]> git.proxmox.com Git - mirror_ubuntu-kernels.git/blame - fs/ext2/balloc.c
Merge tag 'rtc-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/abelloni/linux
[mirror_ubuntu-kernels.git] / fs / ext2 / balloc.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
1da177e4
LT
2/*
3 * linux/fs/ext2/balloc.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller (davem@caip.rutgers.edu), 1995
13 */
14
1da177e4
LT
15#include "ext2.h"
16#include <linux/quotaops.h>
5a0e3ad6 17#include <linux/slab.h>
1da177e4 18#include <linux/sched.h>
5b825c3a 19#include <linux/cred.h>
1da177e4 20#include <linux/buffer_head.h>
16f7e0fe 21#include <linux/capability.h>
1da177e4
LT
22
23/*
24 * balloc.c contains the blocks allocation and deallocation routines
25 */
26
27/*
28 * The free blocks are managed by bitmaps. A file system contains several
29 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
30 * block for inodes, N blocks for the inode table and data blocks.
31 *
32 * The file system contains group descriptors which are located after the
33 * super block. Each descriptor contains the number of the bitmap block and
34 * the free blocks count in the block. The descriptors are loaded in memory
e627432c 35 * when a file system is mounted (see ext2_fill_super).
1da177e4
LT
36 */
37
38
39#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
40
41struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
42 unsigned int block_group,
43 struct buffer_head ** bh)
44{
45 unsigned long group_desc;
46 unsigned long offset;
47 struct ext2_group_desc * desc;
48 struct ext2_sb_info *sbi = EXT2_SB(sb);
49
50 if (block_group >= sbi->s_groups_count) {
51 ext2_error (sb, "ext2_get_group_desc",
52 "block_group >= groups_count - "
53 "block_group = %d, groups_count = %lu",
54 block_group, sbi->s_groups_count);
55
56 return NULL;
57 }
58
59 group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
60 offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
61 if (!sbi->s_group_desc[group_desc]) {
62 ext2_error (sb, "ext2_get_group_desc",
63 "Group descriptor not loaded - "
64 "block_group = %d, group_desc = %lu, desc = %lu",
65 block_group, group_desc, offset);
66 return NULL;
67 }
68
69 desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
70 if (bh)
71 *bh = sbi->s_group_desc[group_desc];
72 return desc + offset;
73}
74
01584fa6
AK
75static int ext2_valid_block_bitmap(struct super_block *sb,
76 struct ext2_group_desc *desc,
77 unsigned int block_group,
78 struct buffer_head *bh)
79{
80 ext2_grpblk_t offset;
81 ext2_grpblk_t next_zero_bit;
82 ext2_fsblk_t bitmap_blk;
83 ext2_fsblk_t group_first_block;
84
85 group_first_block = ext2_group_first_block_no(sb, block_group);
86
87 /* check whether block bitmap block number is set */
88 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
89 offset = bitmap_blk - group_first_block;
90 if (!ext2_test_bit(offset, bh->b_data))
91 /* bad block bitmap */
92 goto err_out;
93
94 /* check whether the inode bitmap block number is set */
95 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
96 offset = bitmap_blk - group_first_block;
97 if (!ext2_test_bit(offset, bh->b_data))
98 /* bad block bitmap */
99 goto err_out;
100
101 /* check whether the inode table block number is set */
102 bitmap_blk = le32_to_cpu(desc->bg_inode_table);
103 offset = bitmap_blk - group_first_block;
104 next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
105 offset + EXT2_SB(sb)->s_itb_per_group,
106 offset);
107 if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
108 /* good bitmap for inode tables */
109 return 1;
110
111err_out:
605afd60 112 ext2_error(sb, __func__,
01584fa6
AK
113 "Invalid block bitmap - "
114 "block_group = %d, block = %lu",
115 block_group, bitmap_blk);
116 return 0;
117}
118
1da177e4 119/*
01584fa6
AK
120 * Read the bitmap for a given block_group,and validate the
121 * bits for block/inode/inode tables are set in the bitmaps
1da177e4
LT
122 *
123 * Return buffer_head on success or NULL in case of failure.
124 */
125static struct buffer_head *
126read_block_bitmap(struct super_block *sb, unsigned int block_group)
127{
128 struct ext2_group_desc * desc;
129 struct buffer_head * bh = NULL;
01584fa6
AK
130 ext2_fsblk_t bitmap_blk;
131
132 desc = ext2_get_group_desc(sb, block_group, NULL);
1da177e4 133 if (!desc)
01584fa6
AK
134 return NULL;
135 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
136 bh = sb_getblk(sb, bitmap_blk);
137 if (unlikely(!bh)) {
605afd60 138 ext2_error(sb, __func__,
01584fa6
AK
139 "Cannot read block bitmap - "
140 "block_group = %d, block_bitmap = %u",
141 block_group, le32_to_cpu(desc->bg_block_bitmap));
142 return NULL;
143 }
144 if (likely(bh_uptodate_or_lock(bh)))
145 return bh;
146
147 if (bh_submit_read(bh) < 0) {
148 brelse(bh);
605afd60 149 ext2_error(sb, __func__,
1da177e4
LT
150 "Cannot read block bitmap - "
151 "block_group = %d, block_bitmap = %u",
152 block_group, le32_to_cpu(desc->bg_block_bitmap));
01584fa6
AK
153 return NULL;
154 }
01584fa6 155
8b915825
AK
156 ext2_valid_block_bitmap(sb, desc, block_group, bh);
157 /*
158 * file system mounted not to panic on error, continue with corrupt
159 * bitmap
160 */
0b832a4b 161 return bh;
1da177e4
LT
162}
163
a686cd89 164static void group_adjust_blocks(struct super_block *sb, int group_no,
1da177e4
LT
165 struct ext2_group_desc *desc, struct buffer_head *bh, int count)
166{
167 if (count) {
168 struct ext2_sb_info *sbi = EXT2_SB(sb);
169 unsigned free_blocks;
170
171 spin_lock(sb_bgl_lock(sbi, group_no));
172 free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
173 desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
174 spin_unlock(sb_bgl_lock(sbi, group_no));
1da177e4
LT
175 mark_buffer_dirty(bh);
176 }
177}
178
a686cd89
MB
179/*
180 * The reservation window structure operations
181 * --------------------------------------------
182 * Operations include:
183 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
184 *
185 * We use a red-black tree to represent per-filesystem reservation
186 * windows.
187 *
188 */
189
190/**
191 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
c53ec7bc 192 * @root: root of per-filesystem reservation rb tree
a686cd89
MB
193 * @verbose: verbose mode
194 * @fn: function which wishes to dump the reservation map
195 *
196 * If verbose is turned on, it will print the whole block reservation
197 * windows(start, end). Otherwise, it will only print out the "bad" windows,
198 * those windows that overlap with their immediate neighbors.
199 */
200#if 1
201static void __rsv_window_dump(struct rb_root *root, int verbose,
202 const char *fn)
203{
204 struct rb_node *n;
205 struct ext2_reserve_window_node *rsv, *prev;
206 int bad;
207
208restart:
209 n = rb_first(root);
210 bad = 0;
211 prev = NULL;
212
213 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
214 while (n) {
215 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
216 if (verbose)
217 printk("reservation window 0x%p "
218 "start: %lu, end: %lu\n",
219 rsv, rsv->rsv_start, rsv->rsv_end);
220 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
221 printk("Bad reservation %p (start >= end)\n",
222 rsv);
223 bad = 1;
224 }
225 if (prev && prev->rsv_end >= rsv->rsv_start) {
226 printk("Bad reservation %p (prev->end >= start)\n",
227 rsv);
228 bad = 1;
229 }
230 if (bad) {
231 if (!verbose) {
232 printk("Restarting reservation walk in verbose mode\n");
233 verbose = 1;
234 goto restart;
235 }
236 }
237 n = rb_next(n);
238 prev = rsv;
239 }
240 printk("Window map complete.\n");
2c11619a 241 BUG_ON(bad);
a686cd89
MB
242}
243#define rsv_window_dump(root, verbose) \
605afd60 244 __rsv_window_dump((root), (verbose), __func__)
a686cd89
MB
245#else
246#define rsv_window_dump(root, verbose) do {} while (0)
247#endif
248
249/**
250 * goal_in_my_reservation()
251 * @rsv: inode's reservation window
252 * @grp_goal: given goal block relative to the allocation block group
253 * @group: the current allocation block group
254 * @sb: filesystem super block
255 *
256 * Test if the given goal block (group relative) is within the file's
257 * own block reservation window range.
258 *
259 * If the reservation window is outside the goal allocation group, return 0;
260 * grp_goal (given goal block) could be -1, which means no specific
261 * goal block. In this case, always return 1.
262 * If the goal block is within the reservation window, return 1;
263 * otherwise, return 0;
264 */
265static int
266goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
267 unsigned int group, struct super_block * sb)
268{
269 ext2_fsblk_t group_first_block, group_last_block;
270
271 group_first_block = ext2_group_first_block_no(sb, group);
90f3741c 272 group_last_block = ext2_group_last_block_no(sb, group);
a686cd89
MB
273
274 if ((rsv->_rsv_start > group_last_block) ||
275 (rsv->_rsv_end < group_first_block))
276 return 0;
277 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
278 || (grp_goal + group_first_block > rsv->_rsv_end)))
279 return 0;
280 return 1;
281}
282
283/**
284 * search_reserve_window()
c53ec7bc 285 * @root: root of reservation tree
a686cd89
MB
286 * @goal: target allocation block
287 *
288 * Find the reserved window which includes the goal, or the previous one
289 * if the goal is not in any window.
290 * Returns NULL if there are no windows or if all windows start after the goal.
291 */
292static struct ext2_reserve_window_node *
293search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
294{
295 struct rb_node *n = root->rb_node;
296 struct ext2_reserve_window_node *rsv;
297
298 if (!n)
299 return NULL;
300
301 do {
302 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
303
304 if (goal < rsv->rsv_start)
305 n = n->rb_left;
306 else if (goal > rsv->rsv_end)
307 n = n->rb_right;
308 else
309 return rsv;
310 } while (n);
311 /*
312 * We've fallen off the end of the tree: the goal wasn't inside
313 * any particular node. OK, the previous node must be to one
314 * side of the interval containing the goal. If it's the RHS,
315 * we need to back up one.
316 */
317 if (rsv->rsv_start > goal) {
318 n = rb_prev(&rsv->rsv_node);
319 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
320 }
321 return rsv;
322}
323
324/*
325 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
326 * @sb: super block
327 * @rsv: reservation window to add
328 *
329 * Must be called with rsv_lock held.
330 */
331void ext2_rsv_window_add(struct super_block *sb,
332 struct ext2_reserve_window_node *rsv)
333{
334 struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
335 struct rb_node *node = &rsv->rsv_node;
336 ext2_fsblk_t start = rsv->rsv_start;
337
338 struct rb_node ** p = &root->rb_node;
339 struct rb_node * parent = NULL;
340 struct ext2_reserve_window_node *this;
341
342 while (*p)
343 {
344 parent = *p;
345 this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
346
347 if (start < this->rsv_start)
348 p = &(*p)->rb_left;
349 else if (start > this->rsv_end)
350 p = &(*p)->rb_right;
351 else {
352 rsv_window_dump(root, 1);
353 BUG();
354 }
355 }
356
357 rb_link_node(node, parent, p);
358 rb_insert_color(node, root);
359}
360
361/**
362 * rsv_window_remove() -- unlink a window from the reservation rb tree
363 * @sb: super block
364 * @rsv: reservation window to remove
365 *
366 * Mark the block reservation window as not allocated, and unlink it
367 * from the filesystem reservation window rb tree. Must be called with
368 * rsv_lock held.
369 */
370static void rsv_window_remove(struct super_block *sb,
371 struct ext2_reserve_window_node *rsv)
372{
373 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
374 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
375 rsv->rsv_alloc_hit = 0;
376 rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
377}
378
379/*
380 * rsv_is_empty() -- Check if the reservation window is allocated.
381 * @rsv: given reservation window to check
382 *
383 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
384 */
385static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
386{
387 /* a valid reservation end block could not be 0 */
388 return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
389}
390
391/**
392 * ext2_init_block_alloc_info()
393 * @inode: file inode structure
394 *
395 * Allocate and initialize the reservation window structure, and
396 * link the window to the ext2 inode structure at last
397 *
398 * The reservation window structure is only dynamically allocated
399 * and linked to ext2 inode the first time the open file
400 * needs a new block. So, before every ext2_new_block(s) call, for
401 * regular files, we should check whether the reservation window
402 * structure exists or not. In the latter case, this function is called.
403 * Fail to do so will result in block reservation being turned off for that
404 * open file.
405 *
406 * This function is called from ext2_get_blocks_handle(), also called
407 * when setting the reservation window size through ioctl before the file
408 * is open for write (needs block allocation).
409 *
410 * Needs truncate_mutex protection prior to calling this function.
411 */
412void ext2_init_block_alloc_info(struct inode *inode)
413{
414 struct ext2_inode_info *ei = EXT2_I(inode);
6e3d6ca0 415 struct ext2_block_alloc_info *block_i;
a686cd89
MB
416 struct super_block *sb = inode->i_sb;
417
418 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
419 if (block_i) {
420 struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
421
422 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
423 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
424
425 /*
426 * if filesystem is mounted with NORESERVATION, the goal
427 * reservation window size is set to zero to indicate
428 * block reservation is off
429 */
430 if (!test_opt(sb, RESERVATION))
431 rsv->rsv_goal_size = 0;
432 else
433 rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
434 rsv->rsv_alloc_hit = 0;
435 block_i->last_alloc_logical_block = 0;
436 block_i->last_alloc_physical_block = 0;
437 }
438 ei->i_block_alloc_info = block_i;
439}
440
441/**
442 * ext2_discard_reservation()
443 * @inode: inode
444 *
445 * Discard(free) block reservation window on last file close, or truncate
446 * or at last iput().
447 *
448 * It is being called in three cases:
449 * ext2_release_file(): last writer closes the file
450 * ext2_clear_inode(): last iput(), when nobody links to this file.
451 * ext2_truncate(): when the block indirect map is about to change.
452 */
453void ext2_discard_reservation(struct inode *inode)
454{
455 struct ext2_inode_info *ei = EXT2_I(inode);
456 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
457 struct ext2_reserve_window_node *rsv;
458 spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
459
460 if (!block_i)
461 return;
462
463 rsv = &block_i->rsv_window_node;
464 if (!rsv_is_empty(&rsv->rsv_window)) {
465 spin_lock(rsv_lock);
466 if (!rsv_is_empty(&rsv->rsv_window))
467 rsv_window_remove(inode->i_sb, rsv);
468 spin_unlock(rsv_lock);
469 }
470}
471
472/**
03248766 473 * ext2_free_blocks() -- Free given blocks and update quota and i_blocks
a686cd89 474 * @inode: inode
4907cb7b 475 * @block: start physical block to free
a686cd89
MB
476 * @count: number of blocks to free
477 */
1da177e4
LT
478void ext2_free_blocks (struct inode * inode, unsigned long block,
479 unsigned long count)
480{
481 struct buffer_head *bitmap_bh = NULL;
482 struct buffer_head * bh2;
483 unsigned long block_group;
484 unsigned long bit;
485 unsigned long i;
486 unsigned long overflow;
487 struct super_block * sb = inode->i_sb;
488 struct ext2_sb_info * sbi = EXT2_SB(sb);
489 struct ext2_group_desc * desc;
490 struct ext2_super_block * es = sbi->s_es;
491 unsigned freed = 0, group_freed;
492
b6aeffc5 493 if (!ext2_data_block_valid(sbi, block, count)) {
1da177e4
LT
494 ext2_error (sb, "ext2_free_blocks",
495 "Freeing blocks not in datazone - "
496 "block = %lu, count = %lu", block, count);
497 goto error_return;
498 }
499
500 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
501
502do_more:
503 overflow = 0;
504 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
505 EXT2_BLOCKS_PER_GROUP(sb);
506 bit = (block - le32_to_cpu(es->s_first_data_block)) %
507 EXT2_BLOCKS_PER_GROUP(sb);
508 /*
509 * Check to see if we are freeing blocks across a group
510 * boundary.
511 */
512 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
513 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
514 count -= overflow;
515 }
516 brelse(bitmap_bh);
517 bitmap_bh = read_block_bitmap(sb, block_group);
518 if (!bitmap_bh)
519 goto error_return;
520
521 desc = ext2_get_group_desc (sb, block_group, &bh2);
522 if (!desc)
523 goto error_return;
524
525 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
526 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
527 in_range (block, le32_to_cpu(desc->bg_inode_table),
528 sbi->s_itb_per_group) ||
529 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
7f0adaec 530 sbi->s_itb_per_group)) {
1da177e4
LT
531 ext2_error (sb, "ext2_free_blocks",
532 "Freeing blocks in system zones - "
533 "Block = %lu, count = %lu",
534 block, count);
7f0adaec
AK
535 goto error_return;
536 }
1da177e4
LT
537
538 for (i = 0, group_freed = 0; i < count; i++) {
539 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
540 bit + i, bitmap_bh->b_data)) {
605afd60 541 ext2_error(sb, __func__,
1da177e4
LT
542 "bit already cleared for block %lu", block + i);
543 } else {
544 group_freed++;
545 }
546 }
547
548 mark_buffer_dirty(bitmap_bh);
1751e8a6 549 if (sb->s_flags & SB_SYNCHRONOUS)
1da177e4
LT
550 sync_dirty_buffer(bitmap_bh);
551
a686cd89 552 group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
1da177e4
LT
553 freed += group_freed;
554
555 if (overflow) {
556 block += count;
557 count = overflow;
558 goto do_more;
559 }
560error_return:
561 brelse(bitmap_bh);
8e3dffc6 562 if (freed) {
712ddc52 563 percpu_counter_add(&sbi->s_freeblocks_counter, freed);
8e3dffc6
WS
564 dquot_free_block_nodirty(inode, freed);
565 mark_inode_dirty(inode);
566 }
1da177e4
LT
567}
568
a686cd89
MB
569/**
570 * bitmap_search_next_usable_block()
571 * @start: the starting block (group relative) of the search
572 * @bh: bufferhead contains the block group bitmap
573 * @maxblocks: the ending block (group relative) of the reservation
574 *
575 * The bitmap search --- search forward through the actual bitmap on disk until
576 * we find a bit free.
577 */
578static ext2_grpblk_t
579bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
580 ext2_grpblk_t maxblocks)
1da177e4 581{
a686cd89 582 ext2_grpblk_t next;
1da177e4 583
a686cd89
MB
584 next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
585 if (next >= maxblocks)
586 return -1;
587 return next;
588}
1da177e4 589
a686cd89
MB
590/**
591 * find_next_usable_block()
592 * @start: the starting block (group relative) to find next
593 * allocatable block in bitmap.
594 * @bh: bufferhead contains the block group bitmap
595 * @maxblocks: the ending block (group relative) for the search
596 *
597 * Find an allocatable block in a bitmap. We perform the "most
598 * appropriate allocation" algorithm of looking for a free block near
599 * the initial goal; then for a free byte somewhere in the bitmap;
600 * then for any free bit in the bitmap.
601 */
602static ext2_grpblk_t
603find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
604{
605 ext2_grpblk_t here, next;
606 char *p, *r;
607
608 if (start > 0) {
1da177e4
LT
609 /*
610 * The goal was occupied; search forward for a free
611 * block within the next XX blocks.
612 *
613 * end_goal is more or less random, but it has to be
614 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
615 * next 64-bit boundary is simple..
616 */
a686cd89
MB
617 ext2_grpblk_t end_goal = (start + 63) & ~63;
618 if (end_goal > maxblocks)
619 end_goal = maxblocks;
620 here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
621 if (here < end_goal)
622 return here;
623 ext2_debug("Bit not found near goal\n");
624 }
625
626 here = start;
627 if (here < 0)
628 here = 0;
629
630 p = ((char *)bh->b_data) + (here >> 3);
631 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
632 next = (r - ((char *)bh->b_data)) << 3;
633
634 if (next < maxblocks && next >= here)
635 return next;
636
637 here = bitmap_search_next_usable_block(here, bh, maxblocks);
638 return here;
639}
640
26f78b7a 641/**
a686cd89
MB
642 * ext2_try_to_allocate()
643 * @sb: superblock
a686cd89
MB
644 * @group: given allocation block group
645 * @bitmap_bh: bufferhead holds the block bitmap
646 * @grp_goal: given target block within the group
647 * @count: target number of blocks to allocate
648 * @my_rsv: reservation window
649 *
650 * Attempt to allocate blocks within a give range. Set the range of allocation
651 * first, then find the first free bit(s) from the bitmap (within the range),
652 * and at last, allocate the blocks by claiming the found free bit as allocated.
653 *
654 * To set the range of this allocation:
655 * if there is a reservation window, only try to allocate block(s)
656 * from the file's own reservation window;
657 * Otherwise, the allocation range starts from the give goal block,
658 * ends at the block group's last block.
659 *
660 * If we failed to allocate the desired block then we may end up crossing to a
661 * new bitmap.
662 */
663static int
664ext2_try_to_allocate(struct super_block *sb, int group,
665 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
666 unsigned long *count,
667 struct ext2_reserve_window *my_rsv)
668{
90f3741c
CX
669 ext2_fsblk_t group_first_block = ext2_group_first_block_no(sb, group);
670 ext2_fsblk_t group_last_block = ext2_group_last_block_no(sb, group);
a686cd89
MB
671 ext2_grpblk_t start, end;
672 unsigned long num = 0;
673
cf4eb321
JK
674 start = 0;
675 end = group_last_block - group_first_block + 1;
a686cd89
MB
676 /* we do allocation within the reservation window if we have a window */
677 if (my_rsv) {
a686cd89
MB
678 if (my_rsv->_rsv_start >= group_first_block)
679 start = my_rsv->_rsv_start - group_first_block;
cf4eb321
JK
680 if (my_rsv->_rsv_end < group_last_block)
681 end = my_rsv->_rsv_end - group_first_block + 1;
682 if (grp_goal < start || grp_goal >= end)
a686cd89 683 grp_goal = -1;
a686cd89 684 }
a686cd89
MB
685 BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
686
a686cd89
MB
687 if (grp_goal < 0) {
688 grp_goal = find_next_usable_block(start, bitmap_bh, end);
689 if (grp_goal < 0)
690 goto fail_access;
691 if (!my_rsv) {
692 int i;
693
694 for (i = 0; i < 7 && grp_goal > start &&
695 !ext2_test_bit(grp_goal - 1,
696 bitmap_bh->b_data);
697 i++, grp_goal--)
698 ;
699 }
700 }
a686cd89 701
44dd6161
CX
702 for (; num < *count && grp_goal < end; grp_goal++) {
703 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
a686cd89 704 grp_goal, bitmap_bh->b_data)) {
44dd6161
CX
705 if (num == 0)
706 continue;
707 break;
708 }
a686cd89 709 num++;
a686cd89 710 }
44dd6161
CX
711
712 if (num == 0)
713 goto fail_access;
714
a686cd89
MB
715 *count = num;
716 return grp_goal - num;
717fail_access:
a686cd89
MB
718 return -1;
719}
720
721/**
722 * find_next_reservable_window():
723 * find a reservable space within the given range.
724 * It does not allocate the reservation window for now:
725 * alloc_new_reservation() will do the work later.
726 *
727 * @search_head: the head of the searching list;
728 * This is not necessarily the list head of the whole filesystem
729 *
730 * We have both head and start_block to assist the search
731 * for the reservable space. The list starts from head,
732 * but we will shift to the place where start_block is,
733 * then start from there, when looking for a reservable space.
734 *
355b9aae 735 * @sb: the super block.
a686cd89 736 *
355b9aae 737 * @start_block: the first block we consider to start the real search from
a686cd89
MB
738 *
739 * @last_block:
740 * the maximum block number that our goal reservable space
741 * could start from. This is normally the last block in this
742 * group. The search will end when we found the start of next
743 * possible reservable space is out of this boundary.
744 * This could handle the cross boundary reservation window
745 * request.
746 *
747 * basically we search from the given range, rather than the whole
748 * reservation double linked list, (start_block, last_block)
749 * to find a free region that is of my size and has not
750 * been reserved.
751 *
752 */
753static int find_next_reservable_window(
754 struct ext2_reserve_window_node *search_head,
755 struct ext2_reserve_window_node *my_rsv,
756 struct super_block * sb,
757 ext2_fsblk_t start_block,
758 ext2_fsblk_t last_block)
759{
760 struct rb_node *next;
761 struct ext2_reserve_window_node *rsv, *prev;
762 ext2_fsblk_t cur;
763 int size = my_rsv->rsv_goal_size;
764
765 /* TODO: make the start of the reservation window byte-aligned */
766 /* cur = *start_block & ~7;*/
767 cur = start_block;
768 rsv = search_head;
769 if (!rsv)
770 return -1;
771
772 while (1) {
773 if (cur <= rsv->rsv_end)
774 cur = rsv->rsv_end + 1;
775
776 /* TODO?
777 * in the case we could not find a reservable space
778 * that is what is expected, during the re-search, we could
779 * remember what's the largest reservable space we could have
780 * and return that one.
781 *
782 * For now it will fail if we could not find the reservable
783 * space with expected-size (or more)...
784 */
785 if (cur > last_block)
786 return -1; /* fail */
787
788 prev = rsv;
789 next = rb_next(&rsv->rsv_node);
790 rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
791
1da177e4 792 /*
a686cd89
MB
793 * Reached the last reservation, we can just append to the
794 * previous one.
1da177e4 795 */
a686cd89
MB
796 if (!next)
797 break;
798
799 if (cur + size <= rsv->rsv_start) {
800 /*
801 * Found a reserveable space big enough. We could
802 * have a reservation across the group boundary here
803 */
804 break;
805 }
1da177e4 806 }
a686cd89
MB
807 /*
808 * we come here either :
809 * when we reach the end of the whole list,
810 * and there is empty reservable space after last entry in the list.
811 * append it to the end of the list.
812 *
813 * or we found one reservable space in the middle of the list,
814 * return the reservation window that we could append to.
815 * succeed.
816 */
817
818 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
819 rsv_window_remove(sb, my_rsv);
820
821 /*
25985edc 822 * Let's book the whole available window for now. We will check the
a686cd89
MB
823 * disk bitmap later and then, if there are free blocks then we adjust
824 * the window size if it's larger than requested.
825 * Otherwise, we will remove this node from the tree next time
826 * call find_next_reservable_window.
827 */
828 my_rsv->rsv_start = cur;
829 my_rsv->rsv_end = cur + size - 1;
830 my_rsv->rsv_alloc_hit = 0;
831
832 if (prev != my_rsv)
833 ext2_rsv_window_add(sb, my_rsv);
834
835 return 0;
836}
837
838/**
839 * alloc_new_reservation()--allocate a new reservation window
840 *
841 * To make a new reservation, we search part of the filesystem
842 * reservation list (the list that inside the group). We try to
843 * allocate a new reservation window near the allocation goal,
844 * or the beginning of the group, if there is no goal.
845 *
846 * We first find a reservable space after the goal, then from
847 * there, we check the bitmap for the first free block after
848 * it. If there is no free block until the end of group, then the
849 * whole group is full, we failed. Otherwise, check if the free
850 * block is inside the expected reservable space, if so, we
851 * succeed.
852 * If the first free block is outside the reservable space, then
853 * start from the first free block, we search for next available
854 * space, and go on.
855 *
856 * on succeed, a new reservation will be found and inserted into the list
857 * It contains at least one free block, and it does not overlap with other
858 * reservation windows.
859 *
860 * failed: we failed to find a reservation window in this group
861 *
c53ec7bc 862 * @my_rsv: the reservation
a686cd89
MB
863 *
864 * @grp_goal: The goal (group-relative). It is where the search for a
865 * free reservable space should start from.
866 * if we have a goal(goal >0 ), then start from there,
867 * no goal(goal = -1), we start from the first block
868 * of the group.
869 *
870 * @sb: the super block
871 * @group: the group we are trying to allocate in
872 * @bitmap_bh: the block group block bitmap
873 *
874 */
875static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
876 ext2_grpblk_t grp_goal, struct super_block *sb,
877 unsigned int group, struct buffer_head *bitmap_bh)
878{
879 struct ext2_reserve_window_node *search_head;
880 ext2_fsblk_t group_first_block, group_end_block, start_block;
881 ext2_grpblk_t first_free_block;
882 struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
883 unsigned long size;
884 int ret;
885 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
886
887 group_first_block = ext2_group_first_block_no(sb, group);
90f3741c 888 group_end_block = ext2_group_last_block_no(sb, group);
1da177e4 889
a686cd89
MB
890 if (grp_goal < 0)
891 start_block = group_first_block;
892 else
893 start_block = grp_goal + group_first_block;
894
895 size = my_rsv->rsv_goal_size;
896
897 if (!rsv_is_empty(&my_rsv->rsv_window)) {
898 /*
899 * if the old reservation is cross group boundary
900 * and if the goal is inside the old reservation window,
901 * we will come here when we just failed to allocate from
902 * the first part of the window. We still have another part
903 * that belongs to the next group. In this case, there is no
904 * point to discard our window and try to allocate a new one
905 * in this group(which will fail). we should
906 * keep the reservation window, just simply move on.
907 *
908 * Maybe we could shift the start block of the reservation
909 * window to the first block of next group.
1da177e4 910 */
a686cd89
MB
911
912 if ((my_rsv->rsv_start <= group_end_block) &&
913 (my_rsv->rsv_end > group_end_block) &&
914 (start_block >= my_rsv->rsv_start))
915 return -1;
916
917 if ((my_rsv->rsv_alloc_hit >
918 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
919 /*
920 * if the previously allocation hit ratio is
921 * greater than 1/2, then we double the size of
922 * the reservation window the next time,
923 * otherwise we keep the same size window
924 */
925 size = size * 2;
926 if (size > EXT2_MAX_RESERVE_BLOCKS)
927 size = EXT2_MAX_RESERVE_BLOCKS;
928 my_rsv->rsv_goal_size= size;
929 }
1da177e4
LT
930 }
931
a686cd89
MB
932 spin_lock(rsv_lock);
933 /*
934 * shift the search start to the window near the goal block
935 */
936 search_head = search_reserve_window(fs_rsv_root, start_block);
937
938 /*
939 * find_next_reservable_window() simply finds a reservable window
940 * inside the given range(start_block, group_end_block).
941 *
942 * To make sure the reservation window has a free bit inside it, we
943 * need to check the bitmap after we found a reservable window.
944 */
945retry:
946 ret = find_next_reservable_window(search_head, my_rsv, sb,
947 start_block, group_end_block);
948
949 if (ret == -1) {
950 if (!rsv_is_empty(&my_rsv->rsv_window))
951 rsv_window_remove(sb, my_rsv);
952 spin_unlock(rsv_lock);
953 return -1;
1da177e4 954 }
a686cd89
MB
955
956 /*
957 * On success, find_next_reservable_window() returns the
958 * reservation window where there is a reservable space after it.
959 * Before we reserve this reservable space, we need
960 * to make sure there is at least a free block inside this region.
961 *
962 * Search the first free bit on the block bitmap. Search starts from
963 * the start block of the reservable space we just found.
964 */
965 spin_unlock(rsv_lock);
966 first_free_block = bitmap_search_next_usable_block(
967 my_rsv->rsv_start - group_first_block,
968 bitmap_bh, group_end_block - group_first_block + 1);
969
970 if (first_free_block < 0) {
971 /*
972 * no free block left on the bitmap, no point
973 * to reserve the space. return failed.
974 */
975 spin_lock(rsv_lock);
976 if (!rsv_is_empty(&my_rsv->rsv_window))
977 rsv_window_remove(sb, my_rsv);
978 spin_unlock(rsv_lock);
979 return -1; /* failed */
980 }
981
982 start_block = first_free_block + group_first_block;
983 /*
984 * check if the first free block is within the
985 * free space we just reserved
986 */
987 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
988 return 0; /* success */
989 /*
990 * if the first free bit we found is out of the reservable space
991 * continue search for next reservable space,
992 * start from where the free block is,
993 * we also shift the list head to where we stopped last time
994 */
995 search_head = my_rsv;
996 spin_lock(rsv_lock);
997 goto retry;
998}
999
1000/**
1001 * try_to_extend_reservation()
1002 * @my_rsv: given reservation window
1003 * @sb: super block
1004 * @size: the delta to extend
1005 *
1006 * Attempt to expand the reservation window large enough to have
1007 * required number of free blocks
1008 *
1009 * Since ext2_try_to_allocate() will always allocate blocks within
1010 * the reservation window range, if the window size is too small,
1011 * multiple blocks allocation has to stop at the end of the reservation
1012 * window. To make this more efficient, given the total number of
1013 * blocks needed and the current size of the window, we try to
1014 * expand the reservation window size if necessary on a best-effort
1015 * basis before ext2_new_blocks() tries to allocate blocks.
1016 */
1017static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1018 struct super_block *sb, int size)
1019{
1020 struct ext2_reserve_window_node *next_rsv;
1021 struct rb_node *next;
1022 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1023
1024 if (!spin_trylock(rsv_lock))
1025 return;
1026
1027 next = rb_next(&my_rsv->rsv_node);
1028
1029 if (!next)
1030 my_rsv->rsv_end += size;
1031 else {
1032 next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1033
1034 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1035 my_rsv->rsv_end += size;
1036 else
1037 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1038 }
1039 spin_unlock(rsv_lock);
1040}
1041
1042/**
1043 * ext2_try_to_allocate_with_rsv()
1044 * @sb: superblock
1045 * @group: given allocation block group
1046 * @bitmap_bh: bufferhead holds the block bitmap
1047 * @grp_goal: given target block within the group
1048 * @count: target number of blocks to allocate
1049 * @my_rsv: reservation window
1050 *
1051 * This is the main function used to allocate a new block and its reservation
1052 * window.
1053 *
1054 * Each time when a new block allocation is need, first try to allocate from
1055 * its own reservation. If it does not have a reservation window, instead of
1056 * looking for a free bit on bitmap first, then look up the reservation list to
1057 * see if it is inside somebody else's reservation window, we try to allocate a
1058 * reservation window for it starting from the goal first. Then do the block
1059 * allocation within the reservation window.
1060 *
1061 * This will avoid keeping on searching the reservation list again and
1062 * again when somebody is looking for a free block (without
1063 * reservation), and there are lots of free blocks, but they are all
1064 * being reserved.
1065 *
1066 * We use a red-black tree for the per-filesystem reservation list.
1067 */
1068static ext2_grpblk_t
1069ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1070 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1071 struct ext2_reserve_window_node * my_rsv,
1072 unsigned long *count)
1073{
1074 ext2_fsblk_t group_first_block, group_last_block;
1075 ext2_grpblk_t ret = 0;
1076 unsigned long num = *count;
1077
1078 /*
1079 * we don't deal with reservation when
1080 * filesystem is mounted without reservation
1081 * or the file is not a regular file
1082 * or last attempt to allocate a block with reservation turned on failed
1083 */
1084 if (my_rsv == NULL) {
1085 return ext2_try_to_allocate(sb, group, bitmap_bh,
1086 grp_goal, count, NULL);
1087 }
1088 /*
1089 * grp_goal is a group relative block number (if there is a goal)
1090 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1091 * first block is a filesystem wide block number
1092 * first block is the block number of the first block in this group
1093 */
1094 group_first_block = ext2_group_first_block_no(sb, group);
90f3741c 1095 group_last_block = ext2_group_last_block_no(sb, group);
a686cd89
MB
1096
1097 /*
1098 * Basically we will allocate a new block from inode's reservation
1099 * window.
1100 *
1101 * We need to allocate a new reservation window, if:
1102 * a) inode does not have a reservation window; or
1103 * b) last attempt to allocate a block from existing reservation
1104 * failed; or
1105 * c) we come here with a goal and with a reservation window
1106 *
1107 * We do not need to allocate a new reservation window if we come here
1108 * at the beginning with a goal and the goal is inside the window, or
1109 * we don't have a goal but already have a reservation window.
1110 * then we could go to allocate from the reservation window directly.
1111 */
1112 while (1) {
1113 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1114 !goal_in_my_reservation(&my_rsv->rsv_window,
1115 grp_goal, group, sb)) {
1116 if (my_rsv->rsv_goal_size < *count)
1117 my_rsv->rsv_goal_size = *count;
1118 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1119 group, bitmap_bh);
1120 if (ret < 0)
1121 break; /* failed */
1122
1123 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1124 grp_goal, group, sb))
1125 grp_goal = -1;
1126 } else if (grp_goal >= 0) {
1127 int curr = my_rsv->rsv_end -
1128 (grp_goal + group_first_block) + 1;
1129
1130 if (curr < *count)
1131 try_to_extend_reservation(my_rsv, sb,
1132 *count - curr);
1133 }
1134
1135 if ((my_rsv->rsv_start > group_last_block) ||
1136 (my_rsv->rsv_end < group_first_block)) {
1137 rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1138 BUG();
1139 }
1140 ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1141 &num, &my_rsv->rsv_window);
1142 if (ret >= 0) {
1143 my_rsv->rsv_alloc_hit += num;
1144 *count = num;
1145 break; /* succeed */
1146 }
1147 num = *count;
1148 }
1149 return ret;
1150}
1151
1152/**
1153 * ext2_has_free_blocks()
1154 * @sbi: in-core super block structure.
1155 *
1156 * Check if filesystem has at least 1 free block available for allocation.
1157 */
1158static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1159{
1160 ext2_fsblk_t free_blocks, root_blocks;
1161
1162 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1163 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1164 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
b8a9f9e1
EB
1165 !uid_eq(sbi->s_resuid, current_fsuid()) &&
1166 (gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1167 !in_group_p (sbi->s_resgid))) {
a686cd89
MB
1168 return 0;
1169 }
1170 return 1;
1da177e4
LT
1171}
1172
ff0031d8
CM
1173/*
1174 * Returns 1 if the passed-in block region is valid; 0 if some part overlaps
1fe03415 1175 * with filesystem metadata blocks.
ff0031d8
CM
1176 */
1177int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk,
1178 unsigned int count)
1179{
1180 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
e5d39597
CX
1181 (start_blk + count - 1 < start_blk) ||
1182 (start_blk + count - 1 >= le32_to_cpu(sbi->s_es->s_blocks_count)))
ff0031d8
CM
1183 return 0;
1184
1185 /* Ensure we do not step over superblock */
1186 if ((start_blk <= sbi->s_sb_block) &&
e5d39597 1187 (start_blk + count - 1 >= sbi->s_sb_block))
ff0031d8
CM
1188 return 0;
1189
ff0031d8
CM
1190 return 1;
1191}
1192
1da177e4 1193/*
a686cd89
MB
1194 * ext2_new_blocks() -- core block(s) allocation function
1195 * @inode: file inode
1196 * @goal: given target block(filesystem wide)
1197 * @count: target number of blocks to allocate
1198 * @errp: error code
1199 *
1200 * ext2_new_blocks uses a goal block to assist allocation. If the goal is
1da177e4
LT
1201 * free, or there is a free block within 32 blocks of the goal, that block
1202 * is allocated. Otherwise a forward search is made for a free block; within
1203 * each block group the search first looks for an entire free byte in the block
1204 * bitmap, and then for any free bit if that fails.
1205 * This function also updates quota and i_blocks field.
1206 */
a686cd89
MB
1207ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1208 unsigned long *count, int *errp)
1da177e4
LT
1209{
1210 struct buffer_head *bitmap_bh = NULL;
a686cd89
MB
1211 struct buffer_head *gdp_bh;
1212 int group_no;
1213 int goal_group;
1214 ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1215 ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1216 ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
1217 int bgi; /* blockgroup iteration index */
1218 int performed_allocation = 0;
1219 ext2_grpblk_t free_blocks; /* number of free blocks in a group */
1220 struct super_block *sb;
1221 struct ext2_group_desc *gdp;
1222 struct ext2_super_block *es;
1223 struct ext2_sb_info *sbi;
1224 struct ext2_reserve_window_node *my_rsv = NULL;
1225 struct ext2_block_alloc_info *block_i;
1226 unsigned short windowsz = 0;
1227 unsigned long ngroups;
1228 unsigned long num = *count;
5dd4056d 1229 int ret;
a686cd89
MB
1230
1231 *errp = -ENOSPC;
1232 sb = inode->i_sb;
1da177e4 1233
a686cd89
MB
1234 /*
1235 * Check quota for allocation of this block.
1236 */
5dd4056d
CH
1237 ret = dquot_alloc_block(inode, num);
1238 if (ret) {
1239 *errp = ret;
a686cd89
MB
1240 return 0;
1241 }
1da177e4 1242
a686cd89
MB
1243 sbi = EXT2_SB(sb);
1244 es = EXT2_SB(sb)->s_es;
1245 ext2_debug("goal=%lu.\n", goal);
1246 /*
1247 * Allocate a block from reservation only when
1248 * filesystem is mounted with reservation(default,-o reservation), and
1249 * it's a regular file, and
1250 * the desired window size is greater than 0 (One could use ioctl
1251 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1252 * reservation on that particular file)
1253 */
1254 block_i = EXT2_I(inode)->i_block_alloc_info;
1255 if (block_i) {
1256 windowsz = block_i->rsv_window_node.rsv_goal_size;
1257 if (windowsz > 0)
1258 my_rsv = &block_i->rsv_window_node;
1da177e4
LT
1259 }
1260
a686cd89
MB
1261 if (!ext2_has_free_blocks(sbi)) {
1262 *errp = -ENOSPC;
1263 goto out;
1264 }
1da177e4 1265
a686cd89
MB
1266 /*
1267 * First, test whether the goal block is free.
1268 */
1da177e4
LT
1269 if (goal < le32_to_cpu(es->s_first_data_block) ||
1270 goal >= le32_to_cpu(es->s_blocks_count))
1271 goal = le32_to_cpu(es->s_first_data_block);
a686cd89
MB
1272 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1273 EXT2_BLOCKS_PER_GROUP(sb);
1274 goal_group = group_no;
1275retry_alloc:
1276 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1277 if (!gdp)
1da177e4 1278 goto io_error;
1da177e4 1279
a686cd89
MB
1280 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1281 /*
1282 * if there is not enough free blocks to make a new resevation
1283 * turn off reservation for this allocation
1284 */
1285 if (my_rsv && (free_blocks < windowsz)
d707d31c 1286 && (free_blocks > 0)
a686cd89
MB
1287 && (rsv_is_empty(&my_rsv->rsv_window)))
1288 my_rsv = NULL;
1289
1290 if (free_blocks > 0) {
1291 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1292 EXT2_BLOCKS_PER_GROUP(sb));
ba1af2e4
CX
1293 /*
1294 * In case we retry allocation (due to fs reservation not
1295 * working out or fs corruption), the bitmap_bh is non-null
1296 * pointer and we have to release it before calling
1297 * read_block_bitmap().
1298 */
1299 brelse(bitmap_bh);
1da177e4
LT
1300 bitmap_bh = read_block_bitmap(sb, group_no);
1301 if (!bitmap_bh)
1302 goto io_error;
a686cd89
MB
1303 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1304 bitmap_bh, grp_target_blk,
1305 my_rsv, &num);
1306 if (grp_alloc_blk >= 0)
1307 goto allocated;
1da177e4
LT
1308 }
1309
a686cd89
MB
1310 ngroups = EXT2_SB(sb)->s_groups_count;
1311 smp_rmb();
1da177e4
LT
1312
1313 /*
144704e5
AM
1314 * Now search the rest of the groups. We assume that
1315 * group_no and gdp correctly point to the last group visited.
1da177e4 1316 */
a686cd89 1317 for (bgi = 0; bgi < ngroups; bgi++) {
1da177e4 1318 group_no++;
a686cd89 1319 if (group_no >= ngroups)
1da177e4 1320 group_no = 0;
a686cd89
MB
1321 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1322 if (!gdp)
1da177e4 1323 goto io_error;
1da177e4 1324
a686cd89 1325 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
46891532
JK
1326 /*
1327 * skip this group (and avoid loading bitmap) if there
1328 * are no free blocks
1329 */
1330 if (!free_blocks)
1331 continue;
1da177e4 1332 /*
a686cd89
MB
1333 * skip this group if the number of
1334 * free blocks is less than half of the reservation
1335 * window size.
1da177e4 1336 */
d707d31c 1337 if (my_rsv && (free_blocks <= (windowsz/2)))
a686cd89
MB
1338 continue;
1339
1340 brelse(bitmap_bh);
1341 bitmap_bh = read_block_bitmap(sb, group_no);
1342 if (!bitmap_bh)
1da177e4 1343 goto io_error;
1da177e4 1344 /*
a686cd89 1345 * try to allocate block(s) from this group, without a goal(-1).
1da177e4 1346 */
a686cd89
MB
1347 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1348 bitmap_bh, -1, my_rsv, &num);
1349 if (grp_alloc_blk >= 0)
1350 goto allocated;
1351 }
1352 /*
25985edc 1353 * We may end up a bogus earlier ENOSPC error due to
a686cd89 1354 * filesystem is "full" of reservations, but
25985edc 1355 * there maybe indeed free blocks available on disk
a686cd89
MB
1356 * In this case, we just forget about the reservations
1357 * just do block allocation as without reservations.
1358 */
1359 if (my_rsv) {
1360 my_rsv = NULL;
1361 windowsz = 0;
1362 group_no = goal_group;
1363 goto retry_alloc;
1da177e4 1364 }
a686cd89
MB
1365 /* No space left on the device */
1366 *errp = -ENOSPC;
1367 goto out;
1368
1369allocated:
1da177e4 1370
1da177e4 1371 ext2_debug("using block group %d(%d)\n",
a686cd89 1372 group_no, gdp->bg_free_blocks_count);
1da177e4 1373
a686cd89 1374 ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1da177e4 1375
a686cd89
MB
1376 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1377 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1378 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1379 EXT2_SB(sb)->s_itb_per_group) ||
1380 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
7f0adaec 1381 EXT2_SB(sb)->s_itb_per_group)) {
a686cd89 1382 ext2_error(sb, "ext2_new_blocks",
1da177e4 1383 "Allocating block in system zone - "
a686cd89
MB
1384 "blocks from "E2FSBLK", length %lu",
1385 ret_block, num);
8b915825
AK
1386 /*
1387 * ext2_try_to_allocate marked the blocks we allocated as in
1388 * use. So we may want to selectively mark some of the blocks
1389 * as free
1390 */
158be76c 1391 num = *count;
8b915825 1392 goto retry_alloc;
7f0adaec 1393 }
a686cd89
MB
1394
1395 performed_allocation = 1;
1da177e4 1396
a686cd89
MB
1397 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1398 ext2_error(sb, "ext2_new_blocks",
1399 "block("E2FSBLK") >= blocks count(%d) - "
1da177e4
LT
1400 "block_group = %d, es == %p ", ret_block,
1401 le32_to_cpu(es->s_blocks_count), group_no, es);
a686cd89 1402 goto out;
1da177e4 1403 }
1da177e4 1404
a686cd89
MB
1405 group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1406 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1da177e4
LT
1407
1408 mark_buffer_dirty(bitmap_bh);
1751e8a6 1409 if (sb->s_flags & SB_SYNCHRONOUS)
1da177e4
LT
1410 sync_dirty_buffer(bitmap_bh);
1411
a686cd89
MB
1412 *errp = 0;
1413 brelse(bitmap_bh);
8e3dffc6
WS
1414 if (num < *count) {
1415 dquot_free_block_nodirty(inode, *count-num);
1416 mark_inode_dirty(inode);
1417 *count = num;
1418 }
a686cd89 1419 return ret_block;
1da177e4 1420
a686cd89
MB
1421io_error:
1422 *errp = -EIO;
1da177e4 1423out:
a686cd89
MB
1424 /*
1425 * Undo the block allocation
1426 */
3889717d
AV
1427 if (!performed_allocation) {
1428 dquot_free_block_nodirty(inode, *count);
1429 mark_inode_dirty(inode);
1430 }
1da177e4 1431 brelse(bitmap_bh);
a686cd89
MB
1432 return 0;
1433}
1da177e4 1434
a686cd89
MB
1435ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1436{
1437 unsigned long count = 1;
1438
1439 return ext2_new_blocks(inode, goal, &count, errp);
1da177e4
LT
1440}
1441
21730eed
VH
1442#ifdef EXT2FS_DEBUG
1443
ecd0afa3 1444unsigned long ext2_count_free(struct buffer_head *map, unsigned int numchars)
21730eed 1445{
ecd0afa3 1446 return numchars * BITS_PER_BYTE - memweight(map->b_data, numchars);
21730eed
VH
1447}
1448
1449#endif /* EXT2FS_DEBUG */
1450
1da177e4
LT
1451unsigned long ext2_count_free_blocks (struct super_block * sb)
1452{
1453 struct ext2_group_desc * desc;
1454 unsigned long desc_count = 0;
1455 int i;
1456#ifdef EXT2FS_DEBUG
1457 unsigned long bitmap_count, x;
1458 struct ext2_super_block *es;
1459
1da177e4
LT
1460 es = EXT2_SB(sb)->s_es;
1461 desc_count = 0;
1462 bitmap_count = 0;
1463 desc = NULL;
1464 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1465 struct buffer_head *bitmap_bh;
1466 desc = ext2_get_group_desc (sb, i, NULL);
1467 if (!desc)
1468 continue;
1469 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1470 bitmap_bh = read_block_bitmap(sb, i);
1471 if (!bitmap_bh)
1472 continue;
1473
1474 x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1475 printk ("group %d: stored = %d, counted = %lu\n",
1476 i, le16_to_cpu(desc->bg_free_blocks_count), x);
1477 bitmap_count += x;
1478 brelse(bitmap_bh);
1479 }
1480 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1481 (long)le32_to_cpu(es->s_free_blocks_count),
1482 desc_count, bitmap_count);
1da177e4
LT
1483 return bitmap_count;
1484#else
1485 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1486 desc = ext2_get_group_desc (sb, i, NULL);
1487 if (!desc)
1488 continue;
1489 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1490 }
1491 return desc_count;
1492#endif
1493}
1494
1da177e4
LT
1495static inline int test_root(int a, int b)
1496{
1497 int num = b;
1498
1499 while (a > num)
1500 num *= b;
1501 return num == a;
1502}
1503
1504static int ext2_group_sparse(int group)
1505{
1506 if (group <= 1)
1507 return 1;
1508 return (test_root(group, 3) || test_root(group, 5) ||
1509 test_root(group, 7));
1510}
1511
1512/**
1513 * ext2_bg_has_super - number of blocks used by the superblock in group
1514 * @sb: superblock for filesystem
1515 * @group: group number to check
1516 *
1517 * Return the number of blocks used by the superblock (primary or backup)
1518 * in this group. Currently this will be only 0 or 1.
1519 */
1520int ext2_bg_has_super(struct super_block *sb, int group)
1521{
1522 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1523 !ext2_group_sparse(group))
1524 return 0;
1525 return 1;
1526}
1527
1528/**
1529 * ext2_bg_num_gdb - number of blocks used by the group table in group
1530 * @sb: superblock for filesystem
1531 * @group: group number to check
1532 *
1533 * Return the number of blocks used by the group descriptor table
1534 * (primary or backup) in this group. In the future there may be a
1535 * different number of descriptor blocks in each group.
1536 */
1537unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1538{
859cb936 1539 return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
1da177e4
LT
1540}
1541