blob: c834f512dce5fc21026a2345ca6366d3092c2e77 [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");
Julia Lawall2c116192008-04-28 02:16:02 -0700248 BUG_ON(bad);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700249}
250#define rsv_window_dump(root, verbose) \
251 __rsv_window_dump((root), (verbose), __FUNCTION__)
252#else
253#define rsv_window_dump(root, verbose) do {} while (0)
254#endif
255
256/**
257 * goal_in_my_reservation()
258 * @rsv: inode's reservation window
259 * @grp_goal: given goal block relative to the allocation block group
260 * @group: the current allocation block group
261 * @sb: filesystem super block
262 *
263 * Test if the given goal block (group relative) is within the file's
264 * own block reservation window range.
265 *
266 * If the reservation window is outside the goal allocation group, return 0;
267 * grp_goal (given goal block) could be -1, which means no specific
268 * goal block. In this case, always return 1.
269 * If the goal block is within the reservation window, return 1;
270 * otherwise, return 0;
271 */
272static int
273goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
274 unsigned int group, struct super_block * sb)
275{
276 ext2_fsblk_t group_first_block, group_last_block;
277
278 group_first_block = ext2_group_first_block_no(sb, group);
279 group_last_block = group_first_block + EXT2_BLOCKS_PER_GROUP(sb) - 1;
280
281 if ((rsv->_rsv_start > group_last_block) ||
282 (rsv->_rsv_end < group_first_block))
283 return 0;
284 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
285 || (grp_goal + group_first_block > rsv->_rsv_end)))
286 return 0;
287 return 1;
288}
289
290/**
291 * search_reserve_window()
292 * @rb_root: root of reservation tree
293 * @goal: target allocation block
294 *
295 * Find the reserved window which includes the goal, or the previous one
296 * if the goal is not in any window.
297 * Returns NULL if there are no windows or if all windows start after the goal.
298 */
299static struct ext2_reserve_window_node *
300search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
301{
302 struct rb_node *n = root->rb_node;
303 struct ext2_reserve_window_node *rsv;
304
305 if (!n)
306 return NULL;
307
308 do {
309 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
310
311 if (goal < rsv->rsv_start)
312 n = n->rb_left;
313 else if (goal > rsv->rsv_end)
314 n = n->rb_right;
315 else
316 return rsv;
317 } while (n);
318 /*
319 * We've fallen off the end of the tree: the goal wasn't inside
320 * any particular node. OK, the previous node must be to one
321 * side of the interval containing the goal. If it's the RHS,
322 * we need to back up one.
323 */
324 if (rsv->rsv_start > goal) {
325 n = rb_prev(&rsv->rsv_node);
326 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
327 }
328 return rsv;
329}
330
331/*
332 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
333 * @sb: super block
334 * @rsv: reservation window to add
335 *
336 * Must be called with rsv_lock held.
337 */
338void ext2_rsv_window_add(struct super_block *sb,
339 struct ext2_reserve_window_node *rsv)
340{
341 struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
342 struct rb_node *node = &rsv->rsv_node;
343 ext2_fsblk_t start = rsv->rsv_start;
344
345 struct rb_node ** p = &root->rb_node;
346 struct rb_node * parent = NULL;
347 struct ext2_reserve_window_node *this;
348
349 while (*p)
350 {
351 parent = *p;
352 this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
353
354 if (start < this->rsv_start)
355 p = &(*p)->rb_left;
356 else if (start > this->rsv_end)
357 p = &(*p)->rb_right;
358 else {
359 rsv_window_dump(root, 1);
360 BUG();
361 }
362 }
363
364 rb_link_node(node, parent, p);
365 rb_insert_color(node, root);
366}
367
368/**
369 * rsv_window_remove() -- unlink a window from the reservation rb tree
370 * @sb: super block
371 * @rsv: reservation window to remove
372 *
373 * Mark the block reservation window as not allocated, and unlink it
374 * from the filesystem reservation window rb tree. Must be called with
375 * rsv_lock held.
376 */
377static void rsv_window_remove(struct super_block *sb,
378 struct ext2_reserve_window_node *rsv)
379{
380 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
381 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
382 rsv->rsv_alloc_hit = 0;
383 rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
384}
385
386/*
387 * rsv_is_empty() -- Check if the reservation window is allocated.
388 * @rsv: given reservation window to check
389 *
390 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
391 */
392static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
393{
394 /* a valid reservation end block could not be 0 */
395 return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
396}
397
398/**
399 * ext2_init_block_alloc_info()
400 * @inode: file inode structure
401 *
402 * Allocate and initialize the reservation window structure, and
403 * link the window to the ext2 inode structure at last
404 *
405 * The reservation window structure is only dynamically allocated
406 * and linked to ext2 inode the first time the open file
407 * needs a new block. So, before every ext2_new_block(s) call, for
408 * regular files, we should check whether the reservation window
409 * structure exists or not. In the latter case, this function is called.
410 * Fail to do so will result in block reservation being turned off for that
411 * open file.
412 *
413 * This function is called from ext2_get_blocks_handle(), also called
414 * when setting the reservation window size through ioctl before the file
415 * is open for write (needs block allocation).
416 *
417 * Needs truncate_mutex protection prior to calling this function.
418 */
419void ext2_init_block_alloc_info(struct inode *inode)
420{
421 struct ext2_inode_info *ei = EXT2_I(inode);
422 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
423 struct super_block *sb = inode->i_sb;
424
425 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
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_sb() -- Free given blocks and update quota and i_blocks
481 * @inode: inode
482 * @block: start physcial block to free
483 * @count: number of blocks to free
484 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485void ext2_free_blocks (struct inode * inode, unsigned long 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 (block < le32_to_cpu(es->s_first_data_block) ||
501 block + count < block ||
502 block + count > le32_to_cpu(es->s_blocks_count)) {
503 ext2_error (sb, "ext2_free_blocks",
504 "Freeing blocks not in datazone - "
505 "block = %lu, count = %lu", block, count);
506 goto error_return;
507 }
508
509 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
510
511do_more:
512 overflow = 0;
513 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
514 EXT2_BLOCKS_PER_GROUP(sb);
515 bit = (block - le32_to_cpu(es->s_first_data_block)) %
516 EXT2_BLOCKS_PER_GROUP(sb);
517 /*
518 * Check to see if we are freeing blocks across a group
519 * boundary.
520 */
521 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
522 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
523 count -= overflow;
524 }
525 brelse(bitmap_bh);
526 bitmap_bh = read_block_bitmap(sb, block_group);
527 if (!bitmap_bh)
528 goto error_return;
529
530 desc = ext2_get_group_desc (sb, block_group, &bh2);
531 if (!desc)
532 goto error_return;
533
534 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
535 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
536 in_range (block, le32_to_cpu(desc->bg_inode_table),
537 sbi->s_itb_per_group) ||
538 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -0800539 sbi->s_itb_per_group)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 ext2_error (sb, "ext2_free_blocks",
541 "Freeing blocks in system zones - "
542 "Block = %lu, count = %lu",
543 block, count);
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -0800544 goto error_return;
545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
547 for (i = 0, group_freed = 0; i < count; i++) {
548 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
549 bit + i, bitmap_bh->b_data)) {
550 ext2_error(sb, __FUNCTION__,
551 "bit already cleared for block %lu", block + i);
552 } else {
553 group_freed++;
554 }
555 }
556
557 mark_buffer_dirty(bitmap_bh);
558 if (sb->s_flags & MS_SYNCHRONOUS)
559 sync_dirty_buffer(bitmap_bh);
560
Martin J. Bligha686cd82007-10-16 23:30:46 -0700561 group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 freed += group_freed;
563
564 if (overflow) {
565 block += count;
566 count = overflow;
567 goto do_more;
568 }
569error_return:
570 brelse(bitmap_bh);
571 release_blocks(sb, freed);
572 DQUOT_FREE_BLOCK(inode, freed);
573}
574
Martin J. Bligha686cd82007-10-16 23:30:46 -0700575/**
576 * bitmap_search_next_usable_block()
577 * @start: the starting block (group relative) of the search
578 * @bh: bufferhead contains the block group bitmap
579 * @maxblocks: the ending block (group relative) of the reservation
580 *
581 * The bitmap search --- search forward through the actual bitmap on disk until
582 * we find a bit free.
583 */
584static ext2_grpblk_t
585bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
586 ext2_grpblk_t maxblocks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587{
Martin J. Bligha686cd82007-10-16 23:30:46 -0700588 ext2_grpblk_t next;
589
590 next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
591 if (next >= maxblocks)
592 return -1;
593 return next;
594}
595
596/**
597 * find_next_usable_block()
598 * @start: the starting block (group relative) to find next
599 * allocatable block in bitmap.
600 * @bh: bufferhead contains the block group bitmap
601 * @maxblocks: the ending block (group relative) for the search
602 *
603 * Find an allocatable block in a bitmap. We perform the "most
604 * appropriate allocation" algorithm of looking for a free block near
605 * the initial goal; then for a free byte somewhere in the bitmap;
606 * then for any free bit in the bitmap.
607 */
608static ext2_grpblk_t
609find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
610{
611 ext2_grpblk_t here, next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 char *p, *r;
613
Martin J. Bligha686cd82007-10-16 23:30:46 -0700614 if (start > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 /*
616 * The goal was occupied; search forward for a free
617 * block within the next XX blocks.
618 *
619 * end_goal is more or less random, but it has to be
620 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
621 * next 64-bit boundary is simple..
622 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700623 ext2_grpblk_t end_goal = (start + 63) & ~63;
624 if (end_goal > maxblocks)
625 end_goal = maxblocks;
626 here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
627 if (here < end_goal)
628 return here;
629 ext2_debug("Bit not found near goal\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 }
631
Martin J. Bligha686cd82007-10-16 23:30:46 -0700632 here = start;
633 if (here < 0)
634 here = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
Martin J. Bligha686cd82007-10-16 23:30:46 -0700636 p = ((char *)bh->b_data) + (here >> 3);
637 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
638 next = (r - ((char *)bh->b_data)) << 3;
639
640 if (next < maxblocks && next >= here)
641 return next;
642
643 here = bitmap_search_next_usable_block(here, bh, maxblocks);
644 return here;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
647/*
Martin J. Bligha686cd82007-10-16 23:30:46 -0700648 * ext2_try_to_allocate()
649 * @sb: superblock
650 * @handle: handle to this transaction
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;
677 ext2_grpblk_t start, end;
678 unsigned long num = 0;
679
680 /* we do allocation within the reservation window if we have a window */
681 if (my_rsv) {
682 group_first_block = ext2_group_first_block_no(sb, group);
683 if (my_rsv->_rsv_start >= group_first_block)
684 start = my_rsv->_rsv_start - group_first_block;
685 else
686 /* reservation window cross group boundary */
687 start = 0;
688 end = my_rsv->_rsv_end - group_first_block + 1;
689 if (end > EXT2_BLOCKS_PER_GROUP(sb))
690 /* reservation window crosses group boundary */
691 end = EXT2_BLOCKS_PER_GROUP(sb);
692 if ((start <= grp_goal) && (grp_goal < end))
693 start = grp_goal;
694 else
695 grp_goal = -1;
696 } else {
697 if (grp_goal > 0)
698 start = grp_goal;
699 else
700 start = 0;
701 end = EXT2_BLOCKS_PER_GROUP(sb);
702 }
703
704 BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
705
706repeat:
707 if (grp_goal < 0) {
708 grp_goal = find_next_usable_block(start, bitmap_bh, end);
709 if (grp_goal < 0)
710 goto fail_access;
711 if (!my_rsv) {
712 int i;
713
714 for (i = 0; i < 7 && grp_goal > start &&
715 !ext2_test_bit(grp_goal - 1,
716 bitmap_bh->b_data);
717 i++, grp_goal--)
718 ;
719 }
720 }
721 start = grp_goal;
722
723 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), grp_goal,
724 bitmap_bh->b_data)) {
725 /*
726 * The block was allocated by another thread, or it was
727 * allocated and then freed by another thread
728 */
729 start++;
730 grp_goal++;
731 if (start >= end)
732 goto fail_access;
733 goto repeat;
734 }
735 num++;
736 grp_goal++;
737 while (num < *count && grp_goal < end
738 && !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
739 grp_goal, bitmap_bh->b_data)) {
740 num++;
741 grp_goal++;
742 }
743 *count = num;
744 return grp_goal - num;
745fail_access:
746 *count = num;
747 return -1;
748}
749
750/**
751 * find_next_reservable_window():
752 * find a reservable space within the given range.
753 * It does not allocate the reservation window for now:
754 * alloc_new_reservation() will do the work later.
755 *
756 * @search_head: the head of the searching list;
757 * This is not necessarily the list head of the whole filesystem
758 *
759 * We have both head and start_block to assist the search
760 * for the reservable space. The list starts from head,
761 * but we will shift to the place where start_block is,
762 * then start from there, when looking for a reservable space.
763 *
764 * @size: the target new reservation window size
765 *
766 * @group_first_block: the first block we consider to start
767 * the real search from
768 *
769 * @last_block:
770 * the maximum block number that our goal reservable space
771 * could start from. This is normally the last block in this
772 * group. The search will end when we found the start of next
773 * possible reservable space is out of this boundary.
774 * This could handle the cross boundary reservation window
775 * request.
776 *
777 * basically we search from the given range, rather than the whole
778 * reservation double linked list, (start_block, last_block)
779 * to find a free region that is of my size and has not
780 * been reserved.
781 *
782 */
783static int find_next_reservable_window(
784 struct ext2_reserve_window_node *search_head,
785 struct ext2_reserve_window_node *my_rsv,
786 struct super_block * sb,
787 ext2_fsblk_t start_block,
788 ext2_fsblk_t last_block)
789{
790 struct rb_node *next;
791 struct ext2_reserve_window_node *rsv, *prev;
792 ext2_fsblk_t cur;
793 int size = my_rsv->rsv_goal_size;
794
795 /* TODO: make the start of the reservation window byte-aligned */
796 /* cur = *start_block & ~7;*/
797 cur = start_block;
798 rsv = search_head;
799 if (!rsv)
800 return -1;
801
802 while (1) {
803 if (cur <= rsv->rsv_end)
804 cur = rsv->rsv_end + 1;
805
806 /* TODO?
807 * in the case we could not find a reservable space
808 * that is what is expected, during the re-search, we could
809 * remember what's the largest reservable space we could have
810 * and return that one.
811 *
812 * For now it will fail if we could not find the reservable
813 * space with expected-size (or more)...
814 */
815 if (cur > last_block)
816 return -1; /* fail */
817
818 prev = rsv;
819 next = rb_next(&rsv->rsv_node);
820 rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
821
822 /*
823 * Reached the last reservation, we can just append to the
824 * previous one.
825 */
826 if (!next)
827 break;
828
829 if (cur + size <= rsv->rsv_start) {
830 /*
831 * Found a reserveable space big enough. We could
832 * have a reservation across the group boundary here
833 */
834 break;
835 }
836 }
837 /*
838 * we come here either :
839 * when we reach the end of the whole list,
840 * and there is empty reservable space after last entry in the list.
841 * append it to the end of the list.
842 *
843 * or we found one reservable space in the middle of the list,
844 * return the reservation window that we could append to.
845 * succeed.
846 */
847
848 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
849 rsv_window_remove(sb, my_rsv);
850
851 /*
852 * Let's book the whole avaliable window for now. We will check the
853 * disk bitmap later and then, if there are free blocks then we adjust
854 * the window size if it's larger than requested.
855 * Otherwise, we will remove this node from the tree next time
856 * call find_next_reservable_window.
857 */
858 my_rsv->rsv_start = cur;
859 my_rsv->rsv_end = cur + size - 1;
860 my_rsv->rsv_alloc_hit = 0;
861
862 if (prev != my_rsv)
863 ext2_rsv_window_add(sb, my_rsv);
864
865 return 0;
866}
867
868/**
869 * alloc_new_reservation()--allocate a new reservation window
870 *
871 * To make a new reservation, we search part of the filesystem
872 * reservation list (the list that inside the group). We try to
873 * allocate a new reservation window near the allocation goal,
874 * or the beginning of the group, if there is no goal.
875 *
876 * We first find a reservable space after the goal, then from
877 * there, we check the bitmap for the first free block after
878 * it. If there is no free block until the end of group, then the
879 * whole group is full, we failed. Otherwise, check if the free
880 * block is inside the expected reservable space, if so, we
881 * succeed.
882 * If the first free block is outside the reservable space, then
883 * start from the first free block, we search for next available
884 * space, and go on.
885 *
886 * on succeed, a new reservation will be found and inserted into the list
887 * It contains at least one free block, and it does not overlap with other
888 * reservation windows.
889 *
890 * failed: we failed to find a reservation window in this group
891 *
892 * @rsv: the reservation
893 *
894 * @grp_goal: The goal (group-relative). It is where the search for a
895 * free reservable space should start from.
896 * if we have a goal(goal >0 ), then start from there,
897 * no goal(goal = -1), we start from the first block
898 * of the group.
899 *
900 * @sb: the super block
901 * @group: the group we are trying to allocate in
902 * @bitmap_bh: the block group block bitmap
903 *
904 */
905static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
906 ext2_grpblk_t grp_goal, struct super_block *sb,
907 unsigned int group, struct buffer_head *bitmap_bh)
908{
909 struct ext2_reserve_window_node *search_head;
910 ext2_fsblk_t group_first_block, group_end_block, start_block;
911 ext2_grpblk_t first_free_block;
912 struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
913 unsigned long size;
914 int ret;
915 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
916
917 group_first_block = ext2_group_first_block_no(sb, group);
918 group_end_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
919
920 if (grp_goal < 0)
921 start_block = group_first_block;
922 else
923 start_block = grp_goal + group_first_block;
924
925 size = my_rsv->rsv_goal_size;
926
927 if (!rsv_is_empty(&my_rsv->rsv_window)) {
928 /*
929 * if the old reservation is cross group boundary
930 * and if the goal is inside the old reservation window,
931 * we will come here when we just failed to allocate from
932 * the first part of the window. We still have another part
933 * that belongs to the next group. In this case, there is no
934 * point to discard our window and try to allocate a new one
935 * in this group(which will fail). we should
936 * keep the reservation window, just simply move on.
937 *
938 * Maybe we could shift the start block of the reservation
939 * window to the first block of next group.
940 */
941
942 if ((my_rsv->rsv_start <= group_end_block) &&
943 (my_rsv->rsv_end > group_end_block) &&
944 (start_block >= my_rsv->rsv_start))
945 return -1;
946
947 if ((my_rsv->rsv_alloc_hit >
948 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
949 /*
950 * if the previously allocation hit ratio is
951 * greater than 1/2, then we double the size of
952 * the reservation window the next time,
953 * otherwise we keep the same size window
954 */
955 size = size * 2;
956 if (size > EXT2_MAX_RESERVE_BLOCKS)
957 size = EXT2_MAX_RESERVE_BLOCKS;
958 my_rsv->rsv_goal_size= size;
959 }
960 }
961
962 spin_lock(rsv_lock);
963 /*
964 * shift the search start to the window near the goal block
965 */
966 search_head = search_reserve_window(fs_rsv_root, start_block);
967
968 /*
969 * find_next_reservable_window() simply finds a reservable window
970 * inside the given range(start_block, group_end_block).
971 *
972 * To make sure the reservation window has a free bit inside it, we
973 * need to check the bitmap after we found a reservable window.
974 */
975retry:
976 ret = find_next_reservable_window(search_head, my_rsv, sb,
977 start_block, group_end_block);
978
979 if (ret == -1) {
980 if (!rsv_is_empty(&my_rsv->rsv_window))
981 rsv_window_remove(sb, my_rsv);
982 spin_unlock(rsv_lock);
983 return -1;
984 }
985
986 /*
987 * On success, find_next_reservable_window() returns the
988 * reservation window where there is a reservable space after it.
989 * Before we reserve this reservable space, we need
990 * to make sure there is at least a free block inside this region.
991 *
992 * Search the first free bit on the block bitmap. Search starts from
993 * the start block of the reservable space we just found.
994 */
995 spin_unlock(rsv_lock);
996 first_free_block = bitmap_search_next_usable_block(
997 my_rsv->rsv_start - group_first_block,
998 bitmap_bh, group_end_block - group_first_block + 1);
999
1000 if (first_free_block < 0) {
1001 /*
1002 * no free block left on the bitmap, no point
1003 * to reserve the space. return failed.
1004 */
1005 spin_lock(rsv_lock);
1006 if (!rsv_is_empty(&my_rsv->rsv_window))
1007 rsv_window_remove(sb, my_rsv);
1008 spin_unlock(rsv_lock);
1009 return -1; /* failed */
1010 }
1011
1012 start_block = first_free_block + group_first_block;
1013 /*
1014 * check if the first free block is within the
1015 * free space we just reserved
1016 */
1017 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1018 return 0; /* success */
1019 /*
1020 * if the first free bit we found is out of the reservable space
1021 * continue search for next reservable space,
1022 * start from where the free block is,
1023 * we also shift the list head to where we stopped last time
1024 */
1025 search_head = my_rsv;
1026 spin_lock(rsv_lock);
1027 goto retry;
1028}
1029
1030/**
1031 * try_to_extend_reservation()
1032 * @my_rsv: given reservation window
1033 * @sb: super block
1034 * @size: the delta to extend
1035 *
1036 * Attempt to expand the reservation window large enough to have
1037 * required number of free blocks
1038 *
1039 * Since ext2_try_to_allocate() will always allocate blocks within
1040 * the reservation window range, if the window size is too small,
1041 * multiple blocks allocation has to stop at the end of the reservation
1042 * window. To make this more efficient, given the total number of
1043 * blocks needed and the current size of the window, we try to
1044 * expand the reservation window size if necessary on a best-effort
1045 * basis before ext2_new_blocks() tries to allocate blocks.
1046 */
1047static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1048 struct super_block *sb, int size)
1049{
1050 struct ext2_reserve_window_node *next_rsv;
1051 struct rb_node *next;
1052 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1053
1054 if (!spin_trylock(rsv_lock))
1055 return;
1056
1057 next = rb_next(&my_rsv->rsv_node);
1058
1059 if (!next)
1060 my_rsv->rsv_end += size;
1061 else {
1062 next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1063
1064 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1065 my_rsv->rsv_end += size;
1066 else
1067 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1068 }
1069 spin_unlock(rsv_lock);
1070}
1071
1072/**
1073 * ext2_try_to_allocate_with_rsv()
1074 * @sb: superblock
1075 * @group: given allocation block group
1076 * @bitmap_bh: bufferhead holds the block bitmap
1077 * @grp_goal: given target block within the group
1078 * @count: target number of blocks to allocate
1079 * @my_rsv: reservation window
1080 *
1081 * This is the main function used to allocate a new block and its reservation
1082 * window.
1083 *
1084 * Each time when a new block allocation is need, first try to allocate from
1085 * its own reservation. If it does not have a reservation window, instead of
1086 * looking for a free bit on bitmap first, then look up the reservation list to
1087 * see if it is inside somebody else's reservation window, we try to allocate a
1088 * reservation window for it starting from the goal first. Then do the block
1089 * allocation within the reservation window.
1090 *
1091 * This will avoid keeping on searching the reservation list again and
1092 * again when somebody is looking for a free block (without
1093 * reservation), and there are lots of free blocks, but they are all
1094 * being reserved.
1095 *
1096 * We use a red-black tree for the per-filesystem reservation list.
1097 */
1098static ext2_grpblk_t
1099ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1100 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1101 struct ext2_reserve_window_node * my_rsv,
1102 unsigned long *count)
1103{
1104 ext2_fsblk_t group_first_block, group_last_block;
1105 ext2_grpblk_t ret = 0;
1106 unsigned long num = *count;
1107
1108 /*
1109 * we don't deal with reservation when
1110 * filesystem is mounted without reservation
1111 * or the file is not a regular file
1112 * or last attempt to allocate a block with reservation turned on failed
1113 */
1114 if (my_rsv == NULL) {
1115 return ext2_try_to_allocate(sb, group, bitmap_bh,
1116 grp_goal, count, NULL);
1117 }
1118 /*
1119 * grp_goal is a group relative block number (if there is a goal)
1120 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1121 * first block is a filesystem wide block number
1122 * first block is the block number of the first block in this group
1123 */
1124 group_first_block = ext2_group_first_block_no(sb, group);
1125 group_last_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
1126
1127 /*
1128 * Basically we will allocate a new block from inode's reservation
1129 * window.
1130 *
1131 * We need to allocate a new reservation window, if:
1132 * a) inode does not have a reservation window; or
1133 * b) last attempt to allocate a block from existing reservation
1134 * failed; or
1135 * c) we come here with a goal and with a reservation window
1136 *
1137 * We do not need to allocate a new reservation window if we come here
1138 * at the beginning with a goal and the goal is inside the window, or
1139 * we don't have a goal but already have a reservation window.
1140 * then we could go to allocate from the reservation window directly.
1141 */
1142 while (1) {
1143 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1144 !goal_in_my_reservation(&my_rsv->rsv_window,
1145 grp_goal, group, sb)) {
1146 if (my_rsv->rsv_goal_size < *count)
1147 my_rsv->rsv_goal_size = *count;
1148 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1149 group, bitmap_bh);
1150 if (ret < 0)
1151 break; /* failed */
1152
1153 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1154 grp_goal, group, sb))
1155 grp_goal = -1;
1156 } else if (grp_goal >= 0) {
1157 int curr = my_rsv->rsv_end -
1158 (grp_goal + group_first_block) + 1;
1159
1160 if (curr < *count)
1161 try_to_extend_reservation(my_rsv, sb,
1162 *count - curr);
1163 }
1164
1165 if ((my_rsv->rsv_start > group_last_block) ||
1166 (my_rsv->rsv_end < group_first_block)) {
1167 rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1168 BUG();
1169 }
1170 ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1171 &num, &my_rsv->rsv_window);
1172 if (ret >= 0) {
1173 my_rsv->rsv_alloc_hit += num;
1174 *count = num;
1175 break; /* succeed */
1176 }
1177 num = *count;
1178 }
1179 return ret;
1180}
1181
1182/**
1183 * ext2_has_free_blocks()
1184 * @sbi: in-core super block structure.
1185 *
1186 * Check if filesystem has at least 1 free block available for allocation.
1187 */
1188static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1189{
1190 ext2_fsblk_t free_blocks, root_blocks;
1191
1192 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1193 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1194 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1195 sbi->s_resuid != current->fsuid &&
1196 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1197 return 0;
1198 }
1199 return 1;
1200}
1201
1202/*
1203 * ext2_new_blocks() -- core block(s) allocation function
1204 * @inode: file inode
1205 * @goal: given target block(filesystem wide)
1206 * @count: target number of blocks to allocate
1207 * @errp: error code
1208 *
1209 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 * free, or there is a free block within 32 blocks of the goal, that block
1211 * is allocated. Otherwise a forward search is made for a free block; within
1212 * each block group the search first looks for an entire free byte in the block
1213 * bitmap, and then for any free bit if that fails.
1214 * This function also updates quota and i_blocks field.
1215 */
Martin J. Bligha686cd82007-10-16 23:30:46 -07001216ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1217 unsigned long *count, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218{
1219 struct buffer_head *bitmap_bh = NULL;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001220 struct buffer_head *gdp_bh;
1221 int group_no;
1222 int goal_group;
1223 ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1224 ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1225 ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
1226 int bgi; /* blockgroup iteration index */
1227 int performed_allocation = 0;
1228 ext2_grpblk_t free_blocks; /* number of free blocks in a group */
1229 struct super_block *sb;
1230 struct ext2_group_desc *gdp;
1231 struct ext2_super_block *es;
1232 struct ext2_sb_info *sbi;
1233 struct ext2_reserve_window_node *my_rsv = NULL;
1234 struct ext2_block_alloc_info *block_i;
1235 unsigned short windowsz = 0;
1236 unsigned long ngroups;
1237 unsigned long num = *count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Martin J. Bligha686cd82007-10-16 23:30:46 -07001239 *errp = -ENOSPC;
1240 sb = inode->i_sb;
1241 if (!sb) {
1242 printk("ext2_new_blocks: nonexistent device");
1243 return 0;
1244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
Martin J. Bligha686cd82007-10-16 23:30:46 -07001246 /*
1247 * Check quota for allocation of this block.
1248 */
1249 if (DQUOT_ALLOC_BLOCK(inode, num)) {
1250 *errp = -EDQUOT;
1251 return 0;
1252 }
1253
1254 sbi = EXT2_SB(sb);
1255 es = EXT2_SB(sb)->s_es;
1256 ext2_debug("goal=%lu.\n", goal);
1257 /*
1258 * Allocate a block from reservation only when
1259 * filesystem is mounted with reservation(default,-o reservation), and
1260 * it's a regular file, and
1261 * the desired window size is greater than 0 (One could use ioctl
1262 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1263 * reservation on that particular file)
1264 */
1265 block_i = EXT2_I(inode)->i_block_alloc_info;
1266 if (block_i) {
1267 windowsz = block_i->rsv_window_node.rsv_goal_size;
1268 if (windowsz > 0)
1269 my_rsv = &block_i->rsv_window_node;
1270 }
1271
1272 if (!ext2_has_free_blocks(sbi)) {
1273 *errp = -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 goto out;
1275 }
1276
Martin J. Bligha686cd82007-10-16 23:30:46 -07001277 /*
1278 * First, test whether the goal block is free.
1279 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 if (goal < le32_to_cpu(es->s_first_data_block) ||
1281 goal >= le32_to_cpu(es->s_blocks_count))
1282 goal = le32_to_cpu(es->s_first_data_block);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001283 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1284 EXT2_BLOCKS_PER_GROUP(sb);
1285 goal_group = group_no;
1286retry_alloc:
1287 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1288 if (!gdp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 goto io_error;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001290
1291 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1292 /*
1293 * if there is not enough free blocks to make a new resevation
1294 * turn off reservation for this allocation
1295 */
1296 if (my_rsv && (free_blocks < windowsz)
1297 && (rsv_is_empty(&my_rsv->rsv_window)))
1298 my_rsv = NULL;
1299
1300 if (free_blocks > 0) {
1301 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1302 EXT2_BLOCKS_PER_GROUP(sb));
1303 bitmap_bh = read_block_bitmap(sb, group_no);
1304 if (!bitmap_bh)
1305 goto io_error;
1306 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1307 bitmap_bh, grp_target_blk,
1308 my_rsv, &num);
1309 if (grp_alloc_blk >= 0)
1310 goto allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 }
1312
Martin J. Bligha686cd82007-10-16 23:30:46 -07001313 ngroups = EXT2_SB(sb)->s_groups_count;
1314 smp_rmb();
1315
1316 /*
Akinobu Mita144704e2008-02-06 01:40:15 -08001317 * Now search the rest of the groups. We assume that
1318 * group_no and gdp correctly point to the last group visited.
Martin J. Bligha686cd82007-10-16 23:30:46 -07001319 */
1320 for (bgi = 0; bgi < ngroups; bgi++) {
1321 group_no++;
1322 if (group_no >= ngroups)
1323 group_no = 0;
1324 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1325 if (!gdp)
1326 goto io_error;
1327
1328 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1329 /*
1330 * skip this group if the number of
1331 * free blocks is less than half of the reservation
1332 * window size.
1333 */
1334 if (free_blocks <= (windowsz/2))
1335 continue;
1336
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 brelse(bitmap_bh);
1338 bitmap_bh = read_block_bitmap(sb, group_no);
1339 if (!bitmap_bh)
1340 goto io_error;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001341 /*
1342 * try to allocate block(s) from this group, without a goal(-1).
1343 */
1344 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1345 bitmap_bh, -1, my_rsv, &num);
1346 if (grp_alloc_blk >= 0)
1347 goto allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 /*
Martin J. Bligha686cd82007-10-16 23:30:46 -07001350 * We may end up a bogus ealier ENOSPC error due to
1351 * filesystem is "full" of reservations, but
1352 * there maybe indeed free blocks avaliable on disk
1353 * In this case, we just forget about the reservations
1354 * just do block allocation as without reservations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 */
Martin J. Bligha686cd82007-10-16 23:30:46 -07001356 if (my_rsv) {
1357 my_rsv = NULL;
1358 windowsz = 0;
1359 group_no = goal_group;
1360 goto retry_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 }
Martin J. Bligha686cd82007-10-16 23:30:46 -07001362 /* No space left on the device */
1363 *errp = -ENOSPC;
1364 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
Martin J. Bligha686cd82007-10-16 23:30:46 -07001366allocated:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 ext2_debug("using block group %d(%d)\n",
Martin J. Bligha686cd82007-10-16 23:30:46 -07001369 group_no, gdp->bg_free_blocks_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Martin J. Bligha686cd82007-10-16 23:30:46 -07001371 ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
Martin J. Bligha686cd82007-10-16 23:30:46 -07001373 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1374 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1375 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1376 EXT2_SB(sb)->s_itb_per_group) ||
1377 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -08001378 EXT2_SB(sb)->s_itb_per_group)) {
Martin J. Bligha686cd82007-10-16 23:30:46 -07001379 ext2_error(sb, "ext2_new_blocks",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 "Allocating block in system zone - "
Martin J. Bligha686cd82007-10-16 23:30:46 -07001381 "blocks from "E2FSBLK", length %lu",
1382 ret_block, num);
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -08001383 goto out;
1384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Martin J. Bligha686cd82007-10-16 23:30:46 -07001386 performed_allocation = 1;
1387
1388 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1389 ext2_error(sb, "ext2_new_blocks",
1390 "block("E2FSBLK") >= blocks count(%d) - "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 "block_group = %d, es == %p ", ret_block,
1392 le32_to_cpu(es->s_blocks_count), group_no, es);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001393 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
Martin J. Bligha686cd82007-10-16 23:30:46 -07001396 group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1397 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
1399 mark_buffer_dirty(bitmap_bh);
1400 if (sb->s_flags & MS_SYNCHRONOUS)
1401 sync_dirty_buffer(bitmap_bh);
1402
Martin J. Bligha686cd82007-10-16 23:30:46 -07001403 *errp = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 brelse(bitmap_bh);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001405 DQUOT_FREE_BLOCK(inode, *count-num);
1406 *count = num;
1407 return ret_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
1409io_error:
Martin J. Bligha686cd82007-10-16 23:30:46 -07001410 *errp = -EIO;
1411out:
1412 /*
1413 * Undo the block allocation
1414 */
1415 if (!performed_allocation)
1416 DQUOT_FREE_BLOCK(inode, *count);
1417 brelse(bitmap_bh);
1418 return 0;
1419}
1420
1421ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1422{
1423 unsigned long count = 1;
1424
1425 return ext2_new_blocks(inode, goal, &count, errp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426}
1427
Valerie Henson21730ee2006-06-25 05:48:12 -07001428#ifdef EXT2FS_DEBUG
1429
Philippe De Muyterfebfcf92007-10-16 23:26:15 -07001430static 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 -07001431
1432unsigned long ext2_count_free (struct buffer_head * map, unsigned int numchars)
1433{
1434 unsigned int i;
1435 unsigned long sum = 0;
1436
1437 if (!map)
1438 return (0);
1439 for (i = 0; i < numchars; i++)
1440 sum += nibblemap[map->b_data[i] & 0xf] +
1441 nibblemap[(map->b_data[i] >> 4) & 0xf];
1442 return (sum);
1443}
1444
1445#endif /* EXT2FS_DEBUG */
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447unsigned long ext2_count_free_blocks (struct super_block * sb)
1448{
1449 struct ext2_group_desc * desc;
1450 unsigned long desc_count = 0;
1451 int i;
1452#ifdef EXT2FS_DEBUG
1453 unsigned long bitmap_count, x;
1454 struct ext2_super_block *es;
1455
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 es = EXT2_SB(sb)->s_es;
1457 desc_count = 0;
1458 bitmap_count = 0;
1459 desc = NULL;
1460 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1461 struct buffer_head *bitmap_bh;
1462 desc = ext2_get_group_desc (sb, i, NULL);
1463 if (!desc)
1464 continue;
1465 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1466 bitmap_bh = read_block_bitmap(sb, i);
1467 if (!bitmap_bh)
1468 continue;
1469
1470 x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1471 printk ("group %d: stored = %d, counted = %lu\n",
1472 i, le16_to_cpu(desc->bg_free_blocks_count), x);
1473 bitmap_count += x;
1474 brelse(bitmap_bh);
1475 }
1476 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1477 (long)le32_to_cpu(es->s_free_blocks_count),
1478 desc_count, bitmap_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 return bitmap_count;
1480#else
1481 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1482 desc = ext2_get_group_desc (sb, i, NULL);
1483 if (!desc)
1484 continue;
1485 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1486 }
1487 return desc_count;
1488#endif
1489}
1490
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491static inline int test_root(int a, int b)
1492{
1493 int num = b;
1494
1495 while (a > num)
1496 num *= b;
1497 return num == a;
1498}
1499
1500static int ext2_group_sparse(int group)
1501{
1502 if (group <= 1)
1503 return 1;
1504 return (test_root(group, 3) || test_root(group, 5) ||
1505 test_root(group, 7));
1506}
1507
1508/**
1509 * ext2_bg_has_super - number of blocks used by the superblock in group
1510 * @sb: superblock for filesystem
1511 * @group: group number to check
1512 *
1513 * Return the number of blocks used by the superblock (primary or backup)
1514 * in this group. Currently this will be only 0 or 1.
1515 */
1516int ext2_bg_has_super(struct super_block *sb, int group)
1517{
1518 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1519 !ext2_group_sparse(group))
1520 return 0;
1521 return 1;
1522}
1523
1524/**
1525 * ext2_bg_num_gdb - number of blocks used by the group table in group
1526 * @sb: superblock for filesystem
1527 * @group: group number to check
1528 *
1529 * Return the number of blocks used by the group descriptor table
1530 * (primary or backup) in this group. In the future there may be a
1531 * different number of descriptor blocks in each group.
1532 */
1533unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1534{
Akinobu Mita859cb932008-02-06 01:40:17 -08001535 return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536}
1537