 count = overflow;
565 goto do_more;
566 }
567error_return:
568 brelse(bitmap_bh);
569 if (freed) {
570 percpu_counter_add(&sbi->s_freeblocks_counter, freed);
571 dquot_free_block_nodirty(inode, freed);
572 mark_inode_dirty(inode);
573 }
574}
575
576/**
577 * bitmap_search_next_usable_block()
578 * @start: the starting block (group relative) of the search
579 * @bh: bufferhead contains the block group bitmap
580 * @maxblocks: the ending block (group relative) of the reservation
581 *
582 * The bitmap search --- search forward through the actual bitmap on disk until
583 * we find a bit free.
584 */
585static ext2_grpblk_t
586bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
587 ext2_grpblk_t maxblocks)
588{
589 ext2_grpblk_t next;
590
591 next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
592 if (next >= maxblocks)
593 return -1;
594 return next;
595}
596
597/**
598 * find_next_usable_block()
599 * @start: the starting block (group relative) to find next
600 * allocatable block in bitmap.
601 * @bh: bufferhead contains the block group bitmap
602 * @maxblocks: the ending block (group relative) for the search
603 *
604 * Find an allocatable block in a bitmap. We perform the "most
605 * appropriate allocation" algorithm of looking for a free block near
606 * the initial goal; then for a free byte somewhere in the bitmap;
607 * then for any free bit in the bitmap.
608 */
609static ext2_grpblk_t
610find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
611{
612 ext2_grpblk_t here, next;
613 char *p, *r;
614
615 if (start > 0) {
616 /*
617 * The goal was occupied; search forward for a free
618 * block within the next XX blocks.
619 *
620 * end_goal is more or less random, but it has to be
621 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
622 * next 64-bit boundary is simple..
623 */
624 ext2_grpblk_t end_goal = (start + 63) & ~63;
625 if (end_goal > maxblocks)
626 end_goal = maxblocks;
627 here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
628 if (here < end_goal)
629 return here;
630 ext2_debug("Bit not found near goal\n");
631 }
632
633 here = start;
634 if (here < 0)
635 here = 0;
636
637 p = ((char *)bh->b_data) + (here >> 3);
638 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
639 next = (r - ((char *)bh->b_data)) << 3;
640
641 if (next < maxblocks && next >= here)
642 return next;
643
644 here = bitmap_search_next_usable_block(here, bh, maxblocks);
645 return here;
646}
647
648/**
649 * ext2_try_to_allocate()
650 * @sb: superblock
651 * @group: given allocation block group
652 * @bitmap_bh: bufferhead holds the block bitmap
653 * @grp_goal: given target block within the group
654 * @count: target number of blocks to allocate
655 * @my_rsv: reservation window
656 *
657 * Attempt to allocate blocks within a give range. Set the range of allocation
658 * first, then find the first free bit(s) from the bitmap (within the range),
659 * and at last, allocate the blocks by claiming the found free bit as allocated.
660 *
661 * To set the range of this allocation:
662 * if there is a reservation window, only try to allocate block(s)
663 * from the file's own reservation window;
664 * Otherwise, the allocation range starts from the give goal block,
665 * ends at the block group's last block.
666 *
667 * If we failed to allocate the desired block then we may end up crossing to a
668 * new bitmap.
669 */
670static int
671ext2_try_to_allocate(struct super_block *sb, int group,
672 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
673 unsigned long *count,
674 struct ext2_reserve_window *my_rsv)
675{
676 ext2_fsblk_t group_first_block = ext2_group_first_block_no(sb, group);
677 ext2_fsblk_t group_last_block = ext2_group_last_block_no(sb, group);
678 ext2_grpblk_t start, end;
679 unsigned long num = 0;
680
681 start = 0;
682 end = group_last_block - group_first_block + 1;
683 /* we do allocation within the reservation window if we have a window */
684 if (my_rsv) {
685 if (my_rsv->_rsv_start >= group_first_block)
686 start = my_rsv->_rsv_start - group_first_block;
687 if (my_rsv->_rsv_end < group_last_block)
688 end = my_rsv->_rsv_end - group_first_block + 1;
689 if (grp_goal < start || grp_goal >= end)
690 grp_goal = -1;
691 }
692 BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
693
694 if (grp_goal < 0) {
695 grp_goal = find_next_usable_block(start, bitmap_bh, end);
696 if (grp_goal < 0)
697 goto fail_access;
698 if (!my_rsv) {
699 int i;
700
701 for (i = 0; i < 7 && grp_goal > start &&
702 !ext2_test_bit(grp_goal - 1,
703 bitmap_bh->b_data);
704 i++, grp_goal--)
705 ;
706 }
707 }
708
709 for (; num < *count && grp_goal < end; grp_goal++) {
710 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
711 grp_goal, bitmap_bh->b_data)) {
712 if (num == 0)
713 continue;
714 break;
715 }
716 num++;
717 }
718
719 if (num == 0)
720 goto fail_access;
721
722 *count = num;
723 return grp_goal - num;
724fail_access:
725 return -1;
726}
727
728/**
729 * find_next_reservable_window - Find a reservable space within the given range.
730 * @search_head: The list to search.
731 * @my_rsv: The reservation we're currently using.
732 * @sb: The super block.
733 * @start_block: The first block we consider to start the real search from
734 * @last_block: The maximum block number that our goal reservable space
735 * could start from.
736 *
737 * It does not allocate the reservation window: alloc_new_reservation()
738 * will do the work later.
739 *
740 * We search the given range, rather than the whole reservation double
741 * linked list, (start_block, last_block) to find a free region that is
742 * of my size and has not been reserved.
743 *
744 * @search_head is not necessarily the list head of the whole filesystem.
745 * We have both head and @start_block to assist the search for the
746 * reservable space. The list starts from head, but we will shift to
747 * the place where start_block is, then start from there, when looking
748 * for a reservable space.
749 *
750 * @last_block is normally the last block in this group. The search will end
751 * when we found the start of next possible reservable space is out
752 * of this boundary. This could handle the cross boundary reservation
753 * window request.
754 *
755 * Return: -1 if we could not find a range of sufficient size. If we could,
756 * return 0 and fill in @my_rsv with the range information.
757 */
758static int find_next_reservable_window(
759 struct ext2_reserve_window_node *search_head,
760 struct ext2_reserve_window_node *my_rsv,
761 struct super_block * sb,
762 ext2_fsblk_t start_block,
763 ext2_fsblk_t last_block)
764 __must_hold(&EXT2_SB(sb)->s_rsv_window_lock)
765{
766 struct rb_node *next;
767 struct ext2_reserve_window_node *rsv, *prev;
768 ext2_fsblk_t cur;
769 int size = my_rsv->rsv_goal_size;
770
771 /* TODO: make the start of the reservation window byte-aligned */
772 /* cur = *start_block & ~7;*/
773 cur = start_block;
774 rsv = search_head;
775 if (!rsv)
776 return -1;
777
778 while (1) {
779 if (cur <= rsv->rsv_end)
780 cur = rsv->rsv_end + 1;
781
782 /* TODO?
783 * in the case we could not find a reservable space
784 * that is what is expected, during the re-search, we could
785 * remember what's the largest reservable space we could have
786 * and return that one.
787 *
788 * For now it will fail if we could not find the reservable
789 * space with expected-size (or more)...
790 */
791 if (cur > last_block)
792 return -1; /* fail */
793
794 prev = rsv;
795 next = rb_next(&rsv->rsv_node);
796 rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
797
798 /*
799 * Reached the last reservation, we can just append to the
800 * previous one.
801 */
802 if (!next)
803 break;
804
805 if (cur + size <= rsv->rsv_start) {
806 /*
807 * Found a reserveable space big enough. We could
808 * have a reservation across the group boundary here
809 */
810 break;
811 }
812 }
813 /*
814 * we come here either :
815 * when we reach the end of the whole list,
816 * and there is empty reservable space after last entry in the list.
817 * append it to the end of the list.
818 *
819 * or we found one reservable space in the middle of the list,
820 * return the reservation window that we could append to.
821 * succeed.
822 */
823
824 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
825 rsv_window_remove(sb, my_rsv);
826
827 /*
828 * Let's book the whole available window for now. We will check the
829 * disk bitmap later and then, if there are free blocks then we adjust
830 * the window size if it's larger than requested.
831 * Otherwise, we will remove this node from the tree next time
832 * call find_next_reservable_window.
833 */
834 my_rsv->rsv_start = cur;
835 my_rsv->rsv_end = cur + size - 1;
836 my_rsv->rsv_alloc_hit = 0;
837
838 if (prev != my_rsv)
839 ext2_rsv_window_add(sb, my_rsv);
840
841 return 0;
842}
843
844/**
845 * alloc_new_reservation - Allocate a new reservation window.
846 * @my_rsv: The reservation we're currently using.
847 * @grp_goal: The goal block relative to the start of the group.
848 * @sb: The super block.
849 * @group: The group we are trying to allocate in.
850 * @bitmap_bh: The block group block bitmap.
851 *
852 * To make a new reservation, we search part of the filesystem reservation
853 * list (the list inside the group). We try to allocate a new
854 * reservation window near @grp_goal, or the beginning of the
855 * group, if @grp_goal is negative.
856 *
857 * We first find a reservable space after the goal, then from there,
858 * we check the bitmap for the first free block after it. If there is
859 * no free block until the end of group, then the whole group is full,
860 * we failed. Otherwise, check if the free block is inside the expected
861 * reservable space, if so, we succeed.
862 *
863 * If the first free block is outside the reservable space, then start
864 * from the first free block, we search for next available space, and
865 * go on.
866 *
867 * on succeed, a new reservation will be found and inserted into the
868 * list. It contains at least one free block, and it does not overlap
869 * with other reservation windows.
870 *
871 * Return: 0 on success, -1 if we failed to find a reservation window
872 * in this group
873 */
874static 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 */
944retry:
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 */
1016static 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 */
1067static ext2_grpblk_t
1068ext2_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 ext2_error(sb, __func__,
1137 "Reservation out of group %u range goal %d fsb[%lu,%lu] rsv[%lu, %lu]",
1138 group, grp_goal, group_first_block,
1139 group_last_block, my_rsv->rsv_start,
1140 my_rsv->rsv_end);
1141 rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1142 return -1;
1143 }
1144 ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1145 &num, &my_rsv->rsv_window);
1146 if (ret >= 0) {
1147 my_rsv->rsv_alloc_hit += num;
1148 *count = num;
1149 break; /* succeed */
1150 }
1151 num = *count;
1152 }
1153 return ret;
1154}
1155
1156/**
1157 * ext2_has_free_blocks()
1158 * @sbi: in-core super block structure.
1159 *
1160 * Check if filesystem has at least 1 free block available for allocation.
1161 */
1162static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1163{
1164 ext2_fsblk_t free_blocks, root_blocks;
1165
1166 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1167 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1168 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1169 !uid_eq(sbi->s_resuid, current_fsuid()) &&
1170 (gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1171 !in_group_p (sbi->s_resgid))) {
1172 return 0;
1173 }
1174 return 1;
1175}
1176
1177/*
1178 * Returns 1 if the passed-in block region is valid; 0 if some part overlaps
1179 * with filesystem metadata blocks.
1180 */
1181int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk,
1182 unsigned int count)
1183{
1184 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
1185 (start_blk + count - 1 < start_blk) ||
1186 (start_blk + count - 1 >= le32_to_cpu(sbi->s_es->s_blocks_count)))
1187 return 0;
1188
1189 /* Ensure we do not step over superblock */
1190 if ((start_blk <= sbi->s_sb_block) &&
1191 (start_blk + count - 1 >= sbi->s_sb_block))
1192 return 0;
1193
1194 return 1;
1195}
1196
1197/*
1198 * ext2_new_blocks() -- core block(s) allocation function
1199 * @inode: file inode
1200 * @goal: given target block(filesystem wide)
1201 * @count: target number of blocks to allocate
1202 * @errp: error code
1203 * @flags: allocate flags
1204 *
1205 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
1206 * free, or there is a free block within 32 blocks of the goal, that block
1207 * is allocated. Otherwise a forward search is made for a free block; within
1208 * each block group the search first looks for an entire free byte in the block
1209 * bitmap, and then for any free bit if that fails.
1210 * This function also updates quota and i_blocks field.
1211 */
1212ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1213 unsigned long *count, int *errp, unsigned int flags)
1214{
1215 struct buffer_head *bitmap_bh = NULL;
1216 struct buffer_head *gdp_bh;
1217 int group_no;
1218 int goal_group;
1219 ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1220 ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1221 ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
1222 int bgi; /* blockgroup iteration index */
1223 int performed_allocation = 0;
1224 ext2_grpblk_t free_blocks; /* number of free blocks in a group */
1225 struct super_block *sb;
1226 struct ext2_group_desc *gdp;
1227 struct ext2_super_block *es;
1228 struct ext2_sb_info *sbi;
1229 struct ext2_reserve_window_node *my_rsv = NULL;
1230 struct ext2_block_alloc_info *block_i;
1231 unsigned short windowsz = 0;
1232 unsigned long ngroups;
1233 unsigned long num = *count;
1234 int ret;
1235
1236 *errp = -ENOSPC;
1237 sb = inode->i_sb;
1238
1239 /*
1240 * Check quota for allocation of this block.
1241 */
1242 ret = dquot_alloc_block(inode, num);
1243 if (ret) {
1244 *errp = ret;
1245 return 0;
1246 }
1247
1248 sbi = EXT2_SB(sb);
1249 es = EXT2_SB(sb)->s_es;
1250 ext2_debug("goal=%lu.\n", goal);
1251 /*
1252 * Allocate a block from reservation only when the filesystem is
1253 * mounted with reservation(default,-o reservation), and it's a regular
1254 * file, and the desired window size is greater than 0 (One could use
1255 * ioctl command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn
1256 * off reservation on that particular file). Also do not use the
1257 * reservation window if the caller asked us not to do it.
1258 */
1259 block_i = EXT2_I(inode)->i_block_alloc_info;
1260 if (!(flags & EXT2_ALLOC_NORESERVE) && block_i) {
1261 windowsz = block_i->rsv_window_node.rsv_goal_size;
1262 if (windowsz > 0)
1263 my_rsv = &block_i->rsv_window_node;
1264 }
1265
1266 if (!ext2_has_free_blocks(sbi)) {
1267 *errp = -ENOSPC;
1268 goto out;
1269 }
1270
1271 /*
1272 * First, test whether the goal block is free.
1273 */
1274 if (goal < le32_to_cpu(es->s_first_data_block) ||
1275 goal >= le32_to_cpu(es->s_blocks_count))
1276 goal = le32_to_cpu(es->s_first_data_block);
1277 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1278 EXT2_BLOCKS_PER_GROUP(sb);
1279 goal_group = group_no;
1280retry_alloc:
1281 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1282 if (!gdp)
1283 goto io_error;
1284
1285 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1286 /*
1287 * if there is not enough free blocks to make a new resevation
1288 * turn off reservation for this allocation
1289 */
1290 if (my_rsv && (free_blocks < windowsz)
1291 && (free_blocks > 0)
1292 && (rsv_is_empty(&my_rsv->rsv_window)))
1293 my_rsv = NULL;
1294
1295 if (free_blocks > 0) {
1296 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1297 EXT2_BLOCKS_PER_GROUP(sb));
1298 /*
1299 * In case we retry allocation (due to fs reservation not
1300 * working out or fs corruption), the bitmap_bh is non-null
1301 * pointer and we have to release it before calling
1302 * read_block_bitmap().
1303 */
1304 brelse(bitmap_bh);
1305 bitmap_bh = read_block_bitmap(sb, group_no);
1306 if (!bitmap_bh)
1307 goto io_error;
1308 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1309 bitmap_bh, grp_target_blk,
1310 my_rsv, &num);
1311 if (grp_alloc_blk >= 0)
1312 goto allocated;
1313 }
1314
1315 ngroups = EXT2_SB(sb)->s_groups_count;
1316 smp_rmb();
1317
1318 /*
1319 * Now search the rest of the groups. We assume that
1320 * group_no and gdp correctly point to the last group visited.
1321 */
1322 for (bgi = 0; bgi < ngroups; bgi++) {
1323 group_no++;
1324 if (group_no >= ngroups)
1325 group_no = 0;
1326 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1327 if (!gdp)
1328 goto io_error;
1329
1330 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1331 /*
1332 * skip this group (and avoid loading bitmap) if there
1333 * are no free blocks
1334 */
1335 if (!free_blocks)
1336 continue;
1337 /*
1338 * skip this group if the number of
1339 * free blocks is less than half of the reservation
1340 * window size.
1341 */
1342 if (my_rsv && (free_blocks <= (windowsz/2)))
1343 continue;
1344
1345 brelse(bitmap_bh);
1346 bitmap_bh = read_block_bitmap(sb, group_no);
1347 if (!bitmap_bh)
1348 goto io_error;
1349 /*
1350 * try to allocate block(s) from this group, without a goal(-1).
1351 */
1352 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1353 bitmap_bh, -1, my_rsv, &num);
1354 if (grp_alloc_blk >= 0)
1355 goto allocated;
1356 }
1357 /*
1358 * We may end up a bogus earlier ENOSPC error due to
1359 * filesystem is "full" of reservations, but
1360 * there maybe indeed free blocks available on disk
1361 * In this case, we just forget about the reservations
1362 * just do block allocation as without reservations.
1363 */
1364 if (my_rsv) {
1365 my_rsv = NULL;
1366 windowsz = 0;
1367 group_no = goal_group;
1368 goto retry_alloc;
1369 }
1370 /* No space left on the device */
1371 *errp = -ENOSPC;
1372 goto out;
1373
1374allocated:
1375
1376 ext2_debug("using block group %d(%d)\n",
1377 group_no, gdp->bg_free_blocks_count);
1378
1379 ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1380
1381 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1382 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1383 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1384 EXT2_SB(sb)->s_itb_per_group) ||
1385 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1386 EXT2_SB(sb)->s_itb_per_group)) {
1387 ext2_error(sb, "ext2_new_blocks",
1388 "Allocating block in system zone - "
1389 "blocks from "E2FSBLK", length %lu",
1390 ret_block, num);
1391 /*
1392 * ext2_try_to_allocate marked the blocks we allocated as in
1393 * use. So we may want to selectively mark some of the blocks
1394 * as free
1395 */
1396 num = *count;
1397 goto retry_alloc;
1398 }
1399
1400 performed_allocation = 1;
1401
1402 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1403 ext2_error(sb, "ext2_new_blocks",
1404 "block("E2FSBLK") >= blocks count(%d) - "
1405 "block_group = %d, es == %p ", ret_block,
1406 le32_to_cpu(es->s_blocks_count), group_no, es);
1407 goto out;
1408 }
1409
1410 group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1411 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1412
1413 mark_buffer_dirty(bitmap_bh);
1414 if (sb->s_flags & SB_SYNCHRONOUS)
1415 sync_dirty_buffer(bitmap_bh);
1416
1417 *errp = 0;
1418 brelse(bitmap_bh);
1419 if (num < *count) {
1420 dquot_free_block_nodirty(inode, *count-num);
1421 mark_inode_dirty(inode);
1422 *count = num;
1423 }
1424 return ret_block;
1425
1426io_error:
1427 *errp = -EIO;
1428out:
1429 /*
1430 * Undo the block allocation
1431 */
1432 if (!performed_allocation) {
1433 dquot_free_block_nodirty(inode, *count);
1434 mark_inode_dirty(inode);
1435 }
1436 brelse(bitmap_bh);
1437 return 0;
1438}
1439
1440#ifdef EXT2FS_DEBUG
1441
1442unsigned 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
1449unsigned 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
1493static 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
1502static 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 */
1518int 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 */
1535unsigned 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