Line data Source code
1 : /*
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 :
14 : #include "ext2.h"
15 : #include <linux/quotaops.h>
16 : #include <linux/sched.h>
17 : #include <linux/buffer_head.h>
18 : #include <linux/capability.h>
19 :
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
32 : * when a file system is mounted (see ext2_fill_super).
33 : */
34 :
35 :
36 : #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
37 :
38 : struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
39 : unsigned int block_group,
40 : struct buffer_head ** bh)
41 10394 : {
42 10394 : unsigned long group_desc;
43 10394 : unsigned long offset;
44 10394 : struct ext2_group_desc * desc;
45 41576 : struct ext2_sb_info *sbi = EXT2_SB(sb);
46 10394 :
47 31182 : if (block_group >= sbi->s_groups_count) {
48 12858 : 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 10394 : return NULL;
54 : }
55 :
56 31182 : group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
57 31182 : offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
58 20788 : if (!sbi->s_group_desc[group_desc]) {
59 12858 : 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 10394 : return NULL;
64 : }
65 :
66 20788 : desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
67 20788 : if (bh)
68 10394 : *bh = sbi->s_group_desc[group_desc];
69 10394 : return desc + offset;
70 : }
71 :
72 : static int ext2_valid_block_bitmap(struct super_block *sb,
73 : struct ext2_group_desc *desc,
74 : unsigned int block_group,
75 502 : struct buffer_head *bh)
76 502 : {
77 502 : ext2_grpblk_t offset;
78 502 : ext2_grpblk_t next_zero_bit;
79 502 : ext2_fsblk_t bitmap_blk;
80 502 : ext2_fsblk_t group_first_block;
81 502 :
82 1506 : group_first_block = ext2_group_first_block_no(sb, block_group);
83 502 :
84 : /* check whether block bitmap block number is set */
85 1004 : bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
86 502 : offset = bitmap_blk - group_first_block;
87 2510 : if (!ext2_test_bit(offset, bh->b_data))
88 : /* bad block bitmap */
89 502 : goto err_out;
90 :
91 : /* check whether the inode bitmap block number is set */
92 1004 : bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
93 502 : offset = bitmap_blk - group_first_block;
94 2510 : if (!ext2_test_bit(offset, bh->b_data))
95 : /* bad block bitmap */
96 502 : goto err_out;
97 :
98 : /* check whether the inode table block number is set */
99 1004 : bitmap_blk = le32_to_cpu(desc->bg_inode_table);
100 502 : offset = bitmap_blk - group_first_block;
101 2510 : next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
102 : offset + EXT2_SB(sb)->s_itb_per_group,
103 : offset);
104 2008 : if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
105 : /* good bitmap for inode tables */
106 502 : return 1;
107 :
108 : err_out:
109 4518 : ext2_error(sb, __func__,
110 502 : "Invalid block bitmap - "
111 : "block_group = %d, block = %lu",
112 : block_group, bitmap_blk);
113 502 : return 0;
114 : }
115 :
116 : /*
117 : * Read the bitmap for a given block_group,and validate the
118 : * bits for block/inode/inode tables are set in the bitmaps
119 : *
120 : * Return buffer_head on success or NULL in case of failure.
121 : */
122 : static struct buffer_head *
123 : read_block_bitmap(struct super_block *sb, unsigned int block_group)
124 : {
125 502 : struct ext2_group_desc * desc;
126 1004 : struct buffer_head * bh = NULL;
127 502 : ext2_fsblk_t bitmap_blk;
128 502 :
129 2510 : desc = ext2_get_group_desc(sb, block_group, NULL);
130 1506 : if (!desc)
131 1004 : return NULL;
132 1004 : bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
133 1004 : bh = sb_getblk(sb, bitmap_blk);
134 2008 : if (unlikely(!bh)) {
135 1506 : ext2_error(sb, __func__,
136 : "Cannot read block bitmap - "
137 : "block_group = %d, block_bitmap = %u",
138 : block_group, le32_to_cpu(desc->bg_block_bitmap));
139 502 : return NULL;
140 : }
141 2510 : if (likely(bh_uptodate_or_lock(bh)))
142 502 : return bh;
143 :
144 1506 : if (bh_submit_read(bh) < 0) {
145 1004 : brelse(bh);
146 1506 : ext2_error(sb, __func__,
147 : "Cannot read block bitmap - "
148 : "block_group = %d, block_bitmap = %u",
149 : block_group, le32_to_cpu(desc->bg_block_bitmap));
150 502 : return NULL;
151 : }
152 :
153 1506 : ext2_valid_block_bitmap(sb, desc, block_group, bh);
154 : /*
155 : * file system mounted not to panic on error, continue with corrupt
156 : * bitmap
157 : */
158 502 : return bh;
159 : }
160 :
161 : static void release_blocks(struct super_block *sb, int count)
162 : {
163 906 : if (count) {
164 1208 : struct ext2_sb_info *sbi = EXT2_SB(sb);
165 :
166 604 : percpu_counter_add(&sbi->s_freeblocks_counter, count);
167 302 : sb->s_dirt = 1;
168 : }
169 604 : }
170 :
171 : static void group_adjust_blocks(struct super_block *sb, int group_no,
172 : struct ext2_group_desc *desc, struct buffer_head *bh, int count)
173 : {
174 1806 : if (count) {
175 2408 : struct ext2_sb_info *sbi = EXT2_SB(sb);
176 602 : unsigned free_blocks;
177 602 :
178 3010 : spin_lock(sb_bgl_lock(sbi, group_no));
179 1204 : free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
180 602 : desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
181 2408 : spin_unlock(sb_bgl_lock(sbi, group_no));
182 602 : sb->s_dirt = 1;
183 602 : mark_buffer_dirty(bh);
184 : }
185 1204 : }
186 :
187 : /*
188 : * The reservation window structure operations
189 : * --------------------------------------------
190 : * Operations include:
191 : * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
192 : *
193 : * We use a red-black tree to represent per-filesystem reservation
194 : * windows.
195 : *
196 : */
197 :
198 : /**
199 : * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
200 : * @rb_root: root of per-filesystem reservation rb tree
201 : * @verbose: verbose mode
202 : * @fn: function which wishes to dump the reservation map
203 : *
204 : * If verbose is turned on, it will print the whole block reservation
205 : * windows(start, end). Otherwise, it will only print out the "bad" windows,
206 : * those windows that overlap with their immediate neighbors.
207 : */
208 : #if 1
209 : static void __rsv_window_dump(struct rb_root *root, int verbose,
210 : const char *fn)
211 1400 : {
212 1400 : struct rb_node *n;
213 1400 : struct ext2_reserve_window_node *rsv, *prev;
214 1400 : int bad;
215 1400 :
216 1400 : restart:
217 1400 : n = rb_first(root);
218 1400 : bad = 0;
219 2800 : prev = NULL;
220 :
221 1400 : printk("Block Allocation Reservation Windows Map (%s):\n", fn);
222 4200 : while (n) {
223 4200 : rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
224 4200 : if (verbose)
225 1400 : printk("reservation window 0x%p "
226 : "start: %lu, end: %lu\n",
227 : rsv, rsv->rsv_start, rsv->rsv_end);
228 5600 : if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
229 1400 : printk("Bad reservation %p (start >= end)\n",
230 : rsv);
231 1400 : bad = 1;
232 : }
233 5600 : if (prev && prev->rsv_end >= rsv->rsv_start) {
234 1400 : printk("Bad reservation %p (prev->end >= start)\n",
235 : rsv);
236 1400 : bad = 1;
237 : }
238 2800 : if (bad) {
239 2800 : if (!verbose) {
240 1400 : printk("Restarting reservation walk in verbose mode\n");
241 1400 : verbose = 1;
242 1400 : goto restart;
243 : }
244 : }
245 1400 : n = rb_next(n);
246 1400 : prev = rsv;
247 : }
248 2800 : printk("Window map complete.\n");
249 8400 : BUG_ON(bad);
250 : }
251 1400 : #define rsv_window_dump(root, verbose) \
252 : __rsv_window_dump((root), (verbose), __func__)
253 : #else
254 : #define rsv_window_dump(root, verbose) do {} while (0)
255 : #endif
256 :
257 : /**
258 : * goal_in_my_reservation()
259 : * @rsv: inode's reservation window
260 : * @grp_goal: given goal block relative to the allocation block group
261 : * @group: the current allocation block group
262 : * @sb: filesystem super block
263 : *
264 : * Test if the given goal block (group relative) is within the file's
265 : * own block reservation window range.
266 : *
267 : * If the reservation window is outside the goal allocation group, return 0;
268 : * grp_goal (given goal block) could be -1, which means no specific
269 : * goal block. In this case, always return 1.
270 : * If the goal block is within the reservation window, return 1;
271 : * otherwise, return 0;
272 : */
273 : static int
274 : goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
275 : unsigned int group, struct super_block * sb)
276 : {
277 400 : ext2_fsblk_t group_first_block, group_last_block;
278 400 :
279 1200 : group_first_block = ext2_group_first_block_no(sb, group);
280 1200 : group_last_block = group_first_block + EXT2_BLOCKS_PER_GROUP(sb) - 1;
281 :
282 1600 : if ((rsv->_rsv_start > group_last_block) ||
283 : (rsv->_rsv_end < group_first_block))
284 400 : return 0;
285 2400 : if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
286 : || (grp_goal + group_first_block > rsv->_rsv_end)))
287 400 : return 0;
288 400 : return 1;
289 : }
290 :
291 : /**
292 : * search_reserve_window()
293 : * @rb_root: root of reservation tree
294 : * @goal: target allocation block
295 : *
296 : * Find the reserved window which includes the goal, or the previous one
297 : * if the goal is not in any window.
298 : * Returns NULL if there are no windows or if all windows start after the goal.
299 : */
300 : static struct ext2_reserve_window_node *
301 : search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
302 : {
303 400 : struct rb_node *n = root->rb_node;
304 400 : struct ext2_reserve_window_node *rsv;
305 400 :
306 1200 : if (!n)
307 800 : return NULL;
308 :
309 : do {
310 800 : rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
311 400 :
312 800 : if (goal < rsv->rsv_start)
313 400 : n = n->rb_left;
314 800 : else if (goal > rsv->rsv_end)
315 400 : n = n->rb_right;
316 : else
317 400 : return rsv;
318 800 : } while (n);
319 : /*
320 400 : * We've fallen off the end of the tree: the goal wasn't inside
321 : * any particular node. OK, the previous node must be to one
322 : * side of the interval containing the goal. If it's the RHS,
323 : * we need to back up one.
324 : */
325 800 : if (rsv->rsv_start > goal) {
326 400 : n = rb_prev(&rsv->rsv_node);
327 800 : rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
328 : }
329 400 : return rsv;
330 : }
331 :
332 : /*
333 : * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
334 : * @sb: super block
335 : * @rsv: reservation window to add
336 : *
337 : * Must be called with rsv_lock held.
338 : */
339 : void ext2_rsv_window_add(struct super_block *sb,
340 : struct ext2_reserve_window_node *rsv)
341 1200 : {
342 4800 : struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
343 2400 : struct rb_node *node = &rsv->rsv_node;
344 2400 : ext2_fsblk_t start = rsv->rsv_start;
345 1200 :
346 2400 : struct rb_node ** p = &root->rb_node;
347 2400 : struct rb_node * parent = NULL;
348 1200 : struct ext2_reserve_window_node *this;
349 :
350 3600 : while (*p)
351 1200 : {
352 2400 : parent = *p;
353 2400 : this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
354 :
355 2400 : if (start < this->rsv_start)
356 1200 : p = &(*p)->rb_left;
357 2400 : else if (start > this->rsv_end)
358 1200 : p = &(*p)->rb_right;
359 : else {
360 2400 : rsv_window_dump(root, 1);
361 2400 : BUG();
362 : }
363 : }
364 2400 :
365 2400 : rb_link_node(node, parent, p);
366 1200 : rb_insert_color(node, root);
367 1200 : }
368 :
369 : /**
370 : * rsv_window_remove() -- unlink a window from the reservation rb tree
371 : * @sb: super block
372 : * @rsv: reservation window to remove
373 : *
374 : * Mark the block reservation window as not allocated, and unlink it
375 : * from the filesystem reservation window rb tree. Must be called with
376 : * rsv_lock held.
377 : */
378 : static void rsv_window_remove(struct super_block *sb,
379 : struct ext2_reserve_window_node *rsv)
380 1227 : {
381 1227 : rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
382 1227 : rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
383 1227 : rsv->rsv_alloc_hit = 0;
384 3681 : rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
385 1227 : }
386 :
387 : /*
388 : * rsv_is_empty() -- Check if the reservation window is allocated.
389 : * @rsv: given reservation window to check
390 : *
391 : * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
392 : */
393 : static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
394 : {
395 : /* a valid reservation end block could not be 0 */
396 1954 : return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
397 : }
398 :
399 : /**
400 : * ext2_init_block_alloc_info()
401 : * @inode: file inode structure
402 : *
403 : * Allocate and initialize the reservation window structure, and
404 : * link the window to the ext2 inode structure at last
405 : *
406 : * The reservation window structure is only dynamically allocated
407 : * and linked to ext2 inode the first time the open file
408 : * needs a new block. So, before every ext2_new_block(s) call, for
409 : * regular files, we should check whether the reservation window
410 : * structure exists or not. In the latter case, this function is called.
411 : * Fail to do so will result in block reservation being turned off for that
412 : * open file.
413 : *
414 : * This function is called from ext2_get_blocks_handle(), also called
415 : * when setting the reservation window size through ioctl before the file
416 : * is open for write (needs block allocation).
417 : *
418 : * Needs truncate_mutex protection prior to calling this function.
419 : */
420 : void ext2_init_block_alloc_info(struct inode *inode)
421 : {
422 16 : struct ext2_inode_info *ei = EXT2_I(inode);
423 8 : struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
424 8 : struct super_block *sb = inode->i_sb;
425 4 :
426 16 : block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
427 12 : if (block_i) {
428 8 : struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
429 :
430 4 : rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
431 4 : rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
432 :
433 : /*
434 : * if filesystem is mounted with NORESERVATION, the goal
435 : * reservation window size is set to zero to indicate
436 : * block reservation is off
437 : */
438 16 : if (!test_opt(sb, RESERVATION))
439 4 : rsv->rsv_goal_size = 0;
440 : else
441 4 : rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
442 4 : rsv->rsv_alloc_hit = 0;
443 4 : block_i->last_alloc_logical_block = 0;
444 4 : block_i->last_alloc_physical_block = 0;
445 : }
446 8 : ei->i_block_alloc_info = block_i;
447 8 : }
448 :
449 : /**
450 : * ext2_discard_reservation()
451 : * @inode: inode
452 : *
453 : * Discard(free) block reservation window on last file close, or truncate
454 : * or at last iput().
455 : *
456 : * It is being called in three cases:
457 : * ext2_release_file(): last writer closes the file
458 : * ext2_clear_inode(): last iput(), when nobody links to this file.
459 : * ext2_truncate(): when the block indirect map is about to change.
460 : */
461 : void ext2_discard_reservation(struct inode *inode)
462 : {
463 108 : struct ext2_inode_info *ei = EXT2_I(inode);
464 54 : struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
465 27 : struct ext2_reserve_window_node *rsv;
466 108 : spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
467 27 :
468 81 : if (!block_i)
469 54 : return;
470 27 :
471 27 : rsv = &block_i->rsv_window_node;
472 108 : if (!rsv_is_empty(&rsv->rsv_window)) {
473 54 : spin_lock(rsv_lock);
474 108 : if (!rsv_is_empty(&rsv->rsv_window))
475 54 : rsv_window_remove(inode->i_sb, rsv);
476 108 : spin_unlock(rsv_lock);
477 : }
478 54 : }
479 :
480 : /**
481 : * ext2_free_blocks_sb() -- Free given blocks and update quota and i_blocks
482 : * @inode: inode
483 : * @block: start physcial block to free
484 : * @count: number of blocks to free
485 : */
486 : void ext2_free_blocks (struct inode * inode, unsigned long block,
487 : unsigned long count)
488 302 : {
489 604 : struct buffer_head *bitmap_bh = NULL;
490 302 : struct buffer_head * bh2;
491 302 : unsigned long block_group;
492 302 : unsigned long bit;
493 302 : unsigned long i;
494 302 : unsigned long overflow;
495 604 : struct super_block * sb = inode->i_sb;
496 1208 : struct ext2_sb_info * sbi = EXT2_SB(sb);
497 302 : struct ext2_group_desc * desc;
498 604 : struct ext2_super_block * es = sbi->s_es;
499 604 : unsigned freed = 0, group_freed;
500 302 :
501 2718 : if (block < le32_to_cpu(es->s_first_data_block) ||
502 302 : block + count < block ||
503 302 : block + count > le32_to_cpu(es->s_blocks_count)) {
504 1208 : ext2_error (sb, "ext2_free_blocks",
505 302 : "Freeing blocks not in datazone - "
506 : "block = %lu, count = %lu", block, count);
507 302 : goto error_return;
508 : }
509 :
510 : ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
511 302 :
512 : do_more:
513 302 : overflow = 0;
514 1208 : block_group = (block - le32_to_cpu(es->s_first_data_block)) /
515 : EXT2_BLOCKS_PER_GROUP(sb);
516 1208 : bit = (block - le32_to_cpu(es->s_first_data_block)) %
517 : EXT2_BLOCKS_PER_GROUP(sb);
518 : /*
519 : * Check to see if we are freeing blocks across a group
520 : * boundary.
521 : */
522 1208 : if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
523 906 : overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
524 302 : count -= overflow;
525 : }
526 1208 : brelse(bitmap_bh);
527 1812 : bitmap_bh = read_block_bitmap(sb, block_group);
528 604 : if (!bitmap_bh)
529 302 : goto error_return;
530 :
531 1208 : desc = ext2_get_group_desc (sb, block_group, &bh2);
532 604 : if (!desc)
533 302 : goto error_return;
534 :
535 7248 : if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
536 : in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
537 : in_range (block, le32_to_cpu(desc->bg_inode_table),
538 : sbi->s_itb_per_group) ||
539 : in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
540 : sbi->s_itb_per_group)) {
541 906 : ext2_error (sb, "ext2_free_blocks",
542 : "Freeing blocks in system zones - "
543 : "Block = %lu, count = %lu",
544 : block, count);
545 302 : goto error_return;
546 : }
547 :
548 2114 : for (i = 0, group_freed = 0; i < count; i++) {
549 2416 : if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
550 302 : bit + i, bitmap_bh->b_data)) {
551 906 : ext2_error(sb, __func__,
552 : "bit already cleared for block %lu", block + i);
553 : } else {
554 302 : group_freed++;
555 : }
556 : }
557 :
558 302 : mark_buffer_dirty(bitmap_bh);
559 604 : if (sb->s_flags & MS_SYNCHRONOUS)
560 302 : sync_dirty_buffer(bitmap_bh);
561 :
562 906 : group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
563 302 : freed += group_freed;
564 :
565 604 : if (overflow) {
566 302 : block += count;
567 302 : count = overflow;
568 302 : goto do_more;
569 : }
570 : error_return:
571 3020 : brelse(bitmap_bh);
572 1208 : release_blocks(sb, freed);
573 604 : vfs_dq_free_block(inode, freed);
574 302 : }
575 :
576 : /**
577 : * bitmap_search_next_usable_block()
578 : * @start: the starting block (group relative) of the search
579 : * @bh: bufferhead contains the block group bitmap
580 : * @maxblocks: the ending block (group relative) of the reservation
581 : *
582 : * The bitmap search --- search forward through the actual bitmap on disk until
583 : * we find a bit free.
584 : */
585 : static ext2_grpblk_t
586 : bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
587 : ext2_grpblk_t maxblocks)
588 : {
589 1400 : ext2_grpblk_t next;
590 1400 :
591 4200 : next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
592 2800 : if (next >= maxblocks)
593 1400 : return -1;
594 1400 : return next;
595 : }
596 :
597 : /**
598 : * find_next_usable_block()
599 : * @start: the starting block (group relative) to find next
600 : * allocatable block in bitmap.
601 : * @bh: bufferhead contains the block group bitmap
602 : * @maxblocks: the ending block (group relative) for the search
603 : *
604 : * Find an allocatable block in a bitmap. We perform the "most
605 : * appropriate allocation" algorithm of looking for a free block near
606 : * the initial goal; then for a free byte somewhere in the bitmap;
607 : * then for any free bit in the bitmap.
608 : */
609 : static ext2_grpblk_t
610 : find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
611 : {
612 1000 : ext2_grpblk_t here, next;
613 1000 : char *p, *r;
614 1000 :
615 3000 : if (start > 0) {
616 1000 : /*
617 1000 : * The goal was occupied; search forward for a free
618 1000 : * block within the next XX blocks.
619 : *
620 : * end_goal is more or less random, but it has to be
621 : * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
622 : * next 64-bit boundary is simple..
623 : */
624 1000 : ext2_grpblk_t end_goal = (start + 63) & ~63;
625 2000 : if (end_goal > maxblocks)
626 1000 : end_goal = maxblocks;
627 3000 : here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
628 2000 : if (here < end_goal)
629 1000 : return here;
630 : ext2_debug("Bit not found near goal\n");
631 : }
632 :
633 1000 : here = start;
634 2000 : if (here < 0)
635 1000 : here = 0;
636 :
637 1000 : p = ((char *)bh->b_data) + (here >> 3);
638 2000 : r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
639 2000 : next = (r - ((char *)bh->b_data)) << 3;
640 :
641 4000 : if (next < maxblocks && next >= here)
642 1000 : return next;
643 :
644 2000 : here = bitmap_search_next_usable_block(here, bh, maxblocks);
645 1000 : return here;
646 : }
647 :
648 : /*
649 : * ext2_try_to_allocate()
650 : * @sb: superblock
651 : * @handle: handle to this transaction
652 : * @group: given allocation block group
653 : * @bitmap_bh: bufferhead holds the block bitmap
654 : * @grp_goal: given target block within the group
655 : * @count: target number of blocks to allocate
656 : * @my_rsv: reservation window
657 : *
658 : * Attempt to allocate blocks within a give range. Set the range of allocation
659 : * first, then find the first free bit(s) from the bitmap (within the range),
660 : * and at last, allocate the blocks by claiming the found free bit as allocated.
661 : *
662 : * To set the range of this allocation:
663 : * if there is a reservation window, only try to allocate block(s)
664 : * from the file's own reservation window;
665 : * Otherwise, the allocation range starts from the give goal block,
666 : * ends at the block group's last block.
667 : *
668 : * If we failed to allocate the desired block then we may end up crossing to a
669 : * new bitmap.
670 : */
671 : static int
672 : ext2_try_to_allocate(struct super_block *sb, int group,
673 : struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
674 : unsigned long *count,
675 1000 : struct ext2_reserve_window *my_rsv)
676 1000 : {
677 1000 : ext2_fsblk_t group_first_block;
678 1000 : ext2_grpblk_t start, end;
679 2000 : unsigned long num = 0;
680 1000 :
681 1000 : /* we do allocation within the reservation window if we have a window */
682 3000 : if (my_rsv) {
683 3000 : group_first_block = ext2_group_first_block_no(sb, group);
684 3000 : if (my_rsv->_rsv_start >= group_first_block)
685 3000 : start = my_rsv->_rsv_start - group_first_block;
686 1000 : else
687 1000 : /* reservation window cross group boundary */
688 1000 : start = 0;
689 2000 : end = my_rsv->_rsv_end - group_first_block + 1;
690 4000 : if (end > EXT2_BLOCKS_PER_GROUP(sb))
691 : /* reservation window crosses group boundary */
692 4000 : end = EXT2_BLOCKS_PER_GROUP(sb);
693 8000 : if ((start <= grp_goal) && (grp_goal < end))
694 2000 : start = grp_goal;
695 : else
696 2000 : grp_goal = -1;
697 : } else {
698 2000 : if (grp_goal > 0)
699 1000 : start = grp_goal;
700 : else
701 1000 : start = 0;
702 4000 : end = EXT2_BLOCKS_PER_GROUP(sb);
703 : }
704 :
705 12000 : BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
706 :
707 : repeat:
708 2000 : if (grp_goal < 0) {
709 3000 : grp_goal = find_next_usable_block(start, bitmap_bh, end);
710 3000 : if (grp_goal < 0)
711 1000 : goto fail_access;
712 2000 : if (!my_rsv) {
713 : int i;
714 :
715 11000 : for (i = 0; i < 7 && grp_goal > start &&
716 1000 : !ext2_test_bit(grp_goal - 1,
717 : bitmap_bh->b_data);
718 3000 : i++, grp_goal--)
719 1000 : ;
720 : }
721 1000 : }
722 4000 : start = grp_goal;
723 :
724 14000 : if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), grp_goal,
725 : bitmap_bh->b_data)) {
726 1000 : /*
727 : * The block was allocated by another thread, or it was
728 : * allocated and then freed by another thread
729 : */
730 1000 : start++;
731 1000 : grp_goal++;
732 2000 : if (start >= end)
733 1000 : goto fail_access;
734 1000 : goto repeat;
735 : }
736 1000 : num++;
737 1000 : grp_goal++;
738 10000 : while (num < *count && grp_goal < end
739 1000 : && !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
740 : grp_goal, bitmap_bh->b_data)) {
741 2000 : num++;
742 1000 : grp_goal++;
743 1000 : }
744 3000 : *count = num;
745 2000 : return grp_goal - num;
746 2000 : fail_access:
747 2000 : *count = num;
748 2000 : return -1;
749 1000 : }
750 :
751 : /**
752 : * find_next_reservable_window():
753 : * find a reservable space within the given range.
754 : * It does not allocate the reservation window for now:
755 : * alloc_new_reservation() will do the work later.
756 : *
757 : * @search_head: the head of the searching list;
758 : * This is not necessarily the list head of the whole filesystem
759 : *
760 : * We have both head and start_block to assist the search
761 : * for the reservable space. The list starts from head,
762 : * but we will shift to the place where start_block is,
763 : * then start from there, when looking for a reservable space.
764 : *
765 : * @size: the target new reservation window size
766 : *
767 : * @group_first_block: the first block we consider to start
768 : * the real search from
769 : *
770 : * @last_block:
771 : * the maximum block number that our goal reservable space
772 : * could start from. This is normally the last block in this
773 : * group. The search will end when we found the start of next
774 : * possible reservable space is out of this boundary.
775 : * This could handle the cross boundary reservation window
776 : * request.
777 : *
778 : * basically we search from the given range, rather than the whole
779 : * reservation double linked list, (start_block, last_block)
780 : * to find a free region that is of my size and has not
781 : * been reserved.
782 : *
783 : */
784 : static int find_next_reservable_window(
785 : struct ext2_reserve_window_node *search_head,
786 : struct ext2_reserve_window_node *my_rsv,
787 : struct super_block * sb,
788 400 : ext2_fsblk_t start_block,
789 400 : ext2_fsblk_t last_block)
790 400 : {
791 400 : struct rb_node *next;
792 400 : struct ext2_reserve_window_node *rsv, *prev;
793 400 : ext2_fsblk_t cur;
794 1200 : int size = my_rsv->rsv_goal_size;
795 :
796 : /* TODO: make the start of the reservation window byte-aligned */
797 : /* cur = *start_block & ~7;*/
798 400 : cur = start_block;
799 400 : rsv = search_head;
800 800 : if (!rsv)
801 400 : return -1;
802 :
803 : while (1) {
804 800 : if (cur <= rsv->rsv_end)
805 800 : cur = rsv->rsv_end + 1;
806 :
807 : /* TODO?
808 : * in the case we could not find a reservable space
809 : * that is what is expected, during the re-search, we could
810 : * remember what's the largest reservable space we could have
811 : * and return that one.
812 : *
813 : * For now it will fail if we could not find the reservable
814 : * space with expected-size (or more)...
815 : */
816 800 : if (cur > last_block)
817 400 : return -1; /* fail */
818 :
819 400 : prev = rsv;
820 400 : next = rb_next(&rsv->rsv_node);
821 800 : rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
822 :
823 : /*
824 : * Reached the last reservation, we can just append to the
825 : * previous one.
826 : */
827 800 : if (!next)
828 400 : break;
829 :
830 800 : if (cur + size <= rsv->rsv_start) {
831 : /*
832 : * Found a reserveable space big enough. We could
833 : * have a reservation across the group boundary here
834 : */
835 400 : break;
836 : }
837 400 : }
838 400 : /*
839 : * we come here either :
840 : * when we reach the end of the whole list,
841 : * and there is empty reservable space after last entry in the list.
842 : * append it to the end of the list.
843 : *
844 : * or we found one reservable space in the middle of the list,
845 : * return the reservation window that we could append to.
846 : * succeed.
847 : */
848 :
849 2400 : if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
850 800 : rsv_window_remove(sb, my_rsv);
851 :
852 : /*
853 : * Let's book the whole avaliable window for now. We will check the
854 : * disk bitmap later and then, if there are free blocks then we adjust
855 : * the window size if it's larger than requested.
856 : * Otherwise, we will remove this node from the tree next time
857 : * call find_next_reservable_window.
858 : */
859 1200 : my_rsv->rsv_start = cur;
860 1200 : my_rsv->rsv_end = cur + size - 1;
861 1200 : my_rsv->rsv_alloc_hit = 0;
862 :
863 2400 : if (prev != my_rsv)
864 2400 : ext2_rsv_window_add(sb, my_rsv);
865 :
866 1600 : return 0;
867 : }
868 :
869 : /**
870 : * alloc_new_reservation()--allocate a new reservation window
871 : *
872 : * To make a new reservation, we search part of the filesystem
873 : * reservation list (the list that inside the group). We try to
874 : * allocate a new reservation window near the allocation goal,
875 : * or the beginning of the group, if there is no goal.
876 : *
877 : * We first find a reservable space after the goal, then from
878 : * there, we check the bitmap for the first free block after
879 : * it. If there is no free block until the end of group, then the
880 : * whole group is full, we failed. Otherwise, check if the free
881 : * block is inside the expected reservable space, if so, we
882 : * succeed.
883 : * If the first free block is outside the reservable space, then
884 : * start from the first free block, we search for next available
885 : * space, and go on.
886 : *
887 : * on succeed, a new reservation will be found and inserted into the list
888 : * It contains at least one free block, and it does not overlap with other
889 : * reservation windows.
890 : *
891 : * failed: we failed to find a reservation window in this group
892 : *
893 : * @rsv: the reservation
894 : *
895 : * @grp_goal: The goal (group-relative). It is where the search for a
896 : * free reservable space should start from.
897 : * if we have a goal(goal >0 ), then start from there,
898 : * no goal(goal = -1), we start from the first block
899 : * of the group.
900 : *
901 : * @sb: the super block
902 : * @group: the group we are trying to allocate in
903 : * @bitmap_bh: the block group block bitmap
904 : *
905 : */
906 : static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
907 : ext2_grpblk_t grp_goal, struct super_block *sb,
908 : unsigned int group, struct buffer_head *bitmap_bh)
909 400 : {
910 400 : struct ext2_reserve_window_node *search_head;
911 400 : ext2_fsblk_t group_first_block, group_end_block, start_block;
912 400 : ext2_grpblk_t first_free_block;
913 1600 : struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
914 400 : unsigned long size;
915 400 : int ret;
916 1600 : spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
917 400 :
918 1200 : group_first_block = ext2_group_first_block_no(sb, group);
919 1600 : group_end_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
920 400 :
921 1200 : if (grp_goal < 0)
922 800 : start_block = group_first_block;
923 400 : else
924 400 : start_block = grp_goal + group_first_block;
925 :
926 800 : size = my_rsv->rsv_goal_size;
927 :
928 1600 : if (!rsv_is_empty(&my_rsv->rsv_window)) {
929 : /*
930 : * if the old reservation is cross group boundary
931 : * and if the goal is inside the old reservation window,
932 : * we will come here when we just failed to allocate from
933 : * the first part of the window. We still have another part
934 : * that belongs to the next group. In this case, there is no
935 : * point to discard our window and try to allocate a new one
936 : * in this group(which will fail). we should
937 : * keep the reservation window, just simply move on.
938 : *
939 : * Maybe we could shift the start block of the reservation
940 : * window to the first block of next group.
941 : */
942 :
943 2400 : if ((my_rsv->rsv_start <= group_end_block) &&
944 : (my_rsv->rsv_end > group_end_block) &&
945 : (start_block >= my_rsv->rsv_start))
946 400 : return -1;
947 :
948 1200 : if ((my_rsv->rsv_alloc_hit >
949 : (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
950 : /*
951 : * if the previously allocation hit ratio is
952 : * greater than 1/2, then we double the size of
953 : * the reservation window the next time,
954 : * otherwise we keep the same size window
955 : */
956 400 : size = size * 2;
957 800 : if (size > EXT2_MAX_RESERVE_BLOCKS)
958 400 : size = EXT2_MAX_RESERVE_BLOCKS;
959 400 : my_rsv->rsv_goal_size= size;
960 : }
961 : }
962 :
963 800 : spin_lock(rsv_lock);
964 : /*
965 : * shift the search start to the window near the goal block
966 : */
967 1200 : search_head = search_reserve_window(fs_rsv_root, start_block);
968 :
969 400 : /*
970 : * find_next_reservable_window() simply finds a reservable window
971 : * inside the given range(start_block, group_end_block).
972 : *
973 : * To make sure the reservation window has a free bit inside it, we
974 : * need to check the bitmap after we found a reservable window.
975 : */
976 : retry:
977 2400 : ret = find_next_reservable_window(search_head, my_rsv, sb,
978 : start_block, group_end_block);
979 :
980 800 : if (ret == -1) {
981 1600 : if (!rsv_is_empty(&my_rsv->rsv_window))
982 800 : rsv_window_remove(sb, my_rsv);
983 1600 : spin_unlock(rsv_lock);
984 400 : return -1;
985 : }
986 :
987 : /*
988 : * On success, find_next_reservable_window() returns the
989 : * reservation window where there is a reservable space after it.
990 : * Before we reserve this reservable space, we need
991 : * to make sure there is at least a free block inside this region.
992 : *
993 : * Search the first free bit on the block bitmap. Search starts from
994 : * the start block of the reservable space we just found.
995 : */
996 800 : spin_unlock(rsv_lock);
997 800 : first_free_block = bitmap_search_next_usable_block(
998 : my_rsv->rsv_start - group_first_block,
999 : bitmap_bh, group_end_block - group_first_block + 1);
1000 :
1001 800 : if (first_free_block < 0) {
1002 : /*
1003 : * no free block left on the bitmap, no point
1004 : * to reserve the space. return failed.
1005 : */
1006 800 : spin_lock(rsv_lock);
1007 1600 : if (!rsv_is_empty(&my_rsv->rsv_window))
1008 800 : rsv_window_remove(sb, my_rsv);
1009 1600 : spin_unlock(rsv_lock);
1010 400 : return -1; /* failed */
1011 : }
1012 :
1013 400 : start_block = first_free_block + group_first_block;
1014 : /*
1015 : * check if the first free block is within the
1016 : * free space we just reserved
1017 : */
1018 1600 : if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1019 400 : return 0; /* success */
1020 : /*
1021 : * if the first free bit we found is out of the reservable space
1022 : * continue search for next reservable space,
1023 : * start from where the free block is,
1024 : * we also shift the list head to where we stopped last time
1025 : */
1026 400 : search_head = my_rsv;
1027 800 : spin_lock(rsv_lock);
1028 400 : goto retry;
1029 : }
1030 :
1031 : /**
1032 : * try_to_extend_reservation()
1033 : * @my_rsv: given reservation window
1034 : * @sb: super block
1035 : * @size: the delta to extend
1036 : *
1037 : * Attempt to expand the reservation window large enough to have
1038 : * required number of free blocks
1039 : *
1040 : * Since ext2_try_to_allocate() will always allocate blocks within
1041 : * the reservation window range, if the window size is too small,
1042 : * multiple blocks allocation has to stop at the end of the reservation
1043 : * window. To make this more efficient, given the total number of
1044 : * blocks needed and the current size of the window, we try to
1045 : * expand the reservation window size if necessary on a best-effort
1046 : * basis before ext2_new_blocks() tries to allocate blocks.
1047 : */
1048 : static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1049 : struct super_block *sb, int size)
1050 : {
1051 200 : struct ext2_reserve_window_node *next_rsv;
1052 200 : struct rb_node *next;
1053 800 : spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1054 200 :
1055 1000 : if (!spin_trylock(rsv_lock))
1056 400 : return;
1057 :
1058 200 : next = rb_next(&my_rsv->rsv_node);
1059 :
1060 400 : if (!next)
1061 200 : my_rsv->rsv_end += size;
1062 : else {
1063 400 : next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1064 :
1065 400 : if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1066 200 : my_rsv->rsv_end += size;
1067 : else
1068 200 : my_rsv->rsv_end = next_rsv->rsv_start - 1;
1069 : }
1070 400 : spin_unlock(rsv_lock);
1071 200 : }
1072 :
1073 : /**
1074 : * ext2_try_to_allocate_with_rsv()
1075 : * @sb: superblock
1076 : * @group: given allocation block group
1077 : * @bitmap_bh: bufferhead holds the block bitmap
1078 : * @grp_goal: given target block within the group
1079 : * @count: target number of blocks to allocate
1080 : * @my_rsv: reservation window
1081 : *
1082 : * This is the main function used to allocate a new block and its reservation
1083 : * window.
1084 : *
1085 : * Each time when a new block allocation is need, first try to allocate from
1086 : * its own reservation. If it does not have a reservation window, instead of
1087 : * looking for a free bit on bitmap first, then look up the reservation list to
1088 : * see if it is inside somebody else's reservation window, we try to allocate a
1089 : * reservation window for it starting from the goal first. Then do the block
1090 : * allocation within the reservation window.
1091 : *
1092 : * This will avoid keeping on searching the reservation list again and
1093 : * again when somebody is looking for a free block (without
1094 : * reservation), and there are lots of free blocks, but they are all
1095 : * being reserved.
1096 : *
1097 : * We use a red-black tree for the per-filesystem reservation list.
1098 : */
1099 : static ext2_grpblk_t
1100 : ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1101 : struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1102 : struct ext2_reserve_window_node * my_rsv,
1103 : unsigned long *count)
1104 : {
1105 200 : ext2_fsblk_t group_first_block, group_last_block;
1106 400 : ext2_grpblk_t ret = 0;
1107 400 : unsigned long num = *count;
1108 200 :
1109 200 : /*
1110 200 : * we don't deal with reservation when
1111 200 : * filesystem is mounted without reservation
1112 200 : * or the file is not a regular file
1113 200 : * or last attempt to allocate a block with reservation turned on failed
1114 200 : */
1115 600 : if (my_rsv == NULL) {
1116 1200 : return ext2_try_to_allocate(sb, group, bitmap_bh,
1117 : grp_goal, count, NULL);
1118 : }
1119 : /*
1120 : * grp_goal is a group relative block number (if there is a goal)
1121 : * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1122 : * first block is a filesystem wide block number
1123 : * first block is the block number of the first block in this group
1124 : */
1125 400 : group_first_block = ext2_group_first_block_no(sb, group);
1126 600 : group_last_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
1127 :
1128 200 : /*
1129 : * Basically we will allocate a new block from inode's reservation
1130 : * window.
1131 : *
1132 : * We need to allocate a new reservation window, if:
1133 : * a) inode does not have a reservation window; or
1134 : * b) last attempt to allocate a block from existing reservation
1135 : * failed; or
1136 : * c) we come here with a goal and with a reservation window
1137 : *
1138 : * We do not need to allocate a new reservation window if we come here
1139 : * at the beginning with a goal and the goal is inside the window, or
1140 : * we don't have a goal but already have a reservation window.
1141 : * then we could go to allocate from the reservation window directly.
1142 : */
1143 : while (1) {
1144 2200 : if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1145 200 : !goal_in_my_reservation(&my_rsv->rsv_window,
1146 : grp_goal, group, sb)) {
1147 1200 : if (my_rsv->rsv_goal_size < *count)
1148 400 : my_rsv->rsv_goal_size = *count;
1149 2000 : ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1150 : group, bitmap_bh);
1151 400 : if (ret < 0)
1152 200 : break; /* failed */
1153 :
1154 800 : if (!goal_in_my_reservation(&my_rsv->rsv_window,
1155 : grp_goal, group, sb))
1156 200 : grp_goal = -1;
1157 400 : } else if (grp_goal >= 0) {
1158 200 : int curr = my_rsv->rsv_end -
1159 : (grp_goal + group_first_block) + 1;
1160 :
1161 400 : if (curr < *count)
1162 600 : try_to_extend_reservation(my_rsv, sb,
1163 : *count - curr);
1164 : }
1165 :
1166 2400 : if ((my_rsv->rsv_start > group_last_block) ||
1167 : (my_rsv->rsv_end < group_first_block)) {
1168 1600 : rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1169 400 : BUG();
1170 : }
1171 4000 : ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1172 : &num, &my_rsv->rsv_window);
1173 400 : if (ret >= 0) {
1174 200 : my_rsv->rsv_alloc_hit += num;
1175 200 : *count = num;
1176 200 : break; /* succeed */
1177 : }
1178 200 : num = *count;
1179 200 : }
1180 800 : return ret;
1181 : }
1182 :
1183 : /**
1184 : * ext2_has_free_blocks()
1185 : * @sbi: in-core super block structure.
1186 : *
1187 : * Check if filesystem has at least 1 free block available for allocation.
1188 : */
1189 : static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1190 : {
1191 100 : ext2_fsblk_t free_blocks, root_blocks;
1192 100 :
1193 400 : free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1194 300 : root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1195 1600 : if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1196 100 : sbi->s_resuid != current_fsuid() &&
1197 : (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1198 200 : return 0;
1199 : }
1200 200 : return 1;
1201 : }
1202 :
1203 : /*
1204 : * ext2_new_blocks() -- core block(s) allocation function
1205 : * @inode: file inode
1206 : * @goal: given target block(filesystem wide)
1207 : * @count: target number of blocks to allocate
1208 : * @errp: error code
1209 : *
1210 : * ext2_new_blocks uses a goal block to assist allocation. If the goal is
1211 : * free, or there is a free block within 32 blocks of the goal, that block
1212 : * is allocated. Otherwise a forward search is made for a free block; within
1213 : * each block group the search first looks for an entire free byte in the block
1214 : * bitmap, and then for any free bit if that fails.
1215 : * This function also updates quota and i_blocks field.
1216 : */
1217 : ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1218 : unsigned long *count, int *errp)
1219 : {
1220 200 : struct buffer_head *bitmap_bh = NULL;
1221 100 : struct buffer_head *gdp_bh;
1222 100 : int group_no;
1223 100 : int goal_group;
1224 100 : ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1225 100 : ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1226 100 : ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
1227 100 : int bgi; /* blockgroup iteration index */
1228 200 : int performed_allocation = 0;
1229 100 : ext2_grpblk_t free_blocks; /* number of free blocks in a group */
1230 100 : struct super_block *sb;
1231 100 : struct ext2_group_desc *gdp;
1232 100 : struct ext2_super_block *es;
1233 100 : struct ext2_sb_info *sbi;
1234 200 : struct ext2_reserve_window_node *my_rsv = NULL;
1235 100 : struct ext2_block_alloc_info *block_i;
1236 200 : unsigned short windowsz = 0;
1237 100 : unsigned long ngroups;
1238 200 : unsigned long num = *count;
1239 100 :
1240 200 : *errp = -ENOSPC;
1241 200 : sb = inode->i_sb;
1242 300 : if (!sb) {
1243 200 : printk("ext2_new_blocks: nonexistent device");
1244 200 : return 0;
1245 100 : }
1246 100 :
1247 100 : /*
1248 100 : * Check quota for allocation of this block.
1249 100 : */
1250 400 : if (vfs_dq_alloc_block(inode, num)) {
1251 100 : *errp = -EDQUOT;
1252 100 : return 0;
1253 : }
1254 :
1255 200 : sbi = EXT2_SB(sb);
1256 300 : es = EXT2_SB(sb)->s_es;
1257 : ext2_debug("goal=%lu.\n", goal);
1258 : /*
1259 : * Allocate a block from reservation only when
1260 : * filesystem is mounted with reservation(default,-o reservation), and
1261 : * it's a regular file, and
1262 : * the desired window size is greater than 0 (One could use ioctl
1263 : * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1264 : * reservation on that particular file)
1265 : */
1266 300 : block_i = EXT2_I(inode)->i_block_alloc_info;
1267 200 : if (block_i) {
1268 100 : windowsz = block_i->rsv_window_node.rsv_goal_size;
1269 200 : if (windowsz > 0)
1270 100 : my_rsv = &block_i->rsv_window_node;
1271 : }
1272 :
1273 500 : if (!ext2_has_free_blocks(sbi)) {
1274 100 : *errp = -ENOSPC;
1275 100 : goto out;
1276 : }
1277 :
1278 : /*
1279 : * First, test whether the goal block is free.
1280 : */
1281 600 : if (goal < le32_to_cpu(es->s_first_data_block) ||
1282 : goal >= le32_to_cpu(es->s_blocks_count))
1283 200 : goal = le32_to_cpu(es->s_first_data_block);
1284 400 : group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1285 : EXT2_BLOCKS_PER_GROUP(sb);
1286 100 : goal_group = group_no;
1287 : retry_alloc:
1288 500 : gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1289 200 : if (!gdp)
1290 100 : goto io_error;
1291 :
1292 200 : free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1293 : /*
1294 : * if there is not enough free blocks to make a new resevation
1295 : * turn off reservation for this allocation
1296 : */
1297 1000 : if (my_rsv && (free_blocks < windowsz)
1298 : && (free_blocks > 0)
1299 : && (rsv_is_empty(&my_rsv->rsv_window)))
1300 100 : my_rsv = NULL;
1301 :
1302 400 : if (free_blocks > 0) {
1303 600 : grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1304 : EXT2_BLOCKS_PER_GROUP(sb));
1305 600 : bitmap_bh = read_block_bitmap(sb, group_no);
1306 200 : if (!bitmap_bh)
1307 100 : goto io_error;
1308 400 : grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1309 : bitmap_bh, grp_target_blk,
1310 : my_rsv, &num);
1311 200 : if (grp_alloc_blk >= 0)
1312 100 : goto allocated;
1313 : }
1314 :
1315 700 : ngroups = EXT2_SB(sb)->s_groups_count;
1316 100 : smp_rmb();
1317 :
1318 : /*
1319 : * Now search the rest of the groups. We assume that
1320 : * group_no and gdp correctly point to the last group visited.
1321 : */
1322 600 : for (bgi = 0; bgi < ngroups; bgi++) {
1323 400 : group_no++;
1324 300 : if (group_no >= ngroups)
1325 100 : group_no = 0;
1326 400 : gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1327 200 : if (!gdp)
1328 100 : goto io_error;
1329 :
1330 200 : free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1331 : /*
1332 : * skip this group if the number of
1333 : * free blocks is less than half of the reservation
1334 : * window size.
1335 : */
1336 400 : if (my_rsv && (free_blocks <= (windowsz/2)))
1337 100 : continue;
1338 :
1339 200 : brelse(bitmap_bh);
1340 600 : bitmap_bh = read_block_bitmap(sb, group_no);
1341 200 : if (!bitmap_bh)
1342 100 : goto io_error;
1343 : /*
1344 : * try to allocate block(s) from this group, without a goal(-1).
1345 : */
1346 400 : grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1347 : bitmap_bh, -1, my_rsv, &num);
1348 200 : if (grp_alloc_blk >= 0)
1349 100 : goto allocated;
1350 : }
1351 : /*
1352 : * We may end up a bogus ealier ENOSPC error due to
1353 100 : * filesystem is "full" of reservations, but
1354 : * there maybe indeed free blocks avaliable on disk
1355 : * In this case, we just forget about the reservations
1356 : * just do block allocation as without reservations.
1357 : */
1358 200 : if (my_rsv) {
1359 100 : my_rsv = NULL;
1360 100 : windowsz = 0;
1361 100 : group_no = goal_group;
1362 100 : goto retry_alloc;
1363 : }
1364 : /* No space left on the device */
1365 100 : *errp = -ENOSPC;
1366 100 : goto out;
1367 200 :
1368 : allocated:
1369 :
1370 : ext2_debug("using block group %d(%d)\n",
1371 : group_no, gdp->bg_free_blocks_count);
1372 :
1373 500 : ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1374 :
1375 3400 : if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1376 : in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1377 : in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1378 100 : EXT2_SB(sb)->s_itb_per_group) ||
1379 : in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1380 : EXT2_SB(sb)->s_itb_per_group)) {
1381 900 : ext2_error(sb, "ext2_new_blocks",
1382 : "Allocating block in system zone - "
1383 : "blocks from "E2FSBLK", length %lu",
1384 : ret_block, num);
1385 : /*
1386 : * ext2_try_to_allocate marked the blocks we allocated as in
1387 : * use. So we may want to selectively mark some of the blocks
1388 : * as free
1389 : */
1390 300 : goto retry_alloc;
1391 : }
1392 :
1393 300 : performed_allocation = 1;
1394 :
1395 900 : if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1396 900 : ext2_error(sb, "ext2_new_blocks",
1397 : "block("E2FSBLK") >= blocks count(%d) - "
1398 : "block_group = %d, es == %p ", ret_block,
1399 : le32_to_cpu(es->s_blocks_count), group_no, es);
1400 100 : goto out;
1401 : }
1402 :
1403 900 : group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1404 200 : percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1405 :
1406 100 : mark_buffer_dirty(bitmap_bh);
1407 200 : if (sb->s_flags & MS_SYNCHRONOUS)
1408 100 : sync_dirty_buffer(bitmap_bh);
1409 :
1410 100 : *errp = 0;
1411 200 : brelse(bitmap_bh);
1412 200 : vfs_dq_free_block(inode, *count-num);
1413 100 : *count = num;
1414 100 : return ret_block;
1415 400 :
1416 : io_error:
1417 400 : *errp = -EIO;
1418 400 : out:
1419 : /*
1420 : * Undo the block allocation
1421 : */
1422 1400 : if (!performed_allocation)
1423 1400 : vfs_dq_free_block(inode, *count);
1424 1600 : brelse(bitmap_bh);
1425 100 : return 0;
1426 : }
1427 :
1428 : ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1429 : {
1430 200 : unsigned long count = 1;
1431 100 :
1432 600 : return ext2_new_blocks(inode, goal, &count, errp);
1433 : }
1434 :
1435 : #ifdef EXT2FS_DEBUG
1436 :
1437 : static const int nibblemap[] = {4, 3, 3, 2, 3, 2, 2, 1, 3, 2, 2, 1, 2, 1, 1, 0};
1438 :
1439 : unsigned long ext2_count_free (struct buffer_head * map, unsigned int numchars)
1440 : {
1441 : unsigned int i;
1442 : unsigned long sum = 0;
1443 :
1444 : if (!map)
1445 : return (0);
1446 : for (i = 0; i < numchars; i++)
1447 : sum += nibblemap[map->b_data[i] & 0xf] +
1448 : nibblemap[(map->b_data[i] >> 4) & 0xf];
1449 : return (sum);
1450 : }
1451 :
1452 : #endif /* EXT2FS_DEBUG */
1453 :
1454 : unsigned long ext2_count_free_blocks (struct super_block * sb)
1455 : {
1456 7001 : struct ext2_group_desc * desc;
1457 14002 : unsigned long desc_count = 0;
1458 7001 : int i;
1459 7001 : #ifdef EXT2FS_DEBUG
1460 : unsigned long bitmap_count, x;
1461 : struct ext2_super_block *es;
1462 :
1463 : es = EXT2_SB(sb)->s_es;
1464 : desc_count = 0;
1465 : bitmap_count = 0;
1466 : desc = NULL;
1467 : for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1468 : struct buffer_head *bitmap_bh;
1469 : desc = ext2_get_group_desc (sb, i, NULL);
1470 : if (!desc)
1471 : continue;
1472 : desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1473 : bitmap_bh = read_block_bitmap(sb, i);
1474 : if (!bitmap_bh)
1475 : continue;
1476 :
1477 : x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1478 : printk ("group %d: stored = %d, counted = %lu\n",
1479 : i, le16_to_cpu(desc->bg_free_blocks_count), x);
1480 : bitmap_count += x;
1481 : brelse(bitmap_bh);
1482 : }
1483 : printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1484 : (long)le32_to_cpu(es->s_free_blocks_count),
1485 : desc_count, bitmap_count);
1486 : return bitmap_count;
1487 : #else
1488 49007 : for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1489 34818 : desc = ext2_get_group_desc (sb, i, NULL);
1490 21003 : if (!desc)
1491 7001 : continue;
1492 14002 : desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1493 7001 : }
1494 7001 : return desc_count;
1495 : #endif
1496 : }
1497 :
1498 : static inline int test_root(int a, int b)
1499 : {
1500 12 : int num = b;
1501 :
1502 18 : while (a > num)
1503 12 : num *= b;
1504 18 : return num == a;
1505 : }
1506 :
1507 : static int ext2_group_sparse(int group)
1508 : {
1509 6 : if (group <= 1)
1510 4 : return 1;
1511 40 : return (test_root(group, 3) || test_root(group, 5) ||
1512 2 : test_root(group, 7));
1513 : }
1514 :
1515 : /**
1516 : * ext2_bg_has_super - number of blocks used by the superblock in group
1517 : * @sb: superblock for filesystem
1518 : * @group: group number to check
1519 : *
1520 : * Return the number of blocks used by the superblock (primary or backup)
1521 : * in this group. Currently this will be only 0 or 1.
1522 : */
1523 : int ext2_bg_has_super(struct super_block *sb, int group)
1524 : {
1525 26 : if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1526 2 : !ext2_group_sparse(group))
1527 2 : return 0;
1528 4 : return 1;
1529 : }
1530 :
1531 : /**
1532 : * ext2_bg_num_gdb - number of blocks used by the group table in group
1533 : * @sb: superblock for filesystem
1534 : * @group: group number to check
1535 : *
1536 : * Return the number of blocks used by the group descriptor table
1537 : * (primary or backup) in this group. In the future there may be a
1538 : * different number of descriptor blocks in each group.
1539 : */
1540 : unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1541 : {
1542 12 : return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
1543 1 : }
1544 1 :
|