blob: 3cf038c055d7b587290b05bd659624985e9d1056 [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090016#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/sched.h>
18#include <linux/buffer_head.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080019#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
21/*
22 * balloc.c contains the blocks allocation and deallocation routines
23 */
24
25/*
26 * The free blocks are managed by bitmaps. A file system contains several
27 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
28 * block for inodes, N blocks for the inode table and data blocks.
29 *
30 * The file system contains group descriptors which are located after the
31 * super block. Each descriptor contains the number of the bitmap block and
32 * the free blocks count in the block. The descriptors are loaded in memory
Aneesh Kumar K.Ve6274322007-02-20 13:57:58 -080033 * when a file system is mounted (see ext2_fill_super).
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 */
35
36
37#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
38
39struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
40 unsigned int block_group,
41 struct buffer_head ** bh)
42{
43 unsigned long group_desc;
44 unsigned long offset;
45 struct ext2_group_desc * desc;
46 struct ext2_sb_info *sbi = EXT2_SB(sb);
47
48 if (block_group >= sbi->s_groups_count) {
49 ext2_error (sb, "ext2_get_group_desc",
50 "block_group >= groups_count - "
51 "block_group = %d, groups_count = %lu",
52 block_group, sbi->s_groups_count);
53
54 return NULL;
55 }
56
57 group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
58 offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
59 if (!sbi->s_group_desc[group_desc]) {
60 ext2_error (sb, "ext2_get_group_desc",
61 "Group descriptor not loaded - "
62 "block_group = %d, group_desc = %lu, desc = %lu",
63 block_group, group_desc, offset);
64 return NULL;
65 }
66
67 desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
68 if (bh)
69 *bh = sbi->s_group_desc[group_desc];
70 return desc + offset;
71}
72
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -080073static int ext2_valid_block_bitmap(struct super_block *sb,
74 struct ext2_group_desc *desc,
75 unsigned int block_group,
76 struct buffer_head *bh)
77{
78 ext2_grpblk_t offset;
79 ext2_grpblk_t next_zero_bit;
80 ext2_fsblk_t bitmap_blk;
81 ext2_fsblk_t group_first_block;
82
83 group_first_block = ext2_group_first_block_no(sb, block_group);
84
85 /* check whether block bitmap block number is set */
86 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
87 offset = bitmap_blk - group_first_block;
88 if (!ext2_test_bit(offset, bh->b_data))
89 /* bad block bitmap */
90 goto err_out;
91
92 /* check whether the inode bitmap block number is set */
93 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
94 offset = bitmap_blk - group_first_block;
95 if (!ext2_test_bit(offset, bh->b_data))
96 /* bad block bitmap */
97 goto err_out;
98
99 /* check whether the inode table block number is set */
100 bitmap_blk = le32_to_cpu(desc->bg_inode_table);
101 offset = bitmap_blk - group_first_block;
102 next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
103 offset + EXT2_SB(sb)->s_itb_per_group,
104 offset);
105 if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
106 /* good bitmap for inode tables */
107 return 1;
108
109err_out:
Harvey Harrison605afd62008-04-28 02:16:03 -0700110 ext2_error(sb, __func__,
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800111 "Invalid block bitmap - "
112 "block_group = %d, block = %lu",
113 block_group, bitmap_blk);
114 return 0;
115}
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117/*
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800118 * Read the bitmap for a given block_group,and validate the
119 * bits for block/inode/inode tables are set in the bitmaps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 *
121 * Return buffer_head on success or NULL in case of failure.
122 */
123static struct buffer_head *
124read_block_bitmap(struct super_block *sb, unsigned int block_group)
125{
126 struct ext2_group_desc * desc;
127 struct buffer_head * bh = NULL;
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800128 ext2_fsblk_t bitmap_blk;
129
130 desc = ext2_get_group_desc(sb, block_group, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 if (!desc)
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800132 return NULL;
133 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
134 bh = sb_getblk(sb, bitmap_blk);
135 if (unlikely(!bh)) {
Harvey Harrison605afd62008-04-28 02:16:03 -0700136 ext2_error(sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 "Cannot read block bitmap - "
138 "block_group = %d, block_bitmap = %u",
139 block_group, le32_to_cpu(desc->bg_block_bitmap));
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800140 return NULL;
141 }
142 if (likely(bh_uptodate_or_lock(bh)))
143 return bh;
144
145 if (bh_submit_read(bh) < 0) {
146 brelse(bh);
Harvey Harrison605afd62008-04-28 02:16:03 -0700147 ext2_error(sb, __func__,
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800148 "Cannot read block bitmap - "
149 "block_group = %d, block_bitmap = %u",
150 block_group, le32_to_cpu(desc->bg_block_bitmap));
151 return NULL;
152 }
Aneesh Kumar K.V01584fa2008-02-06 01:40:08 -0800153
Aneesh Kumar K.V8b915822008-04-28 02:16:04 -0700154 ext2_valid_block_bitmap(sb, desc, block_group, bh);
155 /*
156 * file system mounted not to panic on error, continue with corrupt
157 * bitmap
158 */
Linus Torvalds0b832a42007-11-13 08:07:31 -0800159 return bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160}
161
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162static void release_blocks(struct super_block *sb, int count)
163{
164 if (count) {
165 struct ext2_sb_info *sbi = EXT2_SB(sb);
166
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700167 percpu_counter_add(&sbi->s_freeblocks_counter, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 sb->s_dirt = 1;
169 }
170}
171
Martin J. Bligha686cd82007-10-16 23:30:46 -0700172static void group_adjust_blocks(struct super_block *sb, int group_no,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 struct ext2_group_desc *desc, struct buffer_head *bh, int count)
174{
175 if (count) {
176 struct ext2_sb_info *sbi = EXT2_SB(sb);
177 unsigned free_blocks;
178
179 spin_lock(sb_bgl_lock(sbi, group_no));
180 free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
181 desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
182 spin_unlock(sb_bgl_lock(sbi, group_no));
183 sb->s_dirt = 1;
184 mark_buffer_dirty(bh);
185 }
186}
187
Martin J. Bligha686cd82007-10-16 23:30:46 -0700188/*
189 * The reservation window structure operations
190 * --------------------------------------------
191 * Operations include:
192 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
193 *
194 * We use a red-black tree to represent per-filesystem reservation
195 * windows.
196 *
197 */
198
199/**
200 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
201 * @rb_root: root of per-filesystem reservation rb tree
202 * @verbose: verbose mode
203 * @fn: function which wishes to dump the reservation map
204 *
205 * If verbose is turned on, it will print the whole block reservation
206 * windows(start, end). Otherwise, it will only print out the "bad" windows,
207 * those windows that overlap with their immediate neighbors.
208 */
209#if 1
210static void __rsv_window_dump(struct rb_root *root, int verbose,
211 const char *fn)
212{
213 struct rb_node *n;
214 struct ext2_reserve_window_node *rsv, *prev;
215 int bad;
216
217restart:
218 n = rb_first(root);
219 bad = 0;
220 prev = NULL;
221
222 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
223 while (n) {
224 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
225 if (verbose)
226 printk("reservation window 0x%p "
227 "start: %lu, end: %lu\n",
228 rsv, rsv->rsv_start, rsv->rsv_end);
229 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
230 printk("Bad reservation %p (start >= end)\n",
231 rsv);
232 bad = 1;
233 }
234 if (prev && prev->rsv_end >= rsv->rsv_start) {
235 printk("Bad reservation %p (prev->end >= start)\n",
236 rsv);
237 bad = 1;
238 }
239 if (bad) {
240 if (!verbose) {
241 printk("Restarting reservation walk in verbose mode\n");
242 verbose = 1;
243 goto restart;
244 }
245 }
246 n = rb_next(n);
247 prev = rsv;
248 }
249 printk("Window map complete.\n");
Julia Lawall2c116192008-04-28 02:16:02 -0700250 BUG_ON(bad);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700251}
252#define rsv_window_dump(root, verbose) \
Harvey Harrison605afd62008-04-28 02:16:03 -0700253 __rsv_window_dump((root), (verbose), __func__)
Martin J. Bligha686cd82007-10-16 23:30:46 -0700254#else
255#define rsv_window_dump(root, verbose) do {} while (0)
256#endif
257
258/**
259 * goal_in_my_reservation()
260 * @rsv: inode's reservation window
261 * @grp_goal: given goal block relative to the allocation block group
262 * @group: the current allocation block group
263 * @sb: filesystem super block
264 *
265 * Test if the given goal block (group relative) is within the file's
266 * own block reservation window range.
267 *
268 * If the reservation window is outside the goal allocation group, return 0;
269 * grp_goal (given goal block) could be -1, which means no specific
270 * goal block. In this case, always return 1.
271 * If the goal block is within the reservation window, return 1;
272 * otherwise, return 0;
273 */
274static int
275goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
276 unsigned int group, struct super_block * sb)
277{
278 ext2_fsblk_t group_first_block, group_last_block;
279
280 group_first_block = ext2_group_first_block_no(sb, group);
281 group_last_block = group_first_block + EXT2_BLOCKS_PER_GROUP(sb) - 1;
282
283 if ((rsv->_rsv_start > group_last_block) ||
284 (rsv->_rsv_end < group_first_block))
285 return 0;
286 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
287 || (grp_goal + group_first_block > rsv->_rsv_end)))
288 return 0;
289 return 1;
290}
291
292/**
293 * search_reserve_window()
294 * @rb_root: root of reservation tree
295 * @goal: target allocation block
296 *
297 * Find the reserved window which includes the goal, or the previous one
298 * if the goal is not in any window.
299 * Returns NULL if there are no windows or if all windows start after the goal.
300 */
301static struct ext2_reserve_window_node *
302search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
303{
304 struct rb_node *n = root->rb_node;
305 struct ext2_reserve_window_node *rsv;
306
307 if (!n)
308 return NULL;
309
310 do {
311 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
312
313 if (goal < rsv->rsv_start)
314 n = n->rb_left;
315 else if (goal > rsv->rsv_end)
316 n = n->rb_right;
317 else
318 return rsv;
319 } while (n);
320 /*
321 * We've fallen off the end of the tree: the goal wasn't inside
322 * any particular node. OK, the previous node must be to one
323 * side of the interval containing the goal. If it's the RHS,
324 * we need to back up one.
325 */
326 if (rsv->rsv_start > goal) {
327 n = rb_prev(&rsv->rsv_node);
328 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
329 }
330 return rsv;
331}
332
333/*
334 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
335 * @sb: super block
336 * @rsv: reservation window to add
337 *
338 * Must be called with rsv_lock held.
339 */
340void ext2_rsv_window_add(struct super_block *sb,
341 struct ext2_reserve_window_node *rsv)
342{
343 struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
344 struct rb_node *node = &rsv->rsv_node;
345 ext2_fsblk_t start = rsv->rsv_start;
346
347 struct rb_node ** p = &root->rb_node;
348 struct rb_node * parent = NULL;
349 struct ext2_reserve_window_node *this;
350
351 while (*p)
352 {
353 parent = *p;
354 this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
355
356 if (start < this->rsv_start)
357 p = &(*p)->rb_left;
358 else if (start > this->rsv_end)
359 p = &(*p)->rb_right;
360 else {
361 rsv_window_dump(root, 1);
362 BUG();
363 }
364 }
365
366 rb_link_node(node, parent, p);
367 rb_insert_color(node, root);
368}
369
370/**
371 * rsv_window_remove() -- unlink a window from the reservation rb tree
372 * @sb: super block
373 * @rsv: reservation window to remove
374 *
375 * Mark the block reservation window as not allocated, and unlink it
376 * from the filesystem reservation window rb tree. Must be called with
377 * rsv_lock held.
378 */
379static void rsv_window_remove(struct super_block *sb,
380 struct ext2_reserve_window_node *rsv)
381{
382 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
383 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
384 rsv->rsv_alloc_hit = 0;
385 rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
386}
387
388/*
389 * rsv_is_empty() -- Check if the reservation window is allocated.
390 * @rsv: given reservation window to check
391 *
392 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
393 */
394static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
395{
396 /* a valid reservation end block could not be 0 */
397 return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
398}
399
400/**
401 * ext2_init_block_alloc_info()
402 * @inode: file inode structure
403 *
404 * Allocate and initialize the reservation window structure, and
405 * link the window to the ext2 inode structure at last
406 *
407 * The reservation window structure is only dynamically allocated
408 * and linked to ext2 inode the first time the open file
409 * needs a new block. So, before every ext2_new_block(s) call, for
410 * regular files, we should check whether the reservation window
411 * structure exists or not. In the latter case, this function is called.
412 * Fail to do so will result in block reservation being turned off for that
413 * open file.
414 *
415 * This function is called from ext2_get_blocks_handle(), also called
416 * when setting the reservation window size through ioctl before the file
417 * is open for write (needs block allocation).
418 *
419 * Needs truncate_mutex protection prior to calling this function.
420 */
421void ext2_init_block_alloc_info(struct inode *inode)
422{
423 struct ext2_inode_info *ei = EXT2_I(inode);
424 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
425 struct super_block *sb = inode->i_sb;
426
427 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
428 if (block_i) {
429 struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
430
431 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
432 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
433
434 /*
435 * if filesystem is mounted with NORESERVATION, the goal
436 * reservation window size is set to zero to indicate
437 * block reservation is off
438 */
439 if (!test_opt(sb, RESERVATION))
440 rsv->rsv_goal_size = 0;
441 else
442 rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
443 rsv->rsv_alloc_hit = 0;
444 block_i->last_alloc_logical_block = 0;
445 block_i->last_alloc_physical_block = 0;
446 }
447 ei->i_block_alloc_info = block_i;
448}
449
450/**
451 * ext2_discard_reservation()
452 * @inode: inode
453 *
454 * Discard(free) block reservation window on last file close, or truncate
455 * or at last iput().
456 *
457 * It is being called in three cases:
458 * ext2_release_file(): last writer closes the file
459 * ext2_clear_inode(): last iput(), when nobody links to this file.
460 * ext2_truncate(): when the block indirect map is about to change.
461 */
462void ext2_discard_reservation(struct inode *inode)
463{
464 struct ext2_inode_info *ei = EXT2_I(inode);
465 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
466 struct ext2_reserve_window_node *rsv;
467 spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
468
469 if (!block_i)
470 return;
471
472 rsv = &block_i->rsv_window_node;
473 if (!rsv_is_empty(&rsv->rsv_window)) {
474 spin_lock(rsv_lock);
475 if (!rsv_is_empty(&rsv->rsv_window))
476 rsv_window_remove(inode->i_sb, rsv);
477 spin_unlock(rsv_lock);
478 }
479}
480
481/**
482 * ext2_free_blocks_sb() -- Free given blocks and update quota and i_blocks
483 * @inode: inode
484 * @block: start physcial block to free
485 * @count: number of blocks to free
486 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487void ext2_free_blocks (struct inode * inode, unsigned long block,
488 unsigned long count)
489{
490 struct buffer_head *bitmap_bh = NULL;
491 struct buffer_head * bh2;
492 unsigned long block_group;
493 unsigned long bit;
494 unsigned long i;
495 unsigned long overflow;
496 struct super_block * sb = inode->i_sb;
497 struct ext2_sb_info * sbi = EXT2_SB(sb);
498 struct ext2_group_desc * desc;
499 struct ext2_super_block * es = sbi->s_es;
500 unsigned freed = 0, group_freed;
501
502 if (block < le32_to_cpu(es->s_first_data_block) ||
503 block + count < block ||
504 block + count > le32_to_cpu(es->s_blocks_count)) {
505 ext2_error (sb, "ext2_free_blocks",
506 "Freeing blocks not in datazone - "
507 "block = %lu, count = %lu", block, count);
508 goto error_return;
509 }
510
511 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
512
513do_more:
514 overflow = 0;
515 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
516 EXT2_BLOCKS_PER_GROUP(sb);
517 bit = (block - le32_to_cpu(es->s_first_data_block)) %
518 EXT2_BLOCKS_PER_GROUP(sb);
519 /*
520 * Check to see if we are freeing blocks across a group
521 * boundary.
522 */
523 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
524 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
525 count -= overflow;
526 }
527 brelse(bitmap_bh);
528 bitmap_bh = read_block_bitmap(sb, block_group);
529 if (!bitmap_bh)
530 goto error_return;
531
532 desc = ext2_get_group_desc (sb, block_group, &bh2);
533 if (!desc)
534 goto error_return;
535
536 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
537 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
538 in_range (block, le32_to_cpu(desc->bg_inode_table),
539 sbi->s_itb_per_group) ||
540 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -0800541 sbi->s_itb_per_group)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 ext2_error (sb, "ext2_free_blocks",
543 "Freeing blocks in system zones - "
544 "Block = %lu, count = %lu",
545 block, count);
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -0800546 goto error_return;
547 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
549 for (i = 0, group_freed = 0; i < count; i++) {
550 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
551 bit + i, bitmap_bh->b_data)) {
Harvey Harrison605afd62008-04-28 02:16:03 -0700552 ext2_error(sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 "bit already cleared for block %lu", block + i);
554 } else {
555 group_freed++;
556 }
557 }
558
559 mark_buffer_dirty(bitmap_bh);
560 if (sb->s_flags & MS_SYNCHRONOUS)
561 sync_dirty_buffer(bitmap_bh);
562
Martin J. Bligha686cd82007-10-16 23:30:46 -0700563 group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 freed += group_freed;
565
566 if (overflow) {
567 block += count;
568 count = overflow;
569 goto do_more;
570 }
571error_return:
572 brelse(bitmap_bh);
573 release_blocks(sb, freed);
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500574 dquot_free_block(inode, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575}
576
Martin J. Bligha686cd82007-10-16 23:30:46 -0700577/**
578 * bitmap_search_next_usable_block()
579 * @start: the starting block (group relative) of the search
580 * @bh: bufferhead contains the block group bitmap
581 * @maxblocks: the ending block (group relative) of the reservation
582 *
583 * The bitmap search --- search forward through the actual bitmap on disk until
584 * we find a bit free.
585 */
586static ext2_grpblk_t
587bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
588 ext2_grpblk_t maxblocks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
Martin J. Bligha686cd82007-10-16 23:30:46 -0700590 ext2_grpblk_t next;
591
592 next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
593 if (next >= maxblocks)
594 return -1;
595 return next;
596}
597
598/**
599 * find_next_usable_block()
600 * @start: the starting block (group relative) to find next
601 * allocatable block in bitmap.
602 * @bh: bufferhead contains the block group bitmap
603 * @maxblocks: the ending block (group relative) for the search
604 *
605 * Find an allocatable block in a bitmap. We perform the "most
606 * appropriate allocation" algorithm of looking for a free block near
607 * the initial goal; then for a free byte somewhere in the bitmap;
608 * then for any free bit in the bitmap.
609 */
610static ext2_grpblk_t
611find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
612{
613 ext2_grpblk_t here, next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 char *p, *r;
615
Martin J. Bligha686cd82007-10-16 23:30:46 -0700616 if (start > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 /*
618 * The goal was occupied; search forward for a free
619 * block within the next XX blocks.
620 *
621 * end_goal is more or less random, but it has to be
622 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
623 * next 64-bit boundary is simple..
624 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700625 ext2_grpblk_t end_goal = (start + 63) & ~63;
626 if (end_goal > maxblocks)
627 end_goal = maxblocks;
628 here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
629 if (here < end_goal)
630 return here;
631 ext2_debug("Bit not found near goal\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 }
633
Martin J. Bligha686cd82007-10-16 23:30:46 -0700634 here = start;
635 if (here < 0)
636 here = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
Martin J. Bligha686cd82007-10-16 23:30:46 -0700638 p = ((char *)bh->b_data) + (here >> 3);
639 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
640 next = (r - ((char *)bh->b_data)) << 3;
641
642 if (next < maxblocks && next >= here)
643 return next;
644
645 here = bitmap_search_next_usable_block(here, bh, maxblocks);
646 return here;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
649/*
Martin J. Bligha686cd82007-10-16 23:30:46 -0700650 * ext2_try_to_allocate()
651 * @sb: superblock
652 * @handle: handle to this transaction
653 * @group: given allocation block group
654 * @bitmap_bh: bufferhead holds the block bitmap
655 * @grp_goal: given target block within the group
656 * @count: target number of blocks to allocate
657 * @my_rsv: reservation window
658 *
659 * Attempt to allocate blocks within a give range. Set the range of allocation
660 * first, then find the first free bit(s) from the bitmap (within the range),
661 * and at last, allocate the blocks by claiming the found free bit as allocated.
662 *
663 * To set the range of this allocation:
664 * if there is a reservation window, only try to allocate block(s)
665 * from the file's own reservation window;
666 * Otherwise, the allocation range starts from the give goal block,
667 * ends at the block group's last block.
668 *
669 * If we failed to allocate the desired block then we may end up crossing to a
670 * new bitmap.
671 */
672static int
673ext2_try_to_allocate(struct super_block *sb, int group,
674 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
675 unsigned long *count,
676 struct ext2_reserve_window *my_rsv)
677{
678 ext2_fsblk_t group_first_block;
679 ext2_grpblk_t start, end;
680 unsigned long num = 0;
681
682 /* we do allocation within the reservation window if we have a window */
683 if (my_rsv) {
684 group_first_block = ext2_group_first_block_no(sb, group);
685 if (my_rsv->_rsv_start >= group_first_block)
686 start = my_rsv->_rsv_start - group_first_block;
687 else
688 /* reservation window cross group boundary */
689 start = 0;
690 end = my_rsv->_rsv_end - group_first_block + 1;
691 if (end > EXT2_BLOCKS_PER_GROUP(sb))
692 /* reservation window crosses group boundary */
693 end = EXT2_BLOCKS_PER_GROUP(sb);
694 if ((start <= grp_goal) && (grp_goal < end))
695 start = grp_goal;
696 else
697 grp_goal = -1;
698 } else {
699 if (grp_goal > 0)
700 start = grp_goal;
701 else
702 start = 0;
703 end = EXT2_BLOCKS_PER_GROUP(sb);
704 }
705
706 BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
707
708repeat:
709 if (grp_goal < 0) {
710 grp_goal = find_next_usable_block(start, bitmap_bh, end);
711 if (grp_goal < 0)
712 goto fail_access;
713 if (!my_rsv) {
714 int i;
715
716 for (i = 0; i < 7 && grp_goal > start &&
717 !ext2_test_bit(grp_goal - 1,
718 bitmap_bh->b_data);
719 i++, grp_goal--)
720 ;
721 }
722 }
723 start = grp_goal;
724
725 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), grp_goal,
726 bitmap_bh->b_data)) {
727 /*
728 * The block was allocated by another thread, or it was
729 * allocated and then freed by another thread
730 */
731 start++;
732 grp_goal++;
733 if (start >= end)
734 goto fail_access;
735 goto repeat;
736 }
737 num++;
738 grp_goal++;
739 while (num < *count && grp_goal < end
740 && !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
741 grp_goal, bitmap_bh->b_data)) {
742 num++;
743 grp_goal++;
744 }
745 *count = num;
746 return grp_goal - num;
747fail_access:
748 *count = num;
749 return -1;
750}
751
752/**
753 * find_next_reservable_window():
754 * find a reservable space within the given range.
755 * It does not allocate the reservation window for now:
756 * alloc_new_reservation() will do the work later.
757 *
758 * @search_head: the head of the searching list;
759 * This is not necessarily the list head of the whole filesystem
760 *
761 * We have both head and start_block to assist the search
762 * for the reservable space. The list starts from head,
763 * but we will shift to the place where start_block is,
764 * then start from there, when looking for a reservable space.
765 *
766 * @size: the target new reservation window size
767 *
768 * @group_first_block: the first block we consider to start
769 * the real search from
770 *
771 * @last_block:
772 * the maximum block number that our goal reservable space
773 * could start from. This is normally the last block in this
774 * group. The search will end when we found the start of next
775 * possible reservable space is out of this boundary.
776 * This could handle the cross boundary reservation window
777 * request.
778 *
779 * basically we search from the given range, rather than the whole
780 * reservation double linked list, (start_block, last_block)
781 * to find a free region that is of my size and has not
782 * been reserved.
783 *
784 */
785static int find_next_reservable_window(
786 struct ext2_reserve_window_node *search_head,
787 struct ext2_reserve_window_node *my_rsv,
788 struct super_block * sb,
789 ext2_fsblk_t start_block,
790 ext2_fsblk_t last_block)
791{
792 struct rb_node *next;
793 struct ext2_reserve_window_node *rsv, *prev;
794 ext2_fsblk_t cur;
795 int size = my_rsv->rsv_goal_size;
796
797 /* TODO: make the start of the reservation window byte-aligned */
798 /* cur = *start_block & ~7;*/
799 cur = start_block;
800 rsv = search_head;
801 if (!rsv)
802 return -1;
803
804 while (1) {
805 if (cur <= rsv->rsv_end)
806 cur = rsv->rsv_end + 1;
807
808 /* TODO?
809 * in the case we could not find a reservable space
810 * that is what is expected, during the re-search, we could
811 * remember what's the largest reservable space we could have
812 * and return that one.
813 *
814 * For now it will fail if we could not find the reservable
815 * space with expected-size (or more)...
816 */
817 if (cur > last_block)
818 return -1; /* fail */
819
820 prev = rsv;
821 next = rb_next(&rsv->rsv_node);
822 rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
823
824 /*
825 * Reached the last reservation, we can just append to the
826 * previous one.
827 */
828 if (!next)
829 break;
830
831 if (cur + size <= rsv->rsv_start) {
832 /*
833 * Found a reserveable space big enough. We could
834 * have a reservation across the group boundary here
835 */
836 break;
837 }
838 }
839 /*
840 * we come here either :
841 * when we reach the end of the whole list,
842 * and there is empty reservable space after last entry in the list.
843 * append it to the end of the list.
844 *
845 * or we found one reservable space in the middle of the list,
846 * return the reservation window that we could append to.
847 * succeed.
848 */
849
850 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
851 rsv_window_remove(sb, my_rsv);
852
853 /*
854 * Let's book the whole avaliable window for now. We will check the
855 * disk bitmap later and then, if there are free blocks then we adjust
856 * the window size if it's larger than requested.
857 * Otherwise, we will remove this node from the tree next time
858 * call find_next_reservable_window.
859 */
860 my_rsv->rsv_start = cur;
861 my_rsv->rsv_end = cur + size - 1;
862 my_rsv->rsv_alloc_hit = 0;
863
864 if (prev != my_rsv)
865 ext2_rsv_window_add(sb, my_rsv);
866
867 return 0;
868}
869
870/**
871 * alloc_new_reservation()--allocate a new reservation window
872 *
873 * To make a new reservation, we search part of the filesystem
874 * reservation list (the list that inside the group). We try to
875 * allocate a new reservation window near the allocation goal,
876 * or the beginning of the group, if there is no goal.
877 *
878 * We first find a reservable space after the goal, then from
879 * there, we check the bitmap for the first free block after
880 * it. If there is no free block until the end of group, then the
881 * whole group is full, we failed. Otherwise, check if the free
882 * block is inside the expected reservable space, if so, we
883 * succeed.
884 * If the first free block is outside the reservable space, then
885 * start from the first free block, we search for next available
886 * space, and go on.
887 *
888 * on succeed, a new reservation will be found and inserted into the list
889 * It contains at least one free block, and it does not overlap with other
890 * reservation windows.
891 *
892 * failed: we failed to find a reservation window in this group
893 *
894 * @rsv: the reservation
895 *
896 * @grp_goal: The goal (group-relative). It is where the search for a
897 * free reservable space should start from.
898 * if we have a goal(goal >0 ), then start from there,
899 * no goal(goal = -1), we start from the first block
900 * of the group.
901 *
902 * @sb: the super block
903 * @group: the group we are trying to allocate in
904 * @bitmap_bh: the block group block bitmap
905 *
906 */
907static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
908 ext2_grpblk_t grp_goal, struct super_block *sb,
909 unsigned int group, struct buffer_head *bitmap_bh)
910{
911 struct ext2_reserve_window_node *search_head;
912 ext2_fsblk_t group_first_block, group_end_block, start_block;
913 ext2_grpblk_t first_free_block;
914 struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
915 unsigned long size;
916 int ret;
917 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
918
919 group_first_block = ext2_group_first_block_no(sb, group);
920 group_end_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
921
922 if (grp_goal < 0)
923 start_block = group_first_block;
924 else
925 start_block = grp_goal + group_first_block;
926
927 size = my_rsv->rsv_goal_size;
928
929 if (!rsv_is_empty(&my_rsv->rsv_window)) {
930 /*
931 * if the old reservation is cross group boundary
932 * and if the goal is inside the old reservation window,
933 * we will come here when we just failed to allocate from
934 * the first part of the window. We still have another part
935 * that belongs to the next group. In this case, there is no
936 * point to discard our window and try to allocate a new one
937 * in this group(which will fail). we should
938 * keep the reservation window, just simply move on.
939 *
940 * Maybe we could shift the start block of the reservation
941 * window to the first block of next group.
942 */
943
944 if ((my_rsv->rsv_start <= group_end_block) &&
945 (my_rsv->rsv_end > group_end_block) &&
946 (start_block >= my_rsv->rsv_start))
947 return -1;
948
949 if ((my_rsv->rsv_alloc_hit >
950 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
951 /*
952 * if the previously allocation hit ratio is
953 * greater than 1/2, then we double the size of
954 * the reservation window the next time,
955 * otherwise we keep the same size window
956 */
957 size = size * 2;
958 if (size > EXT2_MAX_RESERVE_BLOCKS)
959 size = EXT2_MAX_RESERVE_BLOCKS;
960 my_rsv->rsv_goal_size= size;
961 }
962 }
963
964 spin_lock(rsv_lock);
965 /*
966 * shift the search start to the window near the goal block
967 */
968 search_head = search_reserve_window(fs_rsv_root, start_block);
969
970 /*
971 * find_next_reservable_window() simply finds a reservable window
972 * inside the given range(start_block, group_end_block).
973 *
974 * To make sure the reservation window has a free bit inside it, we
975 * need to check the bitmap after we found a reservable window.
976 */
977retry:
978 ret = find_next_reservable_window(search_head, my_rsv, sb,
979 start_block, group_end_block);
980
981 if (ret == -1) {
982 if (!rsv_is_empty(&my_rsv->rsv_window))
983 rsv_window_remove(sb, my_rsv);
984 spin_unlock(rsv_lock);
985 return -1;
986 }
987
988 /*
989 * On success, find_next_reservable_window() returns the
990 * reservation window where there is a reservable space after it.
991 * Before we reserve this reservable space, we need
992 * to make sure there is at least a free block inside this region.
993 *
994 * Search the first free bit on the block bitmap. Search starts from
995 * the start block of the reservable space we just found.
996 */
997 spin_unlock(rsv_lock);
998 first_free_block = bitmap_search_next_usable_block(
999 my_rsv->rsv_start - group_first_block,
1000 bitmap_bh, group_end_block - group_first_block + 1);
1001
1002 if (first_free_block < 0) {
1003 /*
1004 * no free block left on the bitmap, no point
1005 * to reserve the space. return failed.
1006 */
1007 spin_lock(rsv_lock);
1008 if (!rsv_is_empty(&my_rsv->rsv_window))
1009 rsv_window_remove(sb, my_rsv);
1010 spin_unlock(rsv_lock);
1011 return -1; /* failed */
1012 }
1013
1014 start_block = first_free_block + group_first_block;
1015 /*
1016 * check if the first free block is within the
1017 * free space we just reserved
1018 */
1019 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1020 return 0; /* success */
1021 /*
1022 * if the first free bit we found is out of the reservable space
1023 * continue search for next reservable space,
1024 * start from where the free block is,
1025 * we also shift the list head to where we stopped last time
1026 */
1027 search_head = my_rsv;
1028 spin_lock(rsv_lock);
1029 goto retry;
1030}
1031
1032/**
1033 * try_to_extend_reservation()
1034 * @my_rsv: given reservation window
1035 * @sb: super block
1036 * @size: the delta to extend
1037 *
1038 * Attempt to expand the reservation window large enough to have
1039 * required number of free blocks
1040 *
1041 * Since ext2_try_to_allocate() will always allocate blocks within
1042 * the reservation window range, if the window size is too small,
1043 * multiple blocks allocation has to stop at the end of the reservation
1044 * window. To make this more efficient, given the total number of
1045 * blocks needed and the current size of the window, we try to
1046 * expand the reservation window size if necessary on a best-effort
1047 * basis before ext2_new_blocks() tries to allocate blocks.
1048 */
1049static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1050 struct super_block *sb, int size)
1051{
1052 struct ext2_reserve_window_node *next_rsv;
1053 struct rb_node *next;
1054 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1055
1056 if (!spin_trylock(rsv_lock))
1057 return;
1058
1059 next = rb_next(&my_rsv->rsv_node);
1060
1061 if (!next)
1062 my_rsv->rsv_end += size;
1063 else {
1064 next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1065
1066 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1067 my_rsv->rsv_end += size;
1068 else
1069 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1070 }
1071 spin_unlock(rsv_lock);
1072}
1073
1074/**
1075 * ext2_try_to_allocate_with_rsv()
1076 * @sb: superblock
1077 * @group: given allocation block group
1078 * @bitmap_bh: bufferhead holds the block bitmap
1079 * @grp_goal: given target block within the group
1080 * @count: target number of blocks to allocate
1081 * @my_rsv: reservation window
1082 *
1083 * This is the main function used to allocate a new block and its reservation
1084 * window.
1085 *
1086 * Each time when a new block allocation is need, first try to allocate from
1087 * its own reservation. If it does not have a reservation window, instead of
1088 * looking for a free bit on bitmap first, then look up the reservation list to
1089 * see if it is inside somebody else's reservation window, we try to allocate a
1090 * reservation window for it starting from the goal first. Then do the block
1091 * allocation within the reservation window.
1092 *
1093 * This will avoid keeping on searching the reservation list again and
1094 * again when somebody is looking for a free block (without
1095 * reservation), and there are lots of free blocks, but they are all
1096 * being reserved.
1097 *
1098 * We use a red-black tree for the per-filesystem reservation list.
1099 */
1100static ext2_grpblk_t
1101ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1102 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1103 struct ext2_reserve_window_node * my_rsv,
1104 unsigned long *count)
1105{
1106 ext2_fsblk_t group_first_block, group_last_block;
1107 ext2_grpblk_t ret = 0;
1108 unsigned long num = *count;
1109
1110 /*
1111 * we don't deal with reservation when
1112 * filesystem is mounted without reservation
1113 * or the file is not a regular file
1114 * or last attempt to allocate a block with reservation turned on failed
1115 */
1116 if (my_rsv == NULL) {
1117 return ext2_try_to_allocate(sb, group, bitmap_bh,
1118 grp_goal, count, NULL);
1119 }
1120 /*
1121 * grp_goal is a group relative block number (if there is a goal)
1122 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1123 * first block is a filesystem wide block number
1124 * first block is the block number of the first block in this group
1125 */
1126 group_first_block = ext2_group_first_block_no(sb, group);
1127 group_last_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
1128
1129 /*
1130 * Basically we will allocate a new block from inode's reservation
1131 * window.
1132 *
1133 * We need to allocate a new reservation window, if:
1134 * a) inode does not have a reservation window; or
1135 * b) last attempt to allocate a block from existing reservation
1136 * failed; or
1137 * c) we come here with a goal and with a reservation window
1138 *
1139 * We do not need to allocate a new reservation window if we come here
1140 * at the beginning with a goal and the goal is inside the window, or
1141 * we don't have a goal but already have a reservation window.
1142 * then we could go to allocate from the reservation window directly.
1143 */
1144 while (1) {
1145 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1146 !goal_in_my_reservation(&my_rsv->rsv_window,
1147 grp_goal, group, sb)) {
1148 if (my_rsv->rsv_goal_size < *count)
1149 my_rsv->rsv_goal_size = *count;
1150 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1151 group, bitmap_bh);
1152 if (ret < 0)
1153 break; /* failed */
1154
1155 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1156 grp_goal, group, sb))
1157 grp_goal = -1;
1158 } else if (grp_goal >= 0) {
1159 int curr = my_rsv->rsv_end -
1160 (grp_goal + group_first_block) + 1;
1161
1162 if (curr < *count)
1163 try_to_extend_reservation(my_rsv, sb,
1164 *count - curr);
1165 }
1166
1167 if ((my_rsv->rsv_start > group_last_block) ||
1168 (my_rsv->rsv_end < group_first_block)) {
1169 rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1170 BUG();
1171 }
1172 ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1173 &num, &my_rsv->rsv_window);
1174 if (ret >= 0) {
1175 my_rsv->rsv_alloc_hit += num;
1176 *count = num;
1177 break; /* succeed */
1178 }
1179 num = *count;
1180 }
1181 return ret;
1182}
1183
1184/**
1185 * ext2_has_free_blocks()
1186 * @sbi: in-core super block structure.
1187 *
1188 * Check if filesystem has at least 1 free block available for allocation.
1189 */
1190static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1191{
1192 ext2_fsblk_t free_blocks, root_blocks;
1193
1194 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1195 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1196 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
David Howellsa8dd4d62008-11-14 10:38:50 +11001197 sbi->s_resuid != current_fsuid() &&
Martin J. Bligha686cd82007-10-16 23:30:46 -07001198 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1199 return 0;
1200 }
1201 return 1;
1202}
1203
1204/*
1205 * ext2_new_blocks() -- core block(s) allocation function
1206 * @inode: file inode
1207 * @goal: given target block(filesystem wide)
1208 * @count: target number of blocks to allocate
1209 * @errp: error code
1210 *
1211 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 * free, or there is a free block within 32 blocks of the goal, that block
1213 * is allocated. Otherwise a forward search is made for a free block; within
1214 * each block group the search first looks for an entire free byte in the block
1215 * bitmap, and then for any free bit if that fails.
1216 * This function also updates quota and i_blocks field.
1217 */
Martin J. Bligha686cd82007-10-16 23:30:46 -07001218ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1219 unsigned long *count, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220{
1221 struct buffer_head *bitmap_bh = NULL;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001222 struct buffer_head *gdp_bh;
1223 int group_no;
1224 int goal_group;
1225 ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1226 ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1227 ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
1228 int bgi; /* blockgroup iteration index */
1229 int performed_allocation = 0;
1230 ext2_grpblk_t free_blocks; /* number of free blocks in a group */
1231 struct super_block *sb;
1232 struct ext2_group_desc *gdp;
1233 struct ext2_super_block *es;
1234 struct ext2_sb_info *sbi;
1235 struct ext2_reserve_window_node *my_rsv = NULL;
1236 struct ext2_block_alloc_info *block_i;
1237 unsigned short windowsz = 0;
1238 unsigned long ngroups;
1239 unsigned long num = *count;
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001240 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Martin J. Bligha686cd82007-10-16 23:30:46 -07001242 *errp = -ENOSPC;
1243 sb = inode->i_sb;
1244 if (!sb) {
1245 printk("ext2_new_blocks: nonexistent device");
1246 return 0;
1247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
Martin J. Bligha686cd82007-10-16 23:30:46 -07001249 /*
1250 * Check quota for allocation of this block.
1251 */
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001252 ret = dquot_alloc_block(inode, num);
1253 if (ret) {
1254 *errp = ret;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001255 return 0;
1256 }
1257
1258 sbi = EXT2_SB(sb);
1259 es = EXT2_SB(sb)->s_es;
1260 ext2_debug("goal=%lu.\n", goal);
1261 /*
1262 * Allocate a block from reservation only when
1263 * filesystem is mounted with reservation(default,-o reservation), and
1264 * it's a regular file, and
1265 * the desired window size is greater than 0 (One could use ioctl
1266 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1267 * reservation on that particular file)
1268 */
1269 block_i = EXT2_I(inode)->i_block_alloc_info;
1270 if (block_i) {
1271 windowsz = block_i->rsv_window_node.rsv_goal_size;
1272 if (windowsz > 0)
1273 my_rsv = &block_i->rsv_window_node;
1274 }
1275
1276 if (!ext2_has_free_blocks(sbi)) {
1277 *errp = -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 goto out;
1279 }
1280
Martin J. Bligha686cd82007-10-16 23:30:46 -07001281 /*
1282 * First, test whether the goal block is free.
1283 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (goal < le32_to_cpu(es->s_first_data_block) ||
1285 goal >= le32_to_cpu(es->s_blocks_count))
1286 goal = le32_to_cpu(es->s_first_data_block);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001287 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1288 EXT2_BLOCKS_PER_GROUP(sb);
1289 goal_group = group_no;
1290retry_alloc:
1291 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1292 if (!gdp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 goto io_error;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001294
1295 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1296 /*
1297 * if there is not enough free blocks to make a new resevation
1298 * turn off reservation for this allocation
1299 */
1300 if (my_rsv && (free_blocks < windowsz)
Mingming Caod707d312008-10-15 22:04:01 -07001301 && (free_blocks > 0)
Martin J. Bligha686cd82007-10-16 23:30:46 -07001302 && (rsv_is_empty(&my_rsv->rsv_window)))
1303 my_rsv = NULL;
1304
1305 if (free_blocks > 0) {
1306 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1307 EXT2_BLOCKS_PER_GROUP(sb));
1308 bitmap_bh = read_block_bitmap(sb, group_no);
1309 if (!bitmap_bh)
1310 goto io_error;
1311 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1312 bitmap_bh, grp_target_blk,
1313 my_rsv, &num);
1314 if (grp_alloc_blk >= 0)
1315 goto allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 }
1317
Martin J. Bligha686cd82007-10-16 23:30:46 -07001318 ngroups = EXT2_SB(sb)->s_groups_count;
1319 smp_rmb();
1320
1321 /*
Akinobu Mita144704e2008-02-06 01:40:15 -08001322 * Now search the rest of the groups. We assume that
1323 * group_no and gdp correctly point to the last group visited.
Martin J. Bligha686cd82007-10-16 23:30:46 -07001324 */
1325 for (bgi = 0; bgi < ngroups; bgi++) {
1326 group_no++;
1327 if (group_no >= ngroups)
1328 group_no = 0;
1329 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1330 if (!gdp)
1331 goto io_error;
1332
1333 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1334 /*
1335 * skip this group if the number of
1336 * free blocks is less than half of the reservation
1337 * window size.
1338 */
Mingming Caod707d312008-10-15 22:04:01 -07001339 if (my_rsv && (free_blocks <= (windowsz/2)))
Martin J. Bligha686cd82007-10-16 23:30:46 -07001340 continue;
1341
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 brelse(bitmap_bh);
1343 bitmap_bh = read_block_bitmap(sb, group_no);
1344 if (!bitmap_bh)
1345 goto io_error;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001346 /*
1347 * try to allocate block(s) from this group, without a goal(-1).
1348 */
1349 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1350 bitmap_bh, -1, my_rsv, &num);
1351 if (grp_alloc_blk >= 0)
1352 goto allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 /*
Martin J. Bligha686cd82007-10-16 23:30:46 -07001355 * We may end up a bogus ealier ENOSPC error due to
1356 * filesystem is "full" of reservations, but
1357 * there maybe indeed free blocks avaliable on disk
1358 * In this case, we just forget about the reservations
1359 * just do block allocation as without reservations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 */
Martin J. Bligha686cd82007-10-16 23:30:46 -07001361 if (my_rsv) {
1362 my_rsv = NULL;
1363 windowsz = 0;
1364 group_no = goal_group;
1365 goto retry_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 }
Martin J. Bligha686cd82007-10-16 23:30:46 -07001367 /* No space left on the device */
1368 *errp = -ENOSPC;
1369 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Martin J. Bligha686cd82007-10-16 23:30:46 -07001371allocated:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 ext2_debug("using block group %d(%d)\n",
Martin J. Bligha686cd82007-10-16 23:30:46 -07001374 group_no, gdp->bg_free_blocks_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
Martin J. Bligha686cd82007-10-16 23:30:46 -07001376 ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
Martin J. Bligha686cd82007-10-16 23:30:46 -07001378 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1379 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1380 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1381 EXT2_SB(sb)->s_itb_per_group) ||
1382 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -08001383 EXT2_SB(sb)->s_itb_per_group)) {
Martin J. Bligha686cd82007-10-16 23:30:46 -07001384 ext2_error(sb, "ext2_new_blocks",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 "Allocating block in system zone - "
Martin J. Bligha686cd82007-10-16 23:30:46 -07001386 "blocks from "E2FSBLK", length %lu",
1387 ret_block, num);
Aneesh Kumar K.V8b915822008-04-28 02:16:04 -07001388 /*
1389 * ext2_try_to_allocate marked the blocks we allocated as in
1390 * use. So we may want to selectively mark some of the blocks
1391 * as free
1392 */
1393 goto retry_alloc;
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -08001394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
Martin J. Bligha686cd82007-10-16 23:30:46 -07001396 performed_allocation = 1;
1397
1398 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1399 ext2_error(sb, "ext2_new_blocks",
1400 "block("E2FSBLK") >= blocks count(%d) - "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 "block_group = %d, es == %p ", ret_block,
1402 le32_to_cpu(es->s_blocks_count), group_no, es);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001403 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Martin J. Bligha686cd82007-10-16 23:30:46 -07001406 group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1407 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
1409 mark_buffer_dirty(bitmap_bh);
1410 if (sb->s_flags & MS_SYNCHRONOUS)
1411 sync_dirty_buffer(bitmap_bh);
1412
Martin J. Bligha686cd82007-10-16 23:30:46 -07001413 *errp = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 brelse(bitmap_bh);
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001415 dquot_free_block(inode, *count-num);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001416 *count = num;
1417 return ret_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
1419io_error:
Martin J. Bligha686cd82007-10-16 23:30:46 -07001420 *errp = -EIO;
1421out:
1422 /*
1423 * Undo the block allocation
1424 */
1425 if (!performed_allocation)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001426 dquot_free_block(inode, *count);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001427 brelse(bitmap_bh);
1428 return 0;
1429}
1430
1431ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1432{
1433 unsigned long count = 1;
1434
1435 return ext2_new_blocks(inode, goal, &count, errp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436}
1437
Valerie Henson21730ee2006-06-25 05:48:12 -07001438#ifdef EXT2FS_DEBUG
1439
Philippe De Muyterfebfcf92007-10-16 23:26:15 -07001440static 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 -07001441
1442unsigned long ext2_count_free (struct buffer_head * map, unsigned int numchars)
1443{
1444 unsigned int i;
1445 unsigned long sum = 0;
1446
1447 if (!map)
1448 return (0);
1449 for (i = 0; i < numchars; i++)
1450 sum += nibblemap[map->b_data[i] & 0xf] +
1451 nibblemap[(map->b_data[i] >> 4) & 0xf];
1452 return (sum);
1453}
1454
1455#endif /* EXT2FS_DEBUG */
1456
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457unsigned long ext2_count_free_blocks (struct super_block * sb)
1458{
1459 struct ext2_group_desc * desc;
1460 unsigned long desc_count = 0;
1461 int i;
1462#ifdef EXT2FS_DEBUG
1463 unsigned long bitmap_count, x;
1464 struct ext2_super_block *es;
1465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 es = EXT2_SB(sb)->s_es;
1467 desc_count = 0;
1468 bitmap_count = 0;
1469 desc = NULL;
1470 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1471 struct buffer_head *bitmap_bh;
1472 desc = ext2_get_group_desc (sb, i, NULL);
1473 if (!desc)
1474 continue;
1475 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1476 bitmap_bh = read_block_bitmap(sb, i);
1477 if (!bitmap_bh)
1478 continue;
1479
1480 x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1481 printk ("group %d: stored = %d, counted = %lu\n",
1482 i, le16_to_cpu(desc->bg_free_blocks_count), x);
1483 bitmap_count += x;
1484 brelse(bitmap_bh);
1485 }
1486 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1487 (long)le32_to_cpu(es->s_free_blocks_count),
1488 desc_count, bitmap_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 return bitmap_count;
1490#else
1491 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1492 desc = ext2_get_group_desc (sb, i, NULL);
1493 if (!desc)
1494 continue;
1495 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1496 }
1497 return desc_count;
1498#endif
1499}
1500
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501static inline int test_root(int a, int b)
1502{
1503 int num = b;
1504
1505 while (a > num)
1506 num *= b;
1507 return num == a;
1508}
1509
1510static int ext2_group_sparse(int group)
1511{
1512 if (group <= 1)
1513 return 1;
1514 return (test_root(group, 3) || test_root(group, 5) ||
1515 test_root(group, 7));
1516}
1517
1518/**
1519 * ext2_bg_has_super - number of blocks used by the superblock in group
1520 * @sb: superblock for filesystem
1521 * @group: group number to check
1522 *
1523 * Return the number of blocks used by the superblock (primary or backup)
1524 * in this group. Currently this will be only 0 or 1.
1525 */
1526int ext2_bg_has_super(struct super_block *sb, int group)
1527{
1528 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1529 !ext2_group_sparse(group))
1530 return 0;
1531 return 1;
1532}
1533
1534/**
1535 * ext2_bg_num_gdb - number of blocks used by the group table in group
1536 * @sb: superblock for filesystem
1537 * @group: group number to check
1538 *
1539 * Return the number of blocks used by the group descriptor table
1540 * (primary or backup) in this group. In the future there may be a
1541 * different number of descriptor blocks in each group.
1542 */
1543unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1544{
Akinobu Mita859cb932008-02-06 01:40:17 -08001545 return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546}
1547