blob: 794008b6ce29779d75a4de2593cd19690e5a323f [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
72/*
73 * Read the bitmap for a given block_group, reading into the specified
74 * slot in the superblock's bitmap cache.
75 *
76 * Return buffer_head on success or NULL in case of failure.
77 */
78static struct buffer_head *
79read_block_bitmap(struct super_block *sb, unsigned int block_group)
80{
81 struct ext2_group_desc * desc;
82 struct buffer_head * bh = NULL;
Linus Torvalds0b832a42007-11-13 08:07:31 -080083
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 desc = ext2_get_group_desc (sb, block_group, NULL);
85 if (!desc)
Linus Torvalds0b832a42007-11-13 08:07:31 -080086 goto error_out;
87 bh = sb_bread(sb, le32_to_cpu(desc->bg_block_bitmap));
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 if (!bh)
Linus Torvalds0b832a42007-11-13 08:07:31 -080089 ext2_error (sb, "read_block_bitmap",
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 "Cannot read block bitmap - "
91 "block_group = %d, block_bitmap = %u",
92 block_group, le32_to_cpu(desc->bg_block_bitmap));
Aneesh Kumar K.V7c9e69f2007-10-16 23:27:02 -070093error_out:
Linus Torvalds0b832a42007-11-13 08:07:31 -080094 return bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095}
96
Linus Torvalds1da177e2005-04-16 15:20:36 -070097static void release_blocks(struct super_block *sb, int count)
98{
99 if (count) {
100 struct ext2_sb_info *sbi = EXT2_SB(sb);
101
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700102 percpu_counter_add(&sbi->s_freeblocks_counter, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 sb->s_dirt = 1;
104 }
105}
106
Martin J. Bligha686cd82007-10-16 23:30:46 -0700107static void group_adjust_blocks(struct super_block *sb, int group_no,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 struct ext2_group_desc *desc, struct buffer_head *bh, int count)
109{
110 if (count) {
111 struct ext2_sb_info *sbi = EXT2_SB(sb);
112 unsigned free_blocks;
113
114 spin_lock(sb_bgl_lock(sbi, group_no));
115 free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
116 desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
117 spin_unlock(sb_bgl_lock(sbi, group_no));
118 sb->s_dirt = 1;
119 mark_buffer_dirty(bh);
120 }
121}
122
Martin J. Bligha686cd82007-10-16 23:30:46 -0700123/*
124 * The reservation window structure operations
125 * --------------------------------------------
126 * Operations include:
127 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
128 *
129 * We use a red-black tree to represent per-filesystem reservation
130 * windows.
131 *
132 */
133
134/**
135 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
136 * @rb_root: root of per-filesystem reservation rb tree
137 * @verbose: verbose mode
138 * @fn: function which wishes to dump the reservation map
139 *
140 * If verbose is turned on, it will print the whole block reservation
141 * windows(start, end). Otherwise, it will only print out the "bad" windows,
142 * those windows that overlap with their immediate neighbors.
143 */
144#if 1
145static void __rsv_window_dump(struct rb_root *root, int verbose,
146 const char *fn)
147{
148 struct rb_node *n;
149 struct ext2_reserve_window_node *rsv, *prev;
150 int bad;
151
152restart:
153 n = rb_first(root);
154 bad = 0;
155 prev = NULL;
156
157 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
158 while (n) {
159 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
160 if (verbose)
161 printk("reservation window 0x%p "
162 "start: %lu, end: %lu\n",
163 rsv, rsv->rsv_start, rsv->rsv_end);
164 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
165 printk("Bad reservation %p (start >= end)\n",
166 rsv);
167 bad = 1;
168 }
169 if (prev && prev->rsv_end >= rsv->rsv_start) {
170 printk("Bad reservation %p (prev->end >= start)\n",
171 rsv);
172 bad = 1;
173 }
174 if (bad) {
175 if (!verbose) {
176 printk("Restarting reservation walk in verbose mode\n");
177 verbose = 1;
178 goto restart;
179 }
180 }
181 n = rb_next(n);
182 prev = rsv;
183 }
184 printk("Window map complete.\n");
185 if (bad)
186 BUG();
187}
188#define rsv_window_dump(root, verbose) \
189 __rsv_window_dump((root), (verbose), __FUNCTION__)
190#else
191#define rsv_window_dump(root, verbose) do {} while (0)
192#endif
193
194/**
195 * goal_in_my_reservation()
196 * @rsv: inode's reservation window
197 * @grp_goal: given goal block relative to the allocation block group
198 * @group: the current allocation block group
199 * @sb: filesystem super block
200 *
201 * Test if the given goal block (group relative) is within the file's
202 * own block reservation window range.
203 *
204 * If the reservation window is outside the goal allocation group, return 0;
205 * grp_goal (given goal block) could be -1, which means no specific
206 * goal block. In this case, always return 1.
207 * If the goal block is within the reservation window, return 1;
208 * otherwise, return 0;
209 */
210static int
211goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
212 unsigned int group, struct super_block * sb)
213{
214 ext2_fsblk_t group_first_block, group_last_block;
215
216 group_first_block = ext2_group_first_block_no(sb, group);
217 group_last_block = group_first_block + EXT2_BLOCKS_PER_GROUP(sb) - 1;
218
219 if ((rsv->_rsv_start > group_last_block) ||
220 (rsv->_rsv_end < group_first_block))
221 return 0;
222 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
223 || (grp_goal + group_first_block > rsv->_rsv_end)))
224 return 0;
225 return 1;
226}
227
228/**
229 * search_reserve_window()
230 * @rb_root: root of reservation tree
231 * @goal: target allocation block
232 *
233 * Find the reserved window which includes the goal, or the previous one
234 * if the goal is not in any window.
235 * Returns NULL if there are no windows or if all windows start after the goal.
236 */
237static struct ext2_reserve_window_node *
238search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
239{
240 struct rb_node *n = root->rb_node;
241 struct ext2_reserve_window_node *rsv;
242
243 if (!n)
244 return NULL;
245
246 do {
247 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
248
249 if (goal < rsv->rsv_start)
250 n = n->rb_left;
251 else if (goal > rsv->rsv_end)
252 n = n->rb_right;
253 else
254 return rsv;
255 } while (n);
256 /*
257 * We've fallen off the end of the tree: the goal wasn't inside
258 * any particular node. OK, the previous node must be to one
259 * side of the interval containing the goal. If it's the RHS,
260 * we need to back up one.
261 */
262 if (rsv->rsv_start > goal) {
263 n = rb_prev(&rsv->rsv_node);
264 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
265 }
266 return rsv;
267}
268
269/*
270 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
271 * @sb: super block
272 * @rsv: reservation window to add
273 *
274 * Must be called with rsv_lock held.
275 */
276void ext2_rsv_window_add(struct super_block *sb,
277 struct ext2_reserve_window_node *rsv)
278{
279 struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
280 struct rb_node *node = &rsv->rsv_node;
281 ext2_fsblk_t start = rsv->rsv_start;
282
283 struct rb_node ** p = &root->rb_node;
284 struct rb_node * parent = NULL;
285 struct ext2_reserve_window_node *this;
286
287 while (*p)
288 {
289 parent = *p;
290 this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
291
292 if (start < this->rsv_start)
293 p = &(*p)->rb_left;
294 else if (start > this->rsv_end)
295 p = &(*p)->rb_right;
296 else {
297 rsv_window_dump(root, 1);
298 BUG();
299 }
300 }
301
302 rb_link_node(node, parent, p);
303 rb_insert_color(node, root);
304}
305
306/**
307 * rsv_window_remove() -- unlink a window from the reservation rb tree
308 * @sb: super block
309 * @rsv: reservation window to remove
310 *
311 * Mark the block reservation window as not allocated, and unlink it
312 * from the filesystem reservation window rb tree. Must be called with
313 * rsv_lock held.
314 */
315static void rsv_window_remove(struct super_block *sb,
316 struct ext2_reserve_window_node *rsv)
317{
318 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
319 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
320 rsv->rsv_alloc_hit = 0;
321 rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
322}
323
324/*
325 * rsv_is_empty() -- Check if the reservation window is allocated.
326 * @rsv: given reservation window to check
327 *
328 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
329 */
330static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
331{
332 /* a valid reservation end block could not be 0 */
333 return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
334}
335
336/**
337 * ext2_init_block_alloc_info()
338 * @inode: file inode structure
339 *
340 * Allocate and initialize the reservation window structure, and
341 * link the window to the ext2 inode structure at last
342 *
343 * The reservation window structure is only dynamically allocated
344 * and linked to ext2 inode the first time the open file
345 * needs a new block. So, before every ext2_new_block(s) call, for
346 * regular files, we should check whether the reservation window
347 * structure exists or not. In the latter case, this function is called.
348 * Fail to do so will result in block reservation being turned off for that
349 * open file.
350 *
351 * This function is called from ext2_get_blocks_handle(), also called
352 * when setting the reservation window size through ioctl before the file
353 * is open for write (needs block allocation).
354 *
355 * Needs truncate_mutex protection prior to calling this function.
356 */
357void ext2_init_block_alloc_info(struct inode *inode)
358{
359 struct ext2_inode_info *ei = EXT2_I(inode);
360 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
361 struct super_block *sb = inode->i_sb;
362
363 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
364 if (block_i) {
365 struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
366
367 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
368 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
369
370 /*
371 * if filesystem is mounted with NORESERVATION, the goal
372 * reservation window size is set to zero to indicate
373 * block reservation is off
374 */
375 if (!test_opt(sb, RESERVATION))
376 rsv->rsv_goal_size = 0;
377 else
378 rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
379 rsv->rsv_alloc_hit = 0;
380 block_i->last_alloc_logical_block = 0;
381 block_i->last_alloc_physical_block = 0;
382 }
383 ei->i_block_alloc_info = block_i;
384}
385
386/**
387 * ext2_discard_reservation()
388 * @inode: inode
389 *
390 * Discard(free) block reservation window on last file close, or truncate
391 * or at last iput().
392 *
393 * It is being called in three cases:
394 * ext2_release_file(): last writer closes the file
395 * ext2_clear_inode(): last iput(), when nobody links to this file.
396 * ext2_truncate(): when the block indirect map is about to change.
397 */
398void ext2_discard_reservation(struct inode *inode)
399{
400 struct ext2_inode_info *ei = EXT2_I(inode);
401 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
402 struct ext2_reserve_window_node *rsv;
403 spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
404
405 if (!block_i)
406 return;
407
408 rsv = &block_i->rsv_window_node;
409 if (!rsv_is_empty(&rsv->rsv_window)) {
410 spin_lock(rsv_lock);
411 if (!rsv_is_empty(&rsv->rsv_window))
412 rsv_window_remove(inode->i_sb, rsv);
413 spin_unlock(rsv_lock);
414 }
415}
416
417/**
418 * ext2_free_blocks_sb() -- Free given blocks and update quota and i_blocks
419 * @inode: inode
420 * @block: start physcial block to free
421 * @count: number of blocks to free
422 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423void ext2_free_blocks (struct inode * inode, unsigned long block,
424 unsigned long count)
425{
426 struct buffer_head *bitmap_bh = NULL;
427 struct buffer_head * bh2;
428 unsigned long block_group;
429 unsigned long bit;
430 unsigned long i;
431 unsigned long overflow;
432 struct super_block * sb = inode->i_sb;
433 struct ext2_sb_info * sbi = EXT2_SB(sb);
434 struct ext2_group_desc * desc;
435 struct ext2_super_block * es = sbi->s_es;
436 unsigned freed = 0, group_freed;
437
438 if (block < le32_to_cpu(es->s_first_data_block) ||
439 block + count < block ||
440 block + count > le32_to_cpu(es->s_blocks_count)) {
441 ext2_error (sb, "ext2_free_blocks",
442 "Freeing blocks not in datazone - "
443 "block = %lu, count = %lu", block, count);
444 goto error_return;
445 }
446
447 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
448
449do_more:
450 overflow = 0;
451 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
452 EXT2_BLOCKS_PER_GROUP(sb);
453 bit = (block - le32_to_cpu(es->s_first_data_block)) %
454 EXT2_BLOCKS_PER_GROUP(sb);
455 /*
456 * Check to see if we are freeing blocks across a group
457 * boundary.
458 */
459 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
460 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
461 count -= overflow;
462 }
463 brelse(bitmap_bh);
464 bitmap_bh = read_block_bitmap(sb, block_group);
465 if (!bitmap_bh)
466 goto error_return;
467
468 desc = ext2_get_group_desc (sb, block_group, &bh2);
469 if (!desc)
470 goto error_return;
471
472 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
473 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
474 in_range (block, le32_to_cpu(desc->bg_inode_table),
475 sbi->s_itb_per_group) ||
476 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -0800477 sbi->s_itb_per_group)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 ext2_error (sb, "ext2_free_blocks",
479 "Freeing blocks in system zones - "
480 "Block = %lu, count = %lu",
481 block, count);
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -0800482 goto error_return;
483 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
485 for (i = 0, group_freed = 0; i < count; i++) {
486 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
487 bit + i, bitmap_bh->b_data)) {
488 ext2_error(sb, __FUNCTION__,
489 "bit already cleared for block %lu", block + i);
490 } else {
491 group_freed++;
492 }
493 }
494
495 mark_buffer_dirty(bitmap_bh);
496 if (sb->s_flags & MS_SYNCHRONOUS)
497 sync_dirty_buffer(bitmap_bh);
498
Martin J. Bligha686cd82007-10-16 23:30:46 -0700499 group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 freed += group_freed;
501
502 if (overflow) {
503 block += count;
504 count = overflow;
505 goto do_more;
506 }
507error_return:
508 brelse(bitmap_bh);
509 release_blocks(sb, freed);
510 DQUOT_FREE_BLOCK(inode, freed);
511}
512
Martin J. Bligha686cd82007-10-16 23:30:46 -0700513/**
514 * bitmap_search_next_usable_block()
515 * @start: the starting block (group relative) of the search
516 * @bh: bufferhead contains the block group bitmap
517 * @maxblocks: the ending block (group relative) of the reservation
518 *
519 * The bitmap search --- search forward through the actual bitmap on disk until
520 * we find a bit free.
521 */
522static ext2_grpblk_t
523bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
524 ext2_grpblk_t maxblocks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
Martin J. Bligha686cd82007-10-16 23:30:46 -0700526 ext2_grpblk_t next;
527
528 next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
529 if (next >= maxblocks)
530 return -1;
531 return next;
532}
533
534/**
535 * find_next_usable_block()
536 * @start: the starting block (group relative) to find next
537 * allocatable block in bitmap.
538 * @bh: bufferhead contains the block group bitmap
539 * @maxblocks: the ending block (group relative) for the search
540 *
541 * Find an allocatable block in a bitmap. We perform the "most
542 * appropriate allocation" algorithm of looking for a free block near
543 * the initial goal; then for a free byte somewhere in the bitmap;
544 * then for any free bit in the bitmap.
545 */
546static ext2_grpblk_t
547find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
548{
549 ext2_grpblk_t here, next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 char *p, *r;
551
Martin J. Bligha686cd82007-10-16 23:30:46 -0700552 if (start > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 /*
554 * The goal was occupied; search forward for a free
555 * block within the next XX blocks.
556 *
557 * end_goal is more or less random, but it has to be
558 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
559 * next 64-bit boundary is simple..
560 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700561 ext2_grpblk_t end_goal = (start + 63) & ~63;
562 if (end_goal > maxblocks)
563 end_goal = maxblocks;
564 here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
565 if (here < end_goal)
566 return here;
567 ext2_debug("Bit not found near goal\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 }
569
Martin J. Bligha686cd82007-10-16 23:30:46 -0700570 here = start;
571 if (here < 0)
572 here = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Martin J. Bligha686cd82007-10-16 23:30:46 -0700574 p = ((char *)bh->b_data) + (here >> 3);
575 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
576 next = (r - ((char *)bh->b_data)) << 3;
577
578 if (next < maxblocks && next >= here)
579 return next;
580
581 here = bitmap_search_next_usable_block(here, bh, maxblocks);
582 return here;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583}
584
585/*
Martin J. Bligha686cd82007-10-16 23:30:46 -0700586 * ext2_try_to_allocate()
587 * @sb: superblock
588 * @handle: handle to this transaction
589 * @group: given allocation block group
590 * @bitmap_bh: bufferhead holds the block bitmap
591 * @grp_goal: given target block within the group
592 * @count: target number of blocks to allocate
593 * @my_rsv: reservation window
594 *
595 * Attempt to allocate blocks within a give range. Set the range of allocation
596 * first, then find the first free bit(s) from the bitmap (within the range),
597 * and at last, allocate the blocks by claiming the found free bit as allocated.
598 *
599 * To set the range of this allocation:
600 * if there is a reservation window, only try to allocate block(s)
601 * from the file's own reservation window;
602 * Otherwise, the allocation range starts from the give goal block,
603 * ends at the block group's last block.
604 *
605 * If we failed to allocate the desired block then we may end up crossing to a
606 * new bitmap.
607 */
608static int
609ext2_try_to_allocate(struct super_block *sb, int group,
610 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
611 unsigned long *count,
612 struct ext2_reserve_window *my_rsv)
613{
614 ext2_fsblk_t group_first_block;
615 ext2_grpblk_t start, end;
616 unsigned long num = 0;
617
618 /* we do allocation within the reservation window if we have a window */
619 if (my_rsv) {
620 group_first_block = ext2_group_first_block_no(sb, group);
621 if (my_rsv->_rsv_start >= group_first_block)
622 start = my_rsv->_rsv_start - group_first_block;
623 else
624 /* reservation window cross group boundary */
625 start = 0;
626 end = my_rsv->_rsv_end - group_first_block + 1;
627 if (end > EXT2_BLOCKS_PER_GROUP(sb))
628 /* reservation window crosses group boundary */
629 end = EXT2_BLOCKS_PER_GROUP(sb);
630 if ((start <= grp_goal) && (grp_goal < end))
631 start = grp_goal;
632 else
633 grp_goal = -1;
634 } else {
635 if (grp_goal > 0)
636 start = grp_goal;
637 else
638 start = 0;
639 end = EXT2_BLOCKS_PER_GROUP(sb);
640 }
641
642 BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
643
644repeat:
645 if (grp_goal < 0) {
646 grp_goal = find_next_usable_block(start, bitmap_bh, end);
647 if (grp_goal < 0)
648 goto fail_access;
649 if (!my_rsv) {
650 int i;
651
652 for (i = 0; i < 7 && grp_goal > start &&
653 !ext2_test_bit(grp_goal - 1,
654 bitmap_bh->b_data);
655 i++, grp_goal--)
656 ;
657 }
658 }
659 start = grp_goal;
660
661 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), grp_goal,
662 bitmap_bh->b_data)) {
663 /*
664 * The block was allocated by another thread, or it was
665 * allocated and then freed by another thread
666 */
667 start++;
668 grp_goal++;
669 if (start >= end)
670 goto fail_access;
671 goto repeat;
672 }
673 num++;
674 grp_goal++;
675 while (num < *count && grp_goal < end
676 && !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
677 grp_goal, bitmap_bh->b_data)) {
678 num++;
679 grp_goal++;
680 }
681 *count = num;
682 return grp_goal - num;
683fail_access:
684 *count = num;
685 return -1;
686}
687
688/**
689 * find_next_reservable_window():
690 * find a reservable space within the given range.
691 * It does not allocate the reservation window for now:
692 * alloc_new_reservation() will do the work later.
693 *
694 * @search_head: the head of the searching list;
695 * This is not necessarily the list head of the whole filesystem
696 *
697 * We have both head and start_block to assist the search
698 * for the reservable space. The list starts from head,
699 * but we will shift to the place where start_block is,
700 * then start from there, when looking for a reservable space.
701 *
702 * @size: the target new reservation window size
703 *
704 * @group_first_block: the first block we consider to start
705 * the real search from
706 *
707 * @last_block:
708 * the maximum block number that our goal reservable space
709 * could start from. This is normally the last block in this
710 * group. The search will end when we found the start of next
711 * possible reservable space is out of this boundary.
712 * This could handle the cross boundary reservation window
713 * request.
714 *
715 * basically we search from the given range, rather than the whole
716 * reservation double linked list, (start_block, last_block)
717 * to find a free region that is of my size and has not
718 * been reserved.
719 *
720 */
721static int find_next_reservable_window(
722 struct ext2_reserve_window_node *search_head,
723 struct ext2_reserve_window_node *my_rsv,
724 struct super_block * sb,
725 ext2_fsblk_t start_block,
726 ext2_fsblk_t last_block)
727{
728 struct rb_node *next;
729 struct ext2_reserve_window_node *rsv, *prev;
730 ext2_fsblk_t cur;
731 int size = my_rsv->rsv_goal_size;
732
733 /* TODO: make the start of the reservation window byte-aligned */
734 /* cur = *start_block & ~7;*/
735 cur = start_block;
736 rsv = search_head;
737 if (!rsv)
738 return -1;
739
740 while (1) {
741 if (cur <= rsv->rsv_end)
742 cur = rsv->rsv_end + 1;
743
744 /* TODO?
745 * in the case we could not find a reservable space
746 * that is what is expected, during the re-search, we could
747 * remember what's the largest reservable space we could have
748 * and return that one.
749 *
750 * For now it will fail if we could not find the reservable
751 * space with expected-size (or more)...
752 */
753 if (cur > last_block)
754 return -1; /* fail */
755
756 prev = rsv;
757 next = rb_next(&rsv->rsv_node);
758 rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
759
760 /*
761 * Reached the last reservation, we can just append to the
762 * previous one.
763 */
764 if (!next)
765 break;
766
767 if (cur + size <= rsv->rsv_start) {
768 /*
769 * Found a reserveable space big enough. We could
770 * have a reservation across the group boundary here
771 */
772 break;
773 }
774 }
775 /*
776 * we come here either :
777 * when we reach the end of the whole list,
778 * and there is empty reservable space after last entry in the list.
779 * append it to the end of the list.
780 *
781 * or we found one reservable space in the middle of the list,
782 * return the reservation window that we could append to.
783 * succeed.
784 */
785
786 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
787 rsv_window_remove(sb, my_rsv);
788
789 /*
790 * Let's book the whole avaliable window for now. We will check the
791 * disk bitmap later and then, if there are free blocks then we adjust
792 * the window size if it's larger than requested.
793 * Otherwise, we will remove this node from the tree next time
794 * call find_next_reservable_window.
795 */
796 my_rsv->rsv_start = cur;
797 my_rsv->rsv_end = cur + size - 1;
798 my_rsv->rsv_alloc_hit = 0;
799
800 if (prev != my_rsv)
801 ext2_rsv_window_add(sb, my_rsv);
802
803 return 0;
804}
805
806/**
807 * alloc_new_reservation()--allocate a new reservation window
808 *
809 * To make a new reservation, we search part of the filesystem
810 * reservation list (the list that inside the group). We try to
811 * allocate a new reservation window near the allocation goal,
812 * or the beginning of the group, if there is no goal.
813 *
814 * We first find a reservable space after the goal, then from
815 * there, we check the bitmap for the first free block after
816 * it. If there is no free block until the end of group, then the
817 * whole group is full, we failed. Otherwise, check if the free
818 * block is inside the expected reservable space, if so, we
819 * succeed.
820 * If the first free block is outside the reservable space, then
821 * start from the first free block, we search for next available
822 * space, and go on.
823 *
824 * on succeed, a new reservation will be found and inserted into the list
825 * It contains at least one free block, and it does not overlap with other
826 * reservation windows.
827 *
828 * failed: we failed to find a reservation window in this group
829 *
830 * @rsv: the reservation
831 *
832 * @grp_goal: The goal (group-relative). It is where the search for a
833 * free reservable space should start from.
834 * if we have a goal(goal >0 ), then start from there,
835 * no goal(goal = -1), we start from the first block
836 * of the group.
837 *
838 * @sb: the super block
839 * @group: the group we are trying to allocate in
840 * @bitmap_bh: the block group block bitmap
841 *
842 */
843static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
844 ext2_grpblk_t grp_goal, struct super_block *sb,
845 unsigned int group, struct buffer_head *bitmap_bh)
846{
847 struct ext2_reserve_window_node *search_head;
848 ext2_fsblk_t group_first_block, group_end_block, start_block;
849 ext2_grpblk_t first_free_block;
850 struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
851 unsigned long size;
852 int ret;
853 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
854
855 group_first_block = ext2_group_first_block_no(sb, group);
856 group_end_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
857
858 if (grp_goal < 0)
859 start_block = group_first_block;
860 else
861 start_block = grp_goal + group_first_block;
862
863 size = my_rsv->rsv_goal_size;
864
865 if (!rsv_is_empty(&my_rsv->rsv_window)) {
866 /*
867 * if the old reservation is cross group boundary
868 * and if the goal is inside the old reservation window,
869 * we will come here when we just failed to allocate from
870 * the first part of the window. We still have another part
871 * that belongs to the next group. In this case, there is no
872 * point to discard our window and try to allocate a new one
873 * in this group(which will fail). we should
874 * keep the reservation window, just simply move on.
875 *
876 * Maybe we could shift the start block of the reservation
877 * window to the first block of next group.
878 */
879
880 if ((my_rsv->rsv_start <= group_end_block) &&
881 (my_rsv->rsv_end > group_end_block) &&
882 (start_block >= my_rsv->rsv_start))
883 return -1;
884
885 if ((my_rsv->rsv_alloc_hit >
886 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
887 /*
888 * if the previously allocation hit ratio is
889 * greater than 1/2, then we double the size of
890 * the reservation window the next time,
891 * otherwise we keep the same size window
892 */
893 size = size * 2;
894 if (size > EXT2_MAX_RESERVE_BLOCKS)
895 size = EXT2_MAX_RESERVE_BLOCKS;
896 my_rsv->rsv_goal_size= size;
897 }
898 }
899
900 spin_lock(rsv_lock);
901 /*
902 * shift the search start to the window near the goal block
903 */
904 search_head = search_reserve_window(fs_rsv_root, start_block);
905
906 /*
907 * find_next_reservable_window() simply finds a reservable window
908 * inside the given range(start_block, group_end_block).
909 *
910 * To make sure the reservation window has a free bit inside it, we
911 * need to check the bitmap after we found a reservable window.
912 */
913retry:
914 ret = find_next_reservable_window(search_head, my_rsv, sb,
915 start_block, group_end_block);
916
917 if (ret == -1) {
918 if (!rsv_is_empty(&my_rsv->rsv_window))
919 rsv_window_remove(sb, my_rsv);
920 spin_unlock(rsv_lock);
921 return -1;
922 }
923
924 /*
925 * On success, find_next_reservable_window() returns the
926 * reservation window where there is a reservable space after it.
927 * Before we reserve this reservable space, we need
928 * to make sure there is at least a free block inside this region.
929 *
930 * Search the first free bit on the block bitmap. Search starts from
931 * the start block of the reservable space we just found.
932 */
933 spin_unlock(rsv_lock);
934 first_free_block = bitmap_search_next_usable_block(
935 my_rsv->rsv_start - group_first_block,
936 bitmap_bh, group_end_block - group_first_block + 1);
937
938 if (first_free_block < 0) {
939 /*
940 * no free block left on the bitmap, no point
941 * to reserve the space. return failed.
942 */
943 spin_lock(rsv_lock);
944 if (!rsv_is_empty(&my_rsv->rsv_window))
945 rsv_window_remove(sb, my_rsv);
946 spin_unlock(rsv_lock);
947 return -1; /* failed */
948 }
949
950 start_block = first_free_block + group_first_block;
951 /*
952 * check if the first free block is within the
953 * free space we just reserved
954 */
955 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
956 return 0; /* success */
957 /*
958 * if the first free bit we found is out of the reservable space
959 * continue search for next reservable space,
960 * start from where the free block is,
961 * we also shift the list head to where we stopped last time
962 */
963 search_head = my_rsv;
964 spin_lock(rsv_lock);
965 goto retry;
966}
967
968/**
969 * try_to_extend_reservation()
970 * @my_rsv: given reservation window
971 * @sb: super block
972 * @size: the delta to extend
973 *
974 * Attempt to expand the reservation window large enough to have
975 * required number of free blocks
976 *
977 * Since ext2_try_to_allocate() will always allocate blocks within
978 * the reservation window range, if the window size is too small,
979 * multiple blocks allocation has to stop at the end of the reservation
980 * window. To make this more efficient, given the total number of
981 * blocks needed and the current size of the window, we try to
982 * expand the reservation window size if necessary on a best-effort
983 * basis before ext2_new_blocks() tries to allocate blocks.
984 */
985static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
986 struct super_block *sb, int size)
987{
988 struct ext2_reserve_window_node *next_rsv;
989 struct rb_node *next;
990 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
991
992 if (!spin_trylock(rsv_lock))
993 return;
994
995 next = rb_next(&my_rsv->rsv_node);
996
997 if (!next)
998 my_rsv->rsv_end += size;
999 else {
1000 next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1001
1002 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1003 my_rsv->rsv_end += size;
1004 else
1005 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1006 }
1007 spin_unlock(rsv_lock);
1008}
1009
1010/**
1011 * ext2_try_to_allocate_with_rsv()
1012 * @sb: superblock
1013 * @group: given allocation block group
1014 * @bitmap_bh: bufferhead holds the block bitmap
1015 * @grp_goal: given target block within the group
1016 * @count: target number of blocks to allocate
1017 * @my_rsv: reservation window
1018 *
1019 * This is the main function used to allocate a new block and its reservation
1020 * window.
1021 *
1022 * Each time when a new block allocation is need, first try to allocate from
1023 * its own reservation. If it does not have a reservation window, instead of
1024 * looking for a free bit on bitmap first, then look up the reservation list to
1025 * see if it is inside somebody else's reservation window, we try to allocate a
1026 * reservation window for it starting from the goal first. Then do the block
1027 * allocation within the reservation window.
1028 *
1029 * This will avoid keeping on searching the reservation list again and
1030 * again when somebody is looking for a free block (without
1031 * reservation), and there are lots of free blocks, but they are all
1032 * being reserved.
1033 *
1034 * We use a red-black tree for the per-filesystem reservation list.
1035 */
1036static ext2_grpblk_t
1037ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1038 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1039 struct ext2_reserve_window_node * my_rsv,
1040 unsigned long *count)
1041{
1042 ext2_fsblk_t group_first_block, group_last_block;
1043 ext2_grpblk_t ret = 0;
1044 unsigned long num = *count;
1045
1046 /*
1047 * we don't deal with reservation when
1048 * filesystem is mounted without reservation
1049 * or the file is not a regular file
1050 * or last attempt to allocate a block with reservation turned on failed
1051 */
1052 if (my_rsv == NULL) {
1053 return ext2_try_to_allocate(sb, group, bitmap_bh,
1054 grp_goal, count, NULL);
1055 }
1056 /*
1057 * grp_goal is a group relative block number (if there is a goal)
1058 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1059 * first block is a filesystem wide block number
1060 * first block is the block number of the first block in this group
1061 */
1062 group_first_block = ext2_group_first_block_no(sb, group);
1063 group_last_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
1064
1065 /*
1066 * Basically we will allocate a new block from inode's reservation
1067 * window.
1068 *
1069 * We need to allocate a new reservation window, if:
1070 * a) inode does not have a reservation window; or
1071 * b) last attempt to allocate a block from existing reservation
1072 * failed; or
1073 * c) we come here with a goal and with a reservation window
1074 *
1075 * We do not need to allocate a new reservation window if we come here
1076 * at the beginning with a goal and the goal is inside the window, or
1077 * we don't have a goal but already have a reservation window.
1078 * then we could go to allocate from the reservation window directly.
1079 */
1080 while (1) {
1081 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1082 !goal_in_my_reservation(&my_rsv->rsv_window,
1083 grp_goal, group, sb)) {
1084 if (my_rsv->rsv_goal_size < *count)
1085 my_rsv->rsv_goal_size = *count;
1086 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1087 group, bitmap_bh);
1088 if (ret < 0)
1089 break; /* failed */
1090
1091 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1092 grp_goal, group, sb))
1093 grp_goal = -1;
1094 } else if (grp_goal >= 0) {
1095 int curr = my_rsv->rsv_end -
1096 (grp_goal + group_first_block) + 1;
1097
1098 if (curr < *count)
1099 try_to_extend_reservation(my_rsv, sb,
1100 *count - curr);
1101 }
1102
1103 if ((my_rsv->rsv_start > group_last_block) ||
1104 (my_rsv->rsv_end < group_first_block)) {
1105 rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1106 BUG();
1107 }
1108 ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1109 &num, &my_rsv->rsv_window);
1110 if (ret >= 0) {
1111 my_rsv->rsv_alloc_hit += num;
1112 *count = num;
1113 break; /* succeed */
1114 }
1115 num = *count;
1116 }
1117 return ret;
1118}
1119
1120/**
1121 * ext2_has_free_blocks()
1122 * @sbi: in-core super block structure.
1123 *
1124 * Check if filesystem has at least 1 free block available for allocation.
1125 */
1126static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1127{
1128 ext2_fsblk_t free_blocks, root_blocks;
1129
1130 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1131 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1132 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1133 sbi->s_resuid != current->fsuid &&
1134 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1135 return 0;
1136 }
1137 return 1;
1138}
1139
1140/*
1141 * ext2_new_blocks() -- core block(s) allocation function
1142 * @inode: file inode
1143 * @goal: given target block(filesystem wide)
1144 * @count: target number of blocks to allocate
1145 * @errp: error code
1146 *
1147 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 * free, or there is a free block within 32 blocks of the goal, that block
1149 * is allocated. Otherwise a forward search is made for a free block; within
1150 * each block group the search first looks for an entire free byte in the block
1151 * bitmap, and then for any free bit if that fails.
1152 * This function also updates quota and i_blocks field.
1153 */
Martin J. Bligha686cd82007-10-16 23:30:46 -07001154ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1155 unsigned long *count, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
1157 struct buffer_head *bitmap_bh = NULL;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001158 struct buffer_head *gdp_bh;
1159 int group_no;
1160 int goal_group;
1161 ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1162 ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1163 ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
1164 int bgi; /* blockgroup iteration index */
1165 int performed_allocation = 0;
1166 ext2_grpblk_t free_blocks; /* number of free blocks in a group */
1167 struct super_block *sb;
1168 struct ext2_group_desc *gdp;
1169 struct ext2_super_block *es;
1170 struct ext2_sb_info *sbi;
1171 struct ext2_reserve_window_node *my_rsv = NULL;
1172 struct ext2_block_alloc_info *block_i;
1173 unsigned short windowsz = 0;
1174 unsigned long ngroups;
1175 unsigned long num = *count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Martin J. Bligha686cd82007-10-16 23:30:46 -07001177 *errp = -ENOSPC;
1178 sb = inode->i_sb;
1179 if (!sb) {
1180 printk("ext2_new_blocks: nonexistent device");
1181 return 0;
1182 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
Martin J. Bligha686cd82007-10-16 23:30:46 -07001184 /*
1185 * Check quota for allocation of this block.
1186 */
1187 if (DQUOT_ALLOC_BLOCK(inode, num)) {
1188 *errp = -EDQUOT;
1189 return 0;
1190 }
1191
1192 sbi = EXT2_SB(sb);
1193 es = EXT2_SB(sb)->s_es;
1194 ext2_debug("goal=%lu.\n", goal);
1195 /*
1196 * Allocate a block from reservation only when
1197 * filesystem is mounted with reservation(default,-o reservation), and
1198 * it's a regular file, and
1199 * the desired window size is greater than 0 (One could use ioctl
1200 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1201 * reservation on that particular file)
1202 */
1203 block_i = EXT2_I(inode)->i_block_alloc_info;
1204 if (block_i) {
1205 windowsz = block_i->rsv_window_node.rsv_goal_size;
1206 if (windowsz > 0)
1207 my_rsv = &block_i->rsv_window_node;
1208 }
1209
1210 if (!ext2_has_free_blocks(sbi)) {
1211 *errp = -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 goto out;
1213 }
1214
Martin J. Bligha686cd82007-10-16 23:30:46 -07001215 /*
1216 * First, test whether the goal block is free.
1217 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 if (goal < le32_to_cpu(es->s_first_data_block) ||
1219 goal >= le32_to_cpu(es->s_blocks_count))
1220 goal = le32_to_cpu(es->s_first_data_block);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001221 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1222 EXT2_BLOCKS_PER_GROUP(sb);
1223 goal_group = group_no;
1224retry_alloc:
1225 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1226 if (!gdp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 goto io_error;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001228
1229 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1230 /*
1231 * if there is not enough free blocks to make a new resevation
1232 * turn off reservation for this allocation
1233 */
1234 if (my_rsv && (free_blocks < windowsz)
1235 && (rsv_is_empty(&my_rsv->rsv_window)))
1236 my_rsv = NULL;
1237
1238 if (free_blocks > 0) {
1239 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1240 EXT2_BLOCKS_PER_GROUP(sb));
1241 bitmap_bh = read_block_bitmap(sb, group_no);
1242 if (!bitmap_bh)
1243 goto io_error;
1244 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1245 bitmap_bh, grp_target_blk,
1246 my_rsv, &num);
1247 if (grp_alloc_blk >= 0)
1248 goto allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 }
1250
Martin J. Bligha686cd82007-10-16 23:30:46 -07001251 ngroups = EXT2_SB(sb)->s_groups_count;
1252 smp_rmb();
1253
1254 /*
1255 * Now search the rest of the groups. We assume that
1256 * i and gdp correctly point to the last group visited.
1257 */
1258 for (bgi = 0; bgi < ngroups; bgi++) {
1259 group_no++;
1260 if (group_no >= ngroups)
1261 group_no = 0;
1262 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1263 if (!gdp)
1264 goto io_error;
1265
1266 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1267 /*
1268 * skip this group if the number of
1269 * free blocks is less than half of the reservation
1270 * window size.
1271 */
1272 if (free_blocks <= (windowsz/2))
1273 continue;
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 brelse(bitmap_bh);
1276 bitmap_bh = read_block_bitmap(sb, group_no);
1277 if (!bitmap_bh)
1278 goto io_error;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001279 /*
1280 * try to allocate block(s) from this group, without a goal(-1).
1281 */
1282 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1283 bitmap_bh, -1, my_rsv, &num);
1284 if (grp_alloc_blk >= 0)
1285 goto allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 /*
Martin J. Bligha686cd82007-10-16 23:30:46 -07001288 * We may end up a bogus ealier ENOSPC error due to
1289 * filesystem is "full" of reservations, but
1290 * there maybe indeed free blocks avaliable on disk
1291 * In this case, we just forget about the reservations
1292 * just do block allocation as without reservations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 */
Martin J. Bligha686cd82007-10-16 23:30:46 -07001294 if (my_rsv) {
1295 my_rsv = NULL;
1296 windowsz = 0;
1297 group_no = goal_group;
1298 goto retry_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 }
Martin J. Bligha686cd82007-10-16 23:30:46 -07001300 /* No space left on the device */
1301 *errp = -ENOSPC;
1302 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Martin J. Bligha686cd82007-10-16 23:30:46 -07001304allocated:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 ext2_debug("using block group %d(%d)\n",
Martin J. Bligha686cd82007-10-16 23:30:46 -07001307 group_no, gdp->bg_free_blocks_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
Martin J. Bligha686cd82007-10-16 23:30:46 -07001309 ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
Martin J. Bligha686cd82007-10-16 23:30:46 -07001311 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1312 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1313 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1314 EXT2_SB(sb)->s_itb_per_group) ||
1315 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -08001316 EXT2_SB(sb)->s_itb_per_group)) {
Martin J. Bligha686cd82007-10-16 23:30:46 -07001317 ext2_error(sb, "ext2_new_blocks",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 "Allocating block in system zone - "
Martin J. Bligha686cd82007-10-16 23:30:46 -07001319 "blocks from "E2FSBLK", length %lu",
1320 ret_block, num);
Aneesh Kumar K.V7f0adae2008-02-06 01:36:17 -08001321 goto out;
1322 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
Martin J. Bligha686cd82007-10-16 23:30:46 -07001324 performed_allocation = 1;
1325
1326 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1327 ext2_error(sb, "ext2_new_blocks",
1328 "block("E2FSBLK") >= blocks count(%d) - "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 "block_group = %d, es == %p ", ret_block,
1330 le32_to_cpu(es->s_blocks_count), group_no, es);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001331 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Martin J. Bligha686cd82007-10-16 23:30:46 -07001334 group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1335 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
1337 mark_buffer_dirty(bitmap_bh);
1338 if (sb->s_flags & MS_SYNCHRONOUS)
1339 sync_dirty_buffer(bitmap_bh);
1340
Martin J. Bligha686cd82007-10-16 23:30:46 -07001341 *errp = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 brelse(bitmap_bh);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001343 DQUOT_FREE_BLOCK(inode, *count-num);
1344 *count = num;
1345 return ret_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347io_error:
Martin J. Bligha686cd82007-10-16 23:30:46 -07001348 *errp = -EIO;
1349out:
1350 /*
1351 * Undo the block allocation
1352 */
1353 if (!performed_allocation)
1354 DQUOT_FREE_BLOCK(inode, *count);
1355 brelse(bitmap_bh);
1356 return 0;
1357}
1358
1359ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1360{
1361 unsigned long count = 1;
1362
1363 return ext2_new_blocks(inode, goal, &count, errp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364}
1365
Valerie Henson21730ee2006-06-25 05:48:12 -07001366#ifdef EXT2FS_DEBUG
1367
Philippe De Muyterfebfcf92007-10-16 23:26:15 -07001368static 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 -07001369
1370unsigned long ext2_count_free (struct buffer_head * map, unsigned int numchars)
1371{
1372 unsigned int i;
1373 unsigned long sum = 0;
1374
1375 if (!map)
1376 return (0);
1377 for (i = 0; i < numchars; i++)
1378 sum += nibblemap[map->b_data[i] & 0xf] +
1379 nibblemap[(map->b_data[i] >> 4) & 0xf];
1380 return (sum);
1381}
1382
1383#endif /* EXT2FS_DEBUG */
1384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385unsigned long ext2_count_free_blocks (struct super_block * sb)
1386{
1387 struct ext2_group_desc * desc;
1388 unsigned long desc_count = 0;
1389 int i;
1390#ifdef EXT2FS_DEBUG
1391 unsigned long bitmap_count, x;
1392 struct ext2_super_block *es;
1393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 es = EXT2_SB(sb)->s_es;
1395 desc_count = 0;
1396 bitmap_count = 0;
1397 desc = NULL;
1398 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1399 struct buffer_head *bitmap_bh;
1400 desc = ext2_get_group_desc (sb, i, NULL);
1401 if (!desc)
1402 continue;
1403 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1404 bitmap_bh = read_block_bitmap(sb, i);
1405 if (!bitmap_bh)
1406 continue;
1407
1408 x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1409 printk ("group %d: stored = %d, counted = %lu\n",
1410 i, le16_to_cpu(desc->bg_free_blocks_count), x);
1411 bitmap_count += x;
1412 brelse(bitmap_bh);
1413 }
1414 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1415 (long)le32_to_cpu(es->s_free_blocks_count),
1416 desc_count, bitmap_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 return bitmap_count;
1418#else
1419 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1420 desc = ext2_get_group_desc (sb, i, NULL);
1421 if (!desc)
1422 continue;
1423 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1424 }
1425 return desc_count;
1426#endif
1427}
1428
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429static inline int test_root(int a, int b)
1430{
1431 int num = b;
1432
1433 while (a > num)
1434 num *= b;
1435 return num == a;
1436}
1437
1438static int ext2_group_sparse(int group)
1439{
1440 if (group <= 1)
1441 return 1;
1442 return (test_root(group, 3) || test_root(group, 5) ||
1443 test_root(group, 7));
1444}
1445
1446/**
1447 * ext2_bg_has_super - number of blocks used by the superblock in group
1448 * @sb: superblock for filesystem
1449 * @group: group number to check
1450 *
1451 * Return the number of blocks used by the superblock (primary or backup)
1452 * in this group. Currently this will be only 0 or 1.
1453 */
1454int ext2_bg_has_super(struct super_block *sb, int group)
1455{
1456 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1457 !ext2_group_sparse(group))
1458 return 0;
1459 return 1;
1460}
1461
1462/**
1463 * ext2_bg_num_gdb - number of blocks used by the group table in group
1464 * @sb: superblock for filesystem
1465 * @group: group number to check
1466 *
1467 * Return the number of blocks used by the group descriptor table
1468 * (primary or backup) in this group. In the future there may be a
1469 * different number of descriptor blocks in each group.
1470 */
1471unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1472{
1473 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1474 !ext2_group_sparse(group))
1475 return 0;
1476 return EXT2_SB(sb)->s_gdb_count;
1477}
1478