blob: e7b2bafa1dd91cd69f8caf4597654a6355c3a666 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext2/balloc.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
10 * Big-endian to little-endian byte-swapping/bitmaps by
11 * David S. Miller (davem@caip.rutgers.edu), 1995
12 */
13
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include "ext2.h"
15#include <linux/quotaops.h>
16#include <linux/sched.h>
17#include <linux/buffer_head.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019
20/*
21 * balloc.c contains the blocks allocation and deallocation routines
22 */
23
24/*
25 * The free blocks are managed by bitmaps. A file system contains several
26 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
27 * block for inodes, N blocks for the inode table and data blocks.
28 *
29 * The file system contains group descriptors which are located after the
30 * super block. Each descriptor contains the number of the bitmap block and
31 * the free blocks count in the block. The descriptors are loaded in memory
Aneesh Kumar K.Ve6274322007-02-20 13:57:58 -080032 * when a file system is mounted (see ext2_fill_super).
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 */
34
35
36#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
37
38struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
39 unsigned int block_group,
40 struct buffer_head ** bh)
41{
42 unsigned long group_desc;
43 unsigned long offset;
44 struct ext2_group_desc * desc;
45 struct ext2_sb_info *sbi = EXT2_SB(sb);
46
47 if (block_group >= sbi->s_groups_count) {
48 ext2_error (sb, "ext2_get_group_desc",
49 "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 ext2_error (sb, "ext2_get_group_desc",
60 "Group descriptor not loaded - "
61 "block_group = %d, group_desc = %lu, desc = %lu",
62 block_group, group_desc, offset);
63 return NULL;
64 }
65
66 desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
67 if (bh)
68 *bh = sbi->s_group_desc[group_desc];
69 return desc + offset;
70}
71
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -080072static int ext2_valid_block_bitmap(struct super_block *sb,
73 struct ext2_group_desc *desc,
74 unsigned int block_group,
75 struct buffer_head *bh)
76{
77 ext2_grpblk_t offset;
78 ext2_grpblk_t next_zero_bit;
79 ext2_fsblk_t bitmap_blk;
80 ext2_fsblk_t group_first_block;
81
82 group_first_block = ext2_group_first_block_no(sb, block_group);
83
84 /* check whether block bitmap block number is set */
85 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
86 offset = bitmap_blk - group_first_block;
87 if (!ext2_test_bit(offset, bh->b_data))
88 /* bad block bitmap */
89 goto err_out;
90
91 /* check whether the inode bitmap block number is set */
92 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
93 offset = bitmap_blk - group_first_block;
94 if (!ext2_test_bit(offset, bh->b_data))
95 /* bad block bitmap */
96 goto err_out;
97
98 /* check whether the inode table block number is set */
99 bitmap_blk = le32_to_cpu(desc->bg_inode_table);
100 offset = bitmap_blk - group_first_block;
101 next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
102 offset + EXT2_SB(sb)->s_itb_per_group,
103 offset);
104 if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
105 /* good bitmap for inode tables */
106 return 1;
107
108err_out:
109 ext2_error(sb, __FUNCTION__,
110 "Invalid block bitmap - "
111 "block_group = %d, block = %lu",
112 block_group, bitmap_blk);
113 return 0;
114}
115
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116/*
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800117 * Read the bitmap for a given block_group,and validate the
118 * bits for block/inode/inode tables are set in the bitmaps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 *
120 * Return buffer_head on success or NULL in case of failure.
121 */
122static struct buffer_head *
123read_block_bitmap(struct super_block *sb, unsigned int block_group)
124{
125 struct ext2_group_desc * desc;
126 struct buffer_head * bh = NULL;
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800127 ext2_fsblk_t bitmap_blk;
128
129 desc = ext2_get_group_desc(sb, block_group, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 if (!desc)
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800131 return NULL;
132 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
133 bh = sb_getblk(sb, bitmap_blk);
134 if (unlikely(!bh)) {
135 ext2_error(sb, __FUNCTION__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 "Cannot read block bitmap - "
137 "block_group = %d, block_bitmap = %u",
138 block_group, le32_to_cpu(desc->bg_block_bitmap));
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800139 return NULL;
140 }
141 if (likely(bh_uptodate_or_lock(bh)))
142 return bh;
143
144 if (bh_submit_read(bh) < 0) {
145 brelse(bh);
146 ext2_error(sb, __FUNCTION__,
147 "Cannot read block bitmap - "
148 "block_group = %d, block_bitmap = %u",
149 block_group, le32_to_cpu(desc->bg_block_bitmap));
150 return NULL;
151 }
152 if (!ext2_valid_block_bitmap(sb, desc, block_group, bh)) {
153 brelse(bh);
154 return NULL;
155 }
156
Linus Torvalds0b832a42007-11-13 08:07:31 -0800157 return bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158}
159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160static void release_blocks(struct super_block *sb, int count)
161{
162 if (count) {
163 struct ext2_sb_info *sbi = EXT2_SB(sb);
164
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700165 percpu_counter_add(&sbi->s_freeblocks_counter, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 sb->s_dirt = 1;
167 }
168}
169
Martin J. Bligha686cd82007-10-16 23:30:46 -0700170static void group_adjust_blocks(struct super_block *sb, int group_no,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 struct ext2_group_desc *desc, struct buffer_head *bh, int count)
172{
173 if (count) {
174 struct ext2_sb_info *sbi = EXT2_SB(sb);
175 unsigned free_blocks;
176
177 spin_lock(sb_bgl_lock(sbi, group_no));
178 free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
179 desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
180 spin_unlock(sb_bgl_lock(sbi, group_no));
181 sb->s_dirt = 1;
182 mark_buffer_dirty(bh);
183 }
184}
185
Martin J. Bligha686cd82007-10-16 23:30:46 -0700186/*
187 * The reservation window structure operations
188 * --------------------------------------------
189 * Operations include:
190 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
191 *
192 * We use a red-black tree to represent per-filesystem reservation
193 * windows.
194 *
195 */
196
197/**
198 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
199 * @rb_root: root of per-filesystem reservation rb tree
200 * @verbose: verbose mode
201 * @fn: function which wishes to dump the reservation map
202 *
203 * If verbose is turned on, it will print the whole block reservation
204 * windows(start, end). Otherwise, it will only print out the "bad" windows,
205 * those windows that overlap with their immediate neighbors.
206 */
207#if 1
208static void __rsv_window_dump(struct rb_root *root, int verbose,
209 const char *fn)
210{
211 struct rb_node *n;
212 struct ext2_reserve_window_node *rsv, *prev;
213 int bad;
214
215restart:
216 n = rb_first(root);
217 bad = 0;
218 prev = NULL;
219
220 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
221 while (n) {
222 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
223 if (verbose)
224 printk("reservation window 0x%p "
225 "start: %lu, end: %lu\n",
226 rsv, rsv->rsv_start, rsv->rsv_end);
227 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
228 printk("Bad reservation %p (start >= end)\n",
229 rsv);
230 bad = 1;
231 }
232 if (prev && prev->rsv_end >= rsv->rsv_start) {
233 printk("Bad reservation %p (prev->end >= start)\n",
234 rsv);
235 bad = 1;
236 }
237 if (bad) {
238 if (!verbose) {
239 printk("Restarting reservation walk in verbose mode\n");
240 verbose = 1;
241 goto restart;
242 }
243 }
244 n = rb_next(n);
245 prev = rsv;
246 }
247 printk("Window map complete.\n");
248 if (bad)
249 BUG();
250}
251#define rsv_window_dump(root, verbose) \
252 __rsv_window_dump((root), (verbose), __FUNCTION__)
253#else
254#define rsv_window_dump(root, verbose) do {} while (0)
255#endif
256
257/**
258 * goal_in_my_reservation()
259 * @rsv: inode's reservation window
260 * @grp_goal: given goal block relative to the allocation block group
261 * @group: the current allocation block group
262 * @sb: filesystem super block
263 *
264 * Test if the given goal block (group relative) is within the file's
265 * own block reservation window range.
266 *
267 * If the reservation window is outside the goal allocation group, return 0;
268 * grp_goal (given goal block) could be -1, which means no specific
269 * goal block. In this case, always return 1.
270 * If the goal block is within the reservation window, return 1;
271 * otherwise, return 0;
272 */
273static int
274goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
275 unsigned int group, struct super_block * sb)
276{
277 ext2_fsblk_t group_first_block, group_last_block;
278
279 group_first_block = ext2_group_first_block_no(sb, group);
280 group_last_block = group_first_block + EXT2_BLOCKS_PER_GROUP(sb) - 1;
281
282 if ((rsv->_rsv_start > group_last_block) ||
283 (rsv->_rsv_end < group_first_block))
284 return 0;
285 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
286 || (grp_goal + group_first_block > rsv->_rsv_end)))
287 return 0;
288 return 1;
289}
290
291/**
292 * search_reserve_window()
293 * @rb_root: root of reservation tree
294 * @goal: target allocation block
295 *
296 * Find the reserved window which includes the goal, or the previous one
297 * if the goal is not in any window.
298 * Returns NULL if there are no windows or if all windows start after the goal.
299 */
300static struct ext2_reserve_window_node *
301search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
302{
303 struct rb_node *n = root->rb_node;
304 struct ext2_reserve_window_node *rsv;
305
306 if (!n)
307 return NULL;
308
309 do {
310 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
311
312 if (goal < rsv->rsv_start)
313 n = n->rb_left;
314 else if (goal > rsv->rsv_end)
315 n = n->rb_right;
316 else
317 return rsv;
318 } while (n);
319 /*
320 * We've fallen off the end of the tree: the goal wasn't inside
321 * any particular node. OK, the previous node must be to one
322 * side of the interval containing the goal. If it's the RHS,
323 * we need to back up one.
324 */
325 if (rsv->rsv_start > goal) {
326 n = rb_prev(&rsv->rsv_node);
327 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
328 }
329 return rsv;
330}
331
332/*
333 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
334 * @sb: super block
335 * @rsv: reservation window to add
336 *
337 * Must be called with rsv_lock held.
338 */
339void ext2_rsv_window_add(struct super_block *sb,
340 struct ext2_reserve_window_node *rsv)
341{
342 struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
343 struct rb_node *node = &rsv->rsv_node;
344 ext2_fsblk_t start = rsv->rsv_start;
345
346 struct rb_node ** p = &root->rb_node;
347 struct rb_node * parent = NULL;
348 struct ext2_reserve_window_node *this;
349
350 while (*p)
351 {
352 parent = *p;
353 this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
354
355 if (start < this->rsv_start)
356 p = &(*p)->rb_left;
357 else if (start > this->rsv_end)
358 p = &(*p)->rb_right;
359 else {
360 rsv_window_dump(root, 1);
361 BUG();
362 }
363 }
364
365 rb_link_node(node, parent, p);
366 rb_insert_color(node, root);
367}
368
369/**
370 * rsv_window_remove() -- unlink a window from the reservation rb tree
371 * @sb: super block
372 * @rsv: reservation window to remove
373 *
374 * Mark the block reservation window as not allocated, and unlink it
375 * from the filesystem reservation window rb tree. Must be called with
376 * rsv_lock held.
377 */
378static void rsv_window_remove(struct super_block *sb,
379 struct ext2_reserve_window_node *rsv)
380{
381 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
382 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
383 rsv->rsv_alloc_hit = 0;
384 rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
385}
386
387/*
388 * rsv_is_empty() -- Check if the reservation window is allocated.
389 * @rsv: given reservation window to check
390 *
391 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
392 */
393static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
394{
395 /* a valid reservation end block could not be 0 */
396 return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
397}
398
399/**
400 * ext2_init_block_alloc_info()
401 * @inode: file inode structure
402 *
403 * Allocate and initialize the reservation window structure, and
404 * link the window to the ext2 inode structure at last
405 *
406 * The reservation window structure is only dynamically allocated
407 * and linked to ext2 inode the first time the open file
408 * needs a new block. So, before every ext2_new_block(s) call, for
409 * regular files, we should check whether the reservation window
410 * structure exists or not. In the latter case, this function is called.
411 * Fail to do so will result in block reservation being turned off for that
412 * open file.
413 *
414 * This function is called from ext2_get_blocks_handle(), also called
415 * when setting the reservation window size through ioctl before the file
416 * is open for write (needs block allocation).
417 *
418 * Needs truncate_mutex protection prior to calling this function.
419 */
420void ext2_init_block_alloc_info(struct inode *inode)
421{
422 struct ext2_inode_info *ei = EXT2_I(inode);
423 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
424 struct super_block *sb = inode->i_sb;
425
426 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
427 if (block_i) {
428 struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
429
430 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
431 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
432
433 /*
434 * if filesystem is mounted with NORESERVATION, the goal
435 * reservation window size is set to zero to indicate
436 * block reservation is off
437 */
438 if (!test_opt(sb, RESERVATION))
439 rsv->rsv_goal_size = 0;
440 else
441 rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
442 rsv->rsv_alloc_hit = 0;
443 block_i->last_alloc_logical_block = 0;
444 block_i->last_alloc_physical_block = 0;
445 }
446 ei->i_block_alloc_info = block_i;
447}
448
449/**
450 * ext2_discard_reservation()
451 * @inode: inode
452 *
453 * Discard(free) block reservation window on last file close, or truncate
454 * or at last iput().
455 *
456 * It is being called in three cases:
457 * ext2_release_file(): last writer closes the file
458 * ext2_clear_inode(): last iput(), when nobody links to this file.
459 * ext2_truncate(): when the block indirect map is about to change.
460 */
461void ext2_discard_reservation(struct inode *inode)
462{
463 struct ext2_inode_info *ei = EXT2_I(inode);
464 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
465 struct ext2_reserve_window_node *rsv;
466 spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
467
468 if (!block_i)
469 return;
470
471 rsv = &block_i->rsv_window_node;
472 if (!rsv_is_empty(&rsv->rsv_window)) {
473 spin_lock(rsv_lock);
474 if (!rsv_is_empty(&rsv->rsv_window))
475 rsv_window_remove(inode->i_sb, rsv);
476 spin_unlock(rsv_lock);
477 }
478}
479
480/**
481 * ext2_free_blocks_sb() -- Free given blocks and update quota and i_blocks
482 * @inode: inode
483 * @block: start physcial block to free
484 * @count: number of blocks to free
485 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486void ext2_free_blocks (struct inode * inode, unsigned long block,
487 unsigned long count)
488{
489 struct buffer_head *bitmap_bh = NULL;
490 struct buffer_head * bh2;
491 unsigned long block_group;
492 unsigned long bit;
493 unsigned long i;
494 unsigned long overflow;
495 struct super_block * sb = inode->i_sb;
496 struct ext2_sb_info * sbi = EXT2_SB(sb);
497 struct ext2_group_desc * desc;
498 struct ext2_super_block * es = sbi->s_es;
499 unsigned freed = 0, group_freed;
500
501 if (block < le32_to_cpu(es->s_first_data_block) ||
502 block + count < block ||
503 block + count > le32_to_cpu(es->s_blocks_count)) {
504 ext2_error (sb, "ext2_free_blocks",
505 "Freeing blocks not in datazone - "
506 "block = %lu, count = %lu", block, count);
507 goto error_return;
508 }
509
510 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
511
512do_more:
513 overflow = 0;
514 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
515 EXT2_BLOCKS_PER_GROUP(sb);
516 bit = (block - le32_to_cpu(es->s_first_data_block)) %
517 EXT2_BLOCKS_PER_GROUP(sb);
518 /*
519 * Check to see if we are freeing blocks across a group
520 * boundary.
521 */
522 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
523 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
524 count -= overflow;
525 }
526 brelse(bitmap_bh);
527 bitmap_bh = read_block_bitmap(sb, block_group);
528 if (!bitmap_bh)
529 goto error_return;
530
531 desc = ext2_get_group_desc (sb, block_group, &bh2);
532 if (!desc)
533 goto error_return;
534
535 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
536 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
537 in_range (block, le32_to_cpu(desc->bg_inode_table),
538 sbi->s_itb_per_group) ||
539 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -0800540 sbi->s_itb_per_group)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 ext2_error (sb, "ext2_free_blocks",
542 "Freeing blocks in system zones - "
543 "Block = %lu, count = %lu",
544 block, count);
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -0800545 goto error_return;
546 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
548 for (i = 0, group_freed = 0; i < count; i++) {
549 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
550 bit + i, bitmap_bh->b_data)) {
551 ext2_error(sb, __FUNCTION__,
552 "bit already cleared for block %lu", block + i);
553 } else {
554 group_freed++;
555 }
556 }
557
558 mark_buffer_dirty(bitmap_bh);
559 if (sb->s_flags & MS_SYNCHRONOUS)
560 sync_dirty_buffer(bitmap_bh);
561
Martin J. Bligha686cd82007-10-16 23:30:46 -0700562 group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 freed += group_freed;
564
565 if (overflow) {
566 block += count;
567 count = overflow;
568 goto do_more;
569 }
570error_return:
571 brelse(bitmap_bh);
572 release_blocks(sb, freed);
573 DQUOT_FREE_BLOCK(inode, freed);
574}
575
Martin J. Bligha686cd82007-10-16 23:30:46 -0700576/**
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)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588{
Martin J. Bligha686cd82007-10-16 23:30:46 -0700589 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 char *p, *r;
614
Martin J. Bligha686cd82007-10-16 23:30:46 -0700615 if (start > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 /*
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 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700624 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");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 }
632
Martin J. Bligha686cd82007-10-16 23:30:46 -0700633 here = start;
634 if (here < 0)
635 here = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
Martin J. Bligha686cd82007-10-16 23:30:46 -0700637 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
648/*
Martin J. Bligha686cd82007-10-16 23:30:46 -0700649 * ext2_try_to_allocate()
650 * @sb: superblock
651 * @handle: handle to this transaction
652 * @group: given allocation block group
653 * @bitmap_bh: bufferhead holds the block bitmap
654 * @grp_goal: given target block within the group
655 * @count: target number of blocks to allocate
656 * @my_rsv: reservation window
657 *
658 * Attempt to allocate blocks within a give range. Set the range of allocation
659 * first, then find the first free bit(s) from the bitmap (within the range),
660 * and at last, allocate the blocks by claiming the found free bit as allocated.
661 *
662 * To set the range of this allocation:
663 * if there is a reservation window, only try to allocate block(s)
664 * from the file's own reservation window;
665 * Otherwise, the allocation range starts from the give goal block,
666 * ends at the block group's last block.
667 *
668 * If we failed to allocate the desired block then we may end up crossing to a
669 * new bitmap.
670 */
671static int
672ext2_try_to_allocate(struct super_block *sb, int group,
673 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
674 unsigned long *count,
675 struct ext2_reserve_window *my_rsv)
676{
677 ext2_fsblk_t group_first_block;
678 ext2_grpblk_t start, end;
679 unsigned long num = 0;
680
681 /* we do allocation within the reservation window if we have a window */
682 if (my_rsv) {
683 group_first_block = ext2_group_first_block_no(sb, group);
684 if (my_rsv->_rsv_start >= group_first_block)
685 start = my_rsv->_rsv_start - group_first_block;
686 else
687 /* reservation window cross group boundary */
688 start = 0;
689 end = my_rsv->_rsv_end - group_first_block + 1;
690 if (end > EXT2_BLOCKS_PER_GROUP(sb))
691 /* reservation window crosses group boundary */
692 end = EXT2_BLOCKS_PER_GROUP(sb);
693 if ((start <= grp_goal) && (grp_goal < end))
694 start = grp_goal;
695 else
696 grp_goal = -1;
697 } else {
698 if (grp_goal > 0)
699 start = grp_goal;
700 else
701 start = 0;
702 end = EXT2_BLOCKS_PER_GROUP(sb);
703 }
704
705 BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
706
707repeat:
708 if (grp_goal < 0) {
709 grp_goal = find_next_usable_block(start, bitmap_bh, end);
710 if (grp_goal < 0)
711 goto fail_access;
712 if (!my_rsv) {
713 int i;
714
715 for (i = 0; i < 7 && grp_goal > start &&
716 !ext2_test_bit(grp_goal - 1,
717 bitmap_bh->b_data);
718 i++, grp_goal--)
719 ;
720 }
721 }
722 start = grp_goal;
723
724 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), grp_goal,
725 bitmap_bh->b_data)) {
726 /*
727 * The block was allocated by another thread, or it was
728 * allocated and then freed by another thread
729 */
730 start++;
731 grp_goal++;
732 if (start >= end)
733 goto fail_access;
734 goto repeat;
735 }
736 num++;
737 grp_goal++;
738 while (num < *count && grp_goal < end
739 && !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
740 grp_goal, bitmap_bh->b_data)) {
741 num++;
742 grp_goal++;
743 }
744 *count = num;
745 return grp_goal - num;
746fail_access:
747 *count = num;
748 return -1;
749}
750
751/**
752 * find_next_reservable_window():
753 * find a reservable space within the given range.
754 * It does not allocate the reservation window for now:
755 * alloc_new_reservation() will do the work later.
756 *
757 * @search_head: the head of the searching list;
758 * This is not necessarily the list head of the whole filesystem
759 *
760 * We have both head and start_block to assist the search
761 * for the reservable space. The list starts from head,
762 * but we will shift to the place where start_block is,
763 * then start from there, when looking for a reservable space.
764 *
765 * @size: the target new reservation window size
766 *
767 * @group_first_block: the first block we consider to start
768 * the real search from
769 *
770 * @last_block:
771 * the maximum block number that our goal reservable space
772 * could start from. This is normally the last block in this
773 * group. The search will end when we found the start of next
774 * possible reservable space is out of this boundary.
775 * This could handle the cross boundary reservation window
776 * request.
777 *
778 * basically we search from the given range, rather than the whole
779 * reservation double linked list, (start_block, last_block)
780 * to find a free region that is of my size and has not
781 * been reserved.
782 *
783 */
784static int find_next_reservable_window(
785 struct ext2_reserve_window_node *search_head,
786 struct ext2_reserve_window_node *my_rsv,
787 struct super_block * sb,
788 ext2_fsblk_t start_block,
789 ext2_fsblk_t last_block)
790{
791 struct rb_node *next;
792 struct ext2_reserve_window_node *rsv, *prev;
793 ext2_fsblk_t cur;
794 int size = my_rsv->rsv_goal_size;
795
796 /* TODO: make the start of the reservation window byte-aligned */
797 /* cur = *start_block & ~7;*/
798 cur = start_block;
799 rsv = search_head;
800 if (!rsv)
801 return -1;
802
803 while (1) {
804 if (cur <= rsv->rsv_end)
805 cur = rsv->rsv_end + 1;
806
807 /* TODO?
808 * in the case we could not find a reservable space
809 * that is what is expected, during the re-search, we could
810 * remember what's the largest reservable space we could have
811 * and return that one.
812 *
813 * For now it will fail if we could not find the reservable
814 * space with expected-size (or more)...
815 */
816 if (cur > last_block)
817 return -1; /* fail */
818
819 prev = rsv;
820 next = rb_next(&rsv->rsv_node);
821 rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
822
823 /*
824 * Reached the last reservation, we can just append to the
825 * previous one.
826 */
827 if (!next)
828 break;
829
830 if (cur + size <= rsv->rsv_start) {
831 /*
832 * Found a reserveable space big enough. We could
833 * have a reservation across the group boundary here
834 */
835 break;
836 }
837 }
838 /*
839 * we come here either :
840 * when we reach the end of the whole list,
841 * and there is empty reservable space after last entry in the list.
842 * append it to the end of the list.
843 *
844 * or we found one reservable space in the middle of the list,
845 * return the reservation window that we could append to.
846 * succeed.
847 */
848
849 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
850 rsv_window_remove(sb, my_rsv);
851
852 /*
853 * Let's book the whole avaliable window for now. We will check the
854 * disk bitmap later and then, if there are free blocks then we adjust
855 * the window size if it's larger than requested.
856 * Otherwise, we will remove this node from the tree next time
857 * call find_next_reservable_window.
858 */
859 my_rsv->rsv_start = cur;
860 my_rsv->rsv_end = cur + size - 1;
861 my_rsv->rsv_alloc_hit = 0;
862
863 if (prev != my_rsv)
864 ext2_rsv_window_add(sb, my_rsv);
865
866 return 0;
867}
868
869/**
870 * alloc_new_reservation()--allocate a new reservation window
871 *
872 * To make a new reservation, we search part of the filesystem
873 * reservation list (the list that inside the group). We try to
874 * allocate a new reservation window near the allocation goal,
875 * or the beginning of the group, if there is no goal.
876 *
877 * We first find a reservable space after the goal, then from
878 * there, we check the bitmap for the first free block after
879 * it. If there is no free block until the end of group, then the
880 * whole group is full, we failed. Otherwise, check if the free
881 * block is inside the expected reservable space, if so, we
882 * succeed.
883 * If the first free block is outside the reservable space, then
884 * start from the first free block, we search for next available
885 * space, and go on.
886 *
887 * on succeed, a new reservation will be found and inserted into the list
888 * It contains at least one free block, and it does not overlap with other
889 * reservation windows.
890 *
891 * failed: we failed to find a reservation window in this group
892 *
893 * @rsv: the reservation
894 *
895 * @grp_goal: The goal (group-relative). It is where the search for a
896 * free reservable space should start from.
897 * if we have a goal(goal >0 ), then start from there,
898 * no goal(goal = -1), we start from the first block
899 * of the group.
900 *
901 * @sb: the super block
902 * @group: the group we are trying to allocate in
903 * @bitmap_bh: the block group block bitmap
904 *
905 */
906static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
907 ext2_grpblk_t grp_goal, struct super_block *sb,
908 unsigned int group, struct buffer_head *bitmap_bh)
909{
910 struct ext2_reserve_window_node *search_head;
911 ext2_fsblk_t group_first_block, group_end_block, start_block;
912 ext2_grpblk_t first_free_block;
913 struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
914 unsigned long size;
915 int ret;
916 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
917
918 group_first_block = ext2_group_first_block_no(sb, group);
919 group_end_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
920
921 if (grp_goal < 0)
922 start_block = group_first_block;
923 else
924 start_block = grp_goal + group_first_block;
925
926 size = my_rsv->rsv_goal_size;
927
928 if (!rsv_is_empty(&my_rsv->rsv_window)) {
929 /*
930 * if the old reservation is cross group boundary
931 * and if the goal is inside the old reservation window,
932 * we will come here when we just failed to allocate from
933 * the first part of the window. We still have another part
934 * that belongs to the next group. In this case, there is no
935 * point to discard our window and try to allocate a new one
936 * in this group(which will fail). we should
937 * keep the reservation window, just simply move on.
938 *
939 * Maybe we could shift the start block of the reservation
940 * window to the first block of next group.
941 */
942
943 if ((my_rsv->rsv_start <= group_end_block) &&
944 (my_rsv->rsv_end > group_end_block) &&
945 (start_block >= my_rsv->rsv_start))
946 return -1;
947
948 if ((my_rsv->rsv_alloc_hit >
949 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
950 /*
951 * if the previously allocation hit ratio is
952 * greater than 1/2, then we double the size of
953 * the reservation window the next time,
954 * otherwise we keep the same size window
955 */
956 size = size * 2;
957 if (size > EXT2_MAX_RESERVE_BLOCKS)
958 size = EXT2_MAX_RESERVE_BLOCKS;
959 my_rsv->rsv_goal_size= size;
960 }
961 }
962
963 spin_lock(rsv_lock);
964 /*
965 * shift the search start to the window near the goal block
966 */
967 search_head = search_reserve_window(fs_rsv_root, start_block);
968
969 /*
970 * find_next_reservable_window() simply finds a reservable window
971 * inside the given range(start_block, group_end_block).
972 *
973 * To make sure the reservation window has a free bit inside it, we
974 * need to check the bitmap after we found a reservable window.
975 */
976retry:
977 ret = find_next_reservable_window(search_head, my_rsv, sb,
978 start_block, group_end_block);
979
980 if (ret == -1) {
981 if (!rsv_is_empty(&my_rsv->rsv_window))
982 rsv_window_remove(sb, my_rsv);
983 spin_unlock(rsv_lock);
984 return -1;
985 }
986
987 /*
988 * On success, find_next_reservable_window() returns the
989 * reservation window where there is a reservable space after it.
990 * Before we reserve this reservable space, we need
991 * to make sure there is at least a free block inside this region.
992 *
993 * Search the first free bit on the block bitmap. Search starts from
994 * the start block of the reservable space we just found.
995 */
996 spin_unlock(rsv_lock);
997 first_free_block = bitmap_search_next_usable_block(
998 my_rsv->rsv_start - group_first_block,
999 bitmap_bh, group_end_block - group_first_block + 1);
1000
1001 if (first_free_block < 0) {
1002 /*
1003 * no free block left on the bitmap, no point
1004 * to reserve the space. return failed.
1005 */
1006 spin_lock(rsv_lock);
1007 if (!rsv_is_empty(&my_rsv->rsv_window))
1008 rsv_window_remove(sb, my_rsv);
1009 spin_unlock(rsv_lock);
1010 return -1; /* failed */
1011 }
1012
1013 start_block = first_free_block + group_first_block;
1014 /*
1015 * check if the first free block is within the
1016 * free space we just reserved
1017 */
1018 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1019 return 0; /* success */
1020 /*
1021 * if the first free bit we found is out of the reservable space
1022 * continue search for next reservable space,
1023 * start from where the free block is,
1024 * we also shift the list head to where we stopped last time
1025 */
1026 search_head = my_rsv;
1027 spin_lock(rsv_lock);
1028 goto retry;
1029}
1030
1031/**
1032 * try_to_extend_reservation()
1033 * @my_rsv: given reservation window
1034 * @sb: super block
1035 * @size: the delta to extend
1036 *
1037 * Attempt to expand the reservation window large enough to have
1038 * required number of free blocks
1039 *
1040 * Since ext2_try_to_allocate() will always allocate blocks within
1041 * the reservation window range, if the window size is too small,
1042 * multiple blocks allocation has to stop at the end of the reservation
1043 * window. To make this more efficient, given the total number of
1044 * blocks needed and the current size of the window, we try to
1045 * expand the reservation window size if necessary on a best-effort
1046 * basis before ext2_new_blocks() tries to allocate blocks.
1047 */
1048static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1049 struct super_block *sb, int size)
1050{
1051 struct ext2_reserve_window_node *next_rsv;
1052 struct rb_node *next;
1053 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1054
1055 if (!spin_trylock(rsv_lock))
1056 return;
1057
1058 next = rb_next(&my_rsv->rsv_node);
1059
1060 if (!next)
1061 my_rsv->rsv_end += size;
1062 else {
1063 next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1064
1065 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1066 my_rsv->rsv_end += size;
1067 else
1068 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1069 }
1070 spin_unlock(rsv_lock);
1071}
1072
1073/**
1074 * ext2_try_to_allocate_with_rsv()
1075 * @sb: superblock
1076 * @group: given allocation block group
1077 * @bitmap_bh: bufferhead holds the block bitmap
1078 * @grp_goal: given target block within the group
1079 * @count: target number of blocks to allocate
1080 * @my_rsv: reservation window
1081 *
1082 * This is the main function used to allocate a new block and its reservation
1083 * window.
1084 *
1085 * Each time when a new block allocation is need, first try to allocate from
1086 * its own reservation. If it does not have a reservation window, instead of
1087 * looking for a free bit on bitmap first, then look up the reservation list to
1088 * see if it is inside somebody else's reservation window, we try to allocate a
1089 * reservation window for it starting from the goal first. Then do the block
1090 * allocation within the reservation window.
1091 *
1092 * This will avoid keeping on searching the reservation list again and
1093 * again when somebody is looking for a free block (without
1094 * reservation), and there are lots of free blocks, but they are all
1095 * being reserved.
1096 *
1097 * We use a red-black tree for the per-filesystem reservation list.
1098 */
1099static ext2_grpblk_t
1100ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1101 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1102 struct ext2_reserve_window_node * my_rsv,
1103 unsigned long *count)
1104{
1105 ext2_fsblk_t group_first_block, group_last_block;
1106 ext2_grpblk_t ret = 0;
1107 unsigned long num = *count;
1108
1109 /*
1110 * we don't deal with reservation when
1111 * filesystem is mounted without reservation
1112 * or the file is not a regular file
1113 * or last attempt to allocate a block with reservation turned on failed
1114 */
1115 if (my_rsv == NULL) {
1116 return ext2_try_to_allocate(sb, group, bitmap_bh,
1117 grp_goal, count, NULL);
1118 }
1119 /*
1120 * grp_goal is a group relative block number (if there is a goal)
1121 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1122 * first block is a filesystem wide block number
1123 * first block is the block number of the first block in this group
1124 */
1125 group_first_block = ext2_group_first_block_no(sb, group);
1126 group_last_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
1127
1128 /*
1129 * Basically we will allocate a new block from inode's reservation
1130 * window.
1131 *
1132 * We need to allocate a new reservation window, if:
1133 * a) inode does not have a reservation window; or
1134 * b) last attempt to allocate a block from existing reservation
1135 * failed; or
1136 * c) we come here with a goal and with a reservation window
1137 *
1138 * We do not need to allocate a new reservation window if we come here
1139 * at the beginning with a goal and the goal is inside the window, or
1140 * we don't have a goal but already have a reservation window.
1141 * then we could go to allocate from the reservation window directly.
1142 */
1143 while (1) {
1144 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1145 !goal_in_my_reservation(&my_rsv->rsv_window,
1146 grp_goal, group, sb)) {
1147 if (my_rsv->rsv_goal_size < *count)
1148 my_rsv->rsv_goal_size = *count;
1149 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1150 group, bitmap_bh);
1151 if (ret < 0)
1152 break; /* failed */
1153
1154 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1155 grp_goal, group, sb))
1156 grp_goal = -1;
1157 } else if (grp_goal >= 0) {
1158 int curr = my_rsv->rsv_end -
1159 (grp_goal + group_first_block) + 1;
1160
1161 if (curr < *count)
1162 try_to_extend_reservation(my_rsv, sb,
1163 *count - curr);
1164 }
1165
1166 if ((my_rsv->rsv_start > group_last_block) ||
1167 (my_rsv->rsv_end < group_first_block)) {
1168 rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1169 BUG();
1170 }
1171 ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1172 &num, &my_rsv->rsv_window);
1173 if (ret >= 0) {
1174 my_rsv->rsv_alloc_hit += num;
1175 *count = num;
1176 break; /* succeed */
1177 }
1178 num = *count;
1179 }
1180 return ret;
1181}
1182
1183/**
1184 * ext2_has_free_blocks()
1185 * @sbi: in-core super block structure.
1186 *
1187 * Check if filesystem has at least 1 free block available for allocation.
1188 */
1189static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1190{
1191 ext2_fsblk_t free_blocks, root_blocks;
1192
1193 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1194 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1195 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1196 sbi->s_resuid != current->fsuid &&
1197 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1198 return 0;
1199 }
1200 return 1;
1201}
1202
1203/*
1204 * ext2_new_blocks() -- core block(s) allocation function
1205 * @inode: file inode
1206 * @goal: given target block(filesystem wide)
1207 * @count: target number of blocks to allocate
1208 * @errp: error code
1209 *
1210 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 * free, or there is a free block within 32 blocks of the goal, that block
1212 * is allocated. Otherwise a forward search is made for a free block; within
1213 * each block group the search first looks for an entire free byte in the block
1214 * bitmap, and then for any free bit if that fails.
1215 * This function also updates quota and i_blocks field.
1216 */
Martin J. Bligha686cd82007-10-16 23:30:46 -07001217ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1218 unsigned long *count, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219{
1220 struct buffer_head *bitmap_bh = NULL;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001221 struct buffer_head *gdp_bh;
1222 int group_no;
1223 int goal_group;
1224 ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1225 ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1226 ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
1227 int bgi; /* blockgroup iteration index */
1228 int performed_allocation = 0;
1229 ext2_grpblk_t free_blocks; /* number of free blocks in a group */
1230 struct super_block *sb;
1231 struct ext2_group_desc *gdp;
1232 struct ext2_super_block *es;
1233 struct ext2_sb_info *sbi;
1234 struct ext2_reserve_window_node *my_rsv = NULL;
1235 struct ext2_block_alloc_info *block_i;
1236 unsigned short windowsz = 0;
1237 unsigned long ngroups;
1238 unsigned long num = *count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Martin J. Bligha686cd82007-10-16 23:30:46 -07001240 *errp = -ENOSPC;
1241 sb = inode->i_sb;
1242 if (!sb) {
1243 printk("ext2_new_blocks: nonexistent device");
1244 return 0;
1245 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Martin J. Bligha686cd82007-10-16 23:30:46 -07001247 /*
1248 * Check quota for allocation of this block.
1249 */
1250 if (DQUOT_ALLOC_BLOCK(inode, num)) {
1251 *errp = -EDQUOT;
1252 return 0;
1253 }
1254
1255 sbi = EXT2_SB(sb);
1256 es = EXT2_SB(sb)->s_es;
1257 ext2_debug("goal=%lu.\n", goal);
1258 /*
1259 * Allocate a block from reservation only when
1260 * filesystem is mounted with reservation(default,-o reservation), and
1261 * it's a regular file, and
1262 * the desired window size is greater than 0 (One could use ioctl
1263 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1264 * reservation on that particular file)
1265 */
1266 block_i = EXT2_I(inode)->i_block_alloc_info;
1267 if (block_i) {
1268 windowsz = block_i->rsv_window_node.rsv_goal_size;
1269 if (windowsz > 0)
1270 my_rsv = &block_i->rsv_window_node;
1271 }
1272
1273 if (!ext2_has_free_blocks(sbi)) {
1274 *errp = -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 goto out;
1276 }
1277
Martin J. Bligha686cd82007-10-16 23:30:46 -07001278 /*
1279 * First, test whether the goal block is free.
1280 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 if (goal < le32_to_cpu(es->s_first_data_block) ||
1282 goal >= le32_to_cpu(es->s_blocks_count))
1283 goal = le32_to_cpu(es->s_first_data_block);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001284 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1285 EXT2_BLOCKS_PER_GROUP(sb);
1286 goal_group = group_no;
1287retry_alloc:
1288 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1289 if (!gdp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 goto io_error;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001291
1292 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1293 /*
1294 * if there is not enough free blocks to make a new resevation
1295 * turn off reservation for this allocation
1296 */
1297 if (my_rsv && (free_blocks < windowsz)
1298 && (rsv_is_empty(&my_rsv->rsv_window)))
1299 my_rsv = NULL;
1300
1301 if (free_blocks > 0) {
1302 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1303 EXT2_BLOCKS_PER_GROUP(sb));
1304 bitmap_bh = read_block_bitmap(sb, group_no);
1305 if (!bitmap_bh)
1306 goto io_error;
1307 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1308 bitmap_bh, grp_target_blk,
1309 my_rsv, &num);
1310 if (grp_alloc_blk >= 0)
1311 goto allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 }
1313
Martin J. Bligha686cd82007-10-16 23:30:46 -07001314 ngroups = EXT2_SB(sb)->s_groups_count;
1315 smp_rmb();
1316
1317 /*
Akinobu Mita144704e2008-02-06 01:40:15 -08001318 * Now search the rest of the groups. We assume that
1319 * group_no and gdp correctly point to the last group visited.
Martin J. Bligha686cd82007-10-16 23:30:46 -07001320 */
1321 for (bgi = 0; bgi < ngroups; bgi++) {
1322 group_no++;
1323 if (group_no >= ngroups)
1324 group_no = 0;
1325 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1326 if (!gdp)
1327 goto io_error;
1328
1329 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
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 (free_blocks <= (windowsz/2))
1336 continue;
1337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 brelse(bitmap_bh);
1339 bitmap_bh = read_block_bitmap(sb, group_no);
1340 if (!bitmap_bh)
1341 goto io_error;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001342 /*
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 /*
Martin J. Bligha686cd82007-10-16 23:30:46 -07001351 * We may end up a bogus ealier ENOSPC error due to
1352 * filesystem is "full" of reservations, but
1353 * there maybe indeed free blocks avaliable on disk
1354 * In this case, we just forget about the reservations
1355 * just do block allocation as without reservations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 */
Martin J. Bligha686cd82007-10-16 23:30:46 -07001357 if (my_rsv) {
1358 my_rsv = NULL;
1359 windowsz = 0;
1360 group_no = goal_group;
1361 goto retry_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 }
Martin J. Bligha686cd82007-10-16 23:30:46 -07001363 /* No space left on the device */
1364 *errp = -ENOSPC;
1365 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
Martin J. Bligha686cd82007-10-16 23:30:46 -07001367allocated:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 ext2_debug("using block group %d(%d)\n",
Martin J. Bligha686cd82007-10-16 23:30:46 -07001370 group_no, gdp->bg_free_blocks_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Martin J. Bligha686cd82007-10-16 23:30:46 -07001372 ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
Martin J. Bligha686cd82007-10-16 23:30:46 -07001374 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),
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -08001379 EXT2_SB(sb)->s_itb_per_group)) {
Martin J. Bligha686cd82007-10-16 23:30:46 -07001380 ext2_error(sb, "ext2_new_blocks",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 "Allocating block in system zone - "
Martin J. Bligha686cd82007-10-16 23:30:46 -07001382 "blocks from "E2FSBLK", length %lu",
1383 ret_block, num);
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -08001384 goto out;
1385 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Martin J. Bligha686cd82007-10-16 23:30:46 -07001387 performed_allocation = 1;
1388
1389 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1390 ext2_error(sb, "ext2_new_blocks",
1391 "block("E2FSBLK") >= blocks count(%d) - "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 "block_group = %d, es == %p ", ret_block,
1393 le32_to_cpu(es->s_blocks_count), group_no, es);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001394 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
Martin J. Bligha686cd82007-10-16 23:30:46 -07001397 group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1398 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
1400 mark_buffer_dirty(bitmap_bh);
1401 if (sb->s_flags & MS_SYNCHRONOUS)
1402 sync_dirty_buffer(bitmap_bh);
1403
Martin J. Bligha686cd82007-10-16 23:30:46 -07001404 *errp = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 brelse(bitmap_bh);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001406 DQUOT_FREE_BLOCK(inode, *count-num);
1407 *count = num;
1408 return ret_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
1410io_error:
Martin J. Bligha686cd82007-10-16 23:30:46 -07001411 *errp = -EIO;
1412out:
1413 /*
1414 * Undo the block allocation
1415 */
1416 if (!performed_allocation)
1417 DQUOT_FREE_BLOCK(inode, *count);
1418 brelse(bitmap_bh);
1419 return 0;
1420}
1421
1422ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1423{
1424 unsigned long count = 1;
1425
1426 return ext2_new_blocks(inode, goal, &count, errp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427}
1428
Valerie Henson21730ee2006-06-25 05:48:12 -07001429#ifdef EXT2FS_DEBUG
1430
Philippe De Muyterfebfcf92007-10-16 23:26:15 -07001431static const int nibblemap[] = {4, 3, 3, 2, 3, 2, 2, 1, 3, 2, 2, 1, 2, 1, 1, 0};
Valerie Henson21730ee2006-06-25 05:48:12 -07001432
1433unsigned long ext2_count_free (struct buffer_head * map, unsigned int numchars)
1434{
1435 unsigned int i;
1436 unsigned long sum = 0;
1437
1438 if (!map)
1439 return (0);
1440 for (i = 0; i < numchars; i++)
1441 sum += nibblemap[map->b_data[i] & 0xf] +
1442 nibblemap[(map->b_data[i] >> 4) & 0xf];
1443 return (sum);
1444}
1445
1446#endif /* EXT2FS_DEBUG */
1447
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448unsigned long ext2_count_free_blocks (struct super_block * sb)
1449{
1450 struct ext2_group_desc * desc;
1451 unsigned long desc_count = 0;
1452 int i;
1453#ifdef EXT2FS_DEBUG
1454 unsigned long bitmap_count, x;
1455 struct ext2_super_block *es;
1456
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 es = EXT2_SB(sb)->s_es;
1458 desc_count = 0;
1459 bitmap_count = 0;
1460 desc = NULL;
1461 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1462 struct buffer_head *bitmap_bh;
1463 desc = ext2_get_group_desc (sb, i, NULL);
1464 if (!desc)
1465 continue;
1466 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1467 bitmap_bh = read_block_bitmap(sb, i);
1468 if (!bitmap_bh)
1469 continue;
1470
1471 x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1472 printk ("group %d: stored = %d, counted = %lu\n",
1473 i, le16_to_cpu(desc->bg_free_blocks_count), x);
1474 bitmap_count += x;
1475 brelse(bitmap_bh);
1476 }
1477 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1478 (long)le32_to_cpu(es->s_free_blocks_count),
1479 desc_count, bitmap_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 return bitmap_count;
1481#else
1482 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1483 desc = ext2_get_group_desc (sb, i, NULL);
1484 if (!desc)
1485 continue;
1486 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1487 }
1488 return desc_count;
1489#endif
1490}
1491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492static inline int test_root(int a, int b)
1493{
1494 int num = b;
1495
1496 while (a > num)
1497 num *= b;
1498 return num == a;
1499}
1500
1501static int ext2_group_sparse(int group)
1502{
1503 if (group <= 1)
1504 return 1;
1505 return (test_root(group, 3) || test_root(group, 5) ||
1506 test_root(group, 7));
1507}
1508
1509/**
1510 * ext2_bg_has_super - number of blocks used by the superblock in group
1511 * @sb: superblock for filesystem
1512 * @group: group number to check
1513 *
1514 * Return the number of blocks used by the superblock (primary or backup)
1515 * in this group. Currently this will be only 0 or 1.
1516 */
1517int ext2_bg_has_super(struct super_block *sb, int group)
1518{
1519 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1520 !ext2_group_sparse(group))
1521 return 0;
1522 return 1;
1523}
1524
1525/**
1526 * ext2_bg_num_gdb - number of blocks used by the group table in group
1527 * @sb: superblock for filesystem
1528 * @group: group number to check
1529 *
1530 * Return the number of blocks used by the group descriptor table
1531 * (primary or backup) in this group. In the future there may be a
1532 * different number of descriptor blocks in each group.
1533 */
1534unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1535{
Akinobu Mita859cb932008-02-06 01:40:17 -08001536 return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537}
1538