Linux kernel & device driver programming

Cross-Referenced Linux and Device Driver Code

[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ]
Version: [ 2.6.11.8 ] [ 2.6.25 ] [ 2.6.25.8 ] [ 2.6.31.13 ] Architecture: [ i386 ]
  1 /*
  2  *  linux/fs/ext3/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 <linux/config.h>
 15 #include <linux/time.h>
 16 #include <linux/fs.h>
 17 #include <linux/jbd.h>
 18 #include <linux/ext3_fs.h>
 19 #include <linux/ext3_jbd.h>
 20 #include <linux/quotaops.h>
 21 #include <linux/buffer_head.h>
 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
 35  * when a file system is mounted (see ext3_read_super).
 36  */
 37 
 38 
 39 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
 40 
 41 struct ext3_group_desc * ext3_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 desc;
 47         struct ext3_group_desc * gdp;
 48 
 49         if (block_group >= EXT3_SB(sb)->s_groups_count) {
 50                 ext3_error (sb, "ext3_get_group_desc",
 51                             "block_group >= groups_count - "
 52                             "block_group = %d, groups_count = %lu",
 53                             block_group, EXT3_SB(sb)->s_groups_count);
 54 
 55                 return NULL;
 56         }
 57         smp_rmb();
 58 
 59         group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
 60         desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
 61         if (!EXT3_SB(sb)->s_group_desc[group_desc]) {
 62                 ext3_error (sb, "ext3_get_group_desc",
 63                             "Group descriptor not loaded - "
 64                             "block_group = %d, group_desc = %lu, desc = %lu",
 65                              block_group, group_desc, desc);
 66                 return NULL;
 67         }
 68 
 69         gdp = (struct ext3_group_desc *) 
 70               EXT3_SB(sb)->s_group_desc[group_desc]->b_data;
 71         if (bh)
 72                 *bh = EXT3_SB(sb)->s_group_desc[group_desc];
 73         return gdp + desc;
 74 }
 75 
 76 /*
 77  * Read the bitmap for a given block_group, reading into the specified 
 78  * slot in the superblock's bitmap cache.
 79  *
 80  * Return buffer_head on success or NULL in case of failure.
 81  */
 82 static struct buffer_head *
 83 read_block_bitmap(struct super_block *sb, unsigned int block_group)
 84 {
 85         struct ext3_group_desc * desc;
 86         struct buffer_head * bh = NULL;
 87 
 88         desc = ext3_get_group_desc (sb, block_group, NULL);
 89         if (!desc)
 90                 goto error_out;
 91         bh = sb_bread(sb, le32_to_cpu(desc->bg_block_bitmap));
 92         if (!bh)
 93                 ext3_error (sb, "read_block_bitmap",
 94                             "Cannot read block bitmap - "
 95                             "block_group = %d, block_bitmap = %u",
 96                             block_group, le32_to_cpu(desc->bg_block_bitmap));
 97 error_out:
 98         return bh;
 99 }
100 /*
101  * The reservation window structure operations
102  * --------------------------------------------
103  * Operations include:
104  * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
105  *
106  * We use sorted double linked list for the per-filesystem reservation
107  * window list. (like in vm_region).
108  *
109  * Initially, we keep those small operations in the abstract functions,
110  * so later if we need a better searching tree than double linked-list,
111  * we could easily switch to that without changing too much
112  * code.
113  */
114 #if 0
115 static void __rsv_window_dump(struct rb_root *root, int verbose,
116                               const char *fn)
117 {
118         struct rb_node *n;
119         struct ext3_reserve_window_node *rsv, *prev;
120         int bad;
121 
122 restart:
123         n = rb_first(root);
124         bad = 0;
125         prev = NULL;
126 
127         printk("Block Allocation Reservation Windows Map (%s):\n", fn);
128         while (n) {
129                 rsv = list_entry(n, struct ext3_reserve_window_node, rsv_node);
130                 if (verbose)
131                         printk("reservation window 0x%p "
132                                "start:  %d, end:  %d\n",
133                                rsv, rsv->rsv_start, rsv->rsv_end);
134                 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
135                         printk("Bad reservation %p (start >= end)\n",
136                                rsv);
137                         bad = 1;
138                 }
139                 if (prev && prev->rsv_end >= rsv->rsv_start) {
140                         printk("Bad reservation %p (prev->end >= start)\n",
141                                rsv);
142                         bad = 1;
143                 }
144                 if (bad) {
145                         if (!verbose) {
146                                 printk("Restarting reservation walk in verbose mode\n");
147                                 verbose = 1;
148                                 goto restart;
149                         }
150                 }
151                 n = rb_next(n);
152                 prev = rsv;
153         }
154         printk("Window map complete.\n");
155         if (bad)
156                 BUG();
157 }
158 #define rsv_window_dump(root, verbose) \
159         __rsv_window_dump((root), (verbose), __FUNCTION__)
160 #else
161 #define rsv_window_dump(root, verbose) do {} while (0)
162 #endif
163 
164 static int
165 goal_in_my_reservation(struct ext3_reserve_window *rsv, int goal,
166                         unsigned int group, struct super_block * sb)
167 {
168         unsigned long group_first_block, group_last_block;
169 
170         group_first_block = le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) +
171                                 group * EXT3_BLOCKS_PER_GROUP(sb);
172         group_last_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1;
173 
174         if ((rsv->_rsv_start > group_last_block) ||
175             (rsv->_rsv_end < group_first_block))
176                 return 0;
177         if ((goal >= 0) && ((goal + group_first_block < rsv->_rsv_start)
178                 || (goal + group_first_block > rsv->_rsv_end)))
179                 return 0;
180         return 1;
181 }
182 
183 /*
184  * Find the reserved window which includes the goal, or the previous one
185  * if the goal is not in any window.
186  * Returns NULL if there are no windows or if all windows start after the goal.
187  */
188 static struct ext3_reserve_window_node *
189 search_reserve_window(struct rb_root *root, unsigned long goal)
190 {
191         struct rb_node *n = root->rb_node;
192         struct ext3_reserve_window_node *rsv;
193 
194         if (!n)
195                 return NULL;
196 
197         do {
198                 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
199 
200                 if (goal < rsv->rsv_start)
201                         n = n->rb_left;
202                 else if (goal > rsv->rsv_end)
203                         n = n->rb_right;
204                 else
205                         return rsv;
206         } while (n);
207         /*
208          * We've fallen off the end of the tree: the goal wasn't inside
209          * any particular node.  OK, the previous node must be to one
210          * side of the interval containing the goal.  If it's the RHS,
211          * we need to back up one.
212          */
213         if (rsv->rsv_start > goal) {
214                 n = rb_prev(&rsv->rsv_node);
215                 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
216         }
217         return rsv;
218 }
219 
220 void ext3_rsv_window_add(struct super_block *sb,
221                     struct ext3_reserve_window_node *rsv)
222 {
223         struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
224         struct rb_node *node = &rsv->rsv_node;
225         unsigned int start = rsv->rsv_start;
226 
227         struct rb_node ** p = &root->rb_node;
228         struct rb_node * parent = NULL;
229         struct ext3_reserve_window_node *this;
230 
231         while (*p)
232         {
233                 parent = *p;
234                 this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
235 
236                 if (start < this->rsv_start)
237                         p = &(*p)->rb_left;
238                 else if (start > this->rsv_end)
239                         p = &(*p)->rb_right;
240                 else
241                         BUG();
242         }
243 
244         rb_link_node(node, parent, p);
245         rb_insert_color(node, root);
246 }
247 
248 static void rsv_window_remove(struct super_block *sb,
249                               struct ext3_reserve_window_node *rsv)
250 {
251         rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
252         rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
253         atomic_set(&rsv->rsv_alloc_hit, 0);
254         rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
255 }
256 
257 static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
258 {
259         /* a valid reservation end block could not be 0 */
260         return (rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED);
261 }
262 
263 void ext3_discard_reservation(struct inode *inode)
264 {
265         struct ext3_inode_info *ei = EXT3_I(inode);
266         struct ext3_reserve_window_node *rsv = &ei->i_rsv_window;
267         spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
268 
269         if (!rsv_is_empty(&rsv->rsv_window)) {
270                 spin_lock(rsv_lock);
271                 rsv_window_remove(inode->i_sb, rsv);
272                 spin_unlock(rsv_lock);
273         }
274 }
275 
276 /* Free given blocks, update quota and i_blocks field */
277 void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
278                          unsigned long block, unsigned long count,
279                          int *pdquot_freed_blocks)
280 {
281         struct buffer_head *bitmap_bh = NULL;
282         struct buffer_head *gd_bh;
283         unsigned long block_group;
284         unsigned long bit;
285         unsigned long i;
286         unsigned long overflow;
287         struct ext3_group_desc * gdp;
288         struct ext3_super_block * es;
289         struct ext3_sb_info *sbi;
290         int err = 0, ret;
291 
292         *pdquot_freed_blocks = 0;
293         sbi = EXT3_SB(sb);
294         es = EXT3_SB(sb)->s_es;
295         if (block < le32_to_cpu(es->s_first_data_block) ||
296             block + count < block ||
297             block + count > le32_to_cpu(es->s_blocks_count)) {
298                 ext3_error (sb, "ext3_free_blocks",
299                             "Freeing blocks not in datazone - "
300                             "block = %lu, count = %lu", block, count);
301                 goto error_return;
302         }
303 
304         ext3_debug ("freeing block %lu\n", block);
305 
306 do_more:
307         overflow = 0;
308         block_group = (block - le32_to_cpu(es->s_first_data_block)) /
309                       EXT3_BLOCKS_PER_GROUP(sb);
310         bit = (block - le32_to_cpu(es->s_first_data_block)) %
311                       EXT3_BLOCKS_PER_GROUP(sb);
312         /*
313          * Check to see if we are freeing blocks across a group
314          * boundary.
315          */
316         if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
317                 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
318                 count -= overflow;
319         }
320         brelse(bitmap_bh);
321         bitmap_bh = read_block_bitmap(sb, block_group);
322         if (!bitmap_bh)
323                 goto error_return;
324         gdp = ext3_get_group_desc (sb, block_group, &gd_bh);
325         if (!gdp)
326                 goto error_return;
327 
328         if (in_range (le32_to_cpu(gdp->bg_block_bitmap), block, count) ||
329             in_range (le32_to_cpu(gdp->bg_inode_bitmap), block, count) ||
330             in_range (block, le32_to_cpu(gdp->bg_inode_table),
331                       EXT3_SB(sb)->s_itb_per_group) ||
332             in_range (block + count - 1, le32_to_cpu(gdp->bg_inode_table),
333                       EXT3_SB(sb)->s_itb_per_group))
334                 ext3_error (sb, "ext3_free_blocks",
335                             "Freeing blocks in system zones - "
336                             "Block = %lu, count = %lu",
337                             block, count);
338 
339         /*
340          * We are about to start releasing blocks in the bitmap,
341          * so we need undo access.
342          */
343         /* @@@ check errors */
344         BUFFER_TRACE(bitmap_bh, "getting undo access");
345         err = ext3_journal_get_undo_access(handle, bitmap_bh, NULL);
346         if (err)
347                 goto error_return;
348 
349         /*
350          * We are about to modify some metadata.  Call the journal APIs
351          * to unshare ->b_data if a currently-committing transaction is
352          * using it
353          */
354         BUFFER_TRACE(gd_bh, "get_write_access");
355         err = ext3_journal_get_write_access(handle, gd_bh);
356         if (err)
357                 goto error_return;
358 
359         jbd_lock_bh_state(bitmap_bh);
360 
361         for (i = 0; i < count; i++) {
362                 /*
363                  * An HJ special.  This is expensive...
364                  */
365 #ifdef CONFIG_JBD_DEBUG
366                 jbd_unlock_bh_state(bitmap_bh);
367                 {
368                         struct buffer_head *debug_bh;
369                         debug_bh = sb_find_get_block(sb, block + i);
370                         if (debug_bh) {
371                                 BUFFER_TRACE(debug_bh, "Deleted!");
372                                 if (!bh2jh(bitmap_bh)->b_committed_data)
373                                         BUFFER_TRACE(debug_bh,
374                                                 "No commited data in bitmap");
375                                 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
376                                 __brelse(debug_bh);
377                         }
378                 }
379                 jbd_lock_bh_state(bitmap_bh);
380 #endif
381                 if (need_resched()) {
382                         jbd_unlock_bh_state(bitmap_bh);
383                         cond_resched();
384                         jbd_lock_bh_state(bitmap_bh);
385                 }
386                 /* @@@ This prevents newly-allocated data from being
387                  * freed and then reallocated within the same
388                  * transaction. 
389                  * 
390                  * Ideally we would want to allow that to happen, but to
391                  * do so requires making journal_forget() capable of
392                  * revoking the queued write of a data block, which
393                  * implies blocking on the journal lock.  *forget()
394                  * cannot block due to truncate races.
395                  *
396                  * Eventually we can fix this by making journal_forget()
397                  * return a status indicating whether or not it was able
398                  * to revoke the buffer.  On successful revoke, it is
399                  * safe not to set the allocation bit in the committed
400                  * bitmap, because we know that there is no outstanding
401                  * activity on the buffer any more and so it is safe to
402                  * reallocate it.  
403                  */
404                 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
405                 J_ASSERT_BH(bitmap_bh,
406                                 bh2jh(bitmap_bh)->b_committed_data != NULL);
407                 ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
408                                 bh2jh(bitmap_bh)->b_committed_data);
409 
410                 /*
411                  * We clear the bit in the bitmap after setting the committed
412                  * data bit, because this is the reverse order to that which
413                  * the allocator uses.
414                  */
415                 BUFFER_TRACE(bitmap_bh, "clear bit");
416                 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
417                                                 bit + i, bitmap_bh->b_data)) {
418                         jbd_unlock_bh_state(bitmap_bh);
419                         ext3_error(sb, __FUNCTION__,
420                                 "bit already cleared for block %lu", block + i);
421                         jbd_lock_bh_state(bitmap_bh);
422                         BUFFER_TRACE(bitmap_bh, "bit already cleared");
423                 } else {
424                         (*pdquot_freed_blocks)++;
425                 }
426         }
427         jbd_unlock_bh_state(bitmap_bh);
428 
429         spin_lock(sb_bgl_lock(sbi, block_group));
430         gdp->bg_free_blocks_count =
431                 cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count) +
432                         *pdquot_freed_blocks);
433         spin_unlock(sb_bgl_lock(sbi, block_group));
434         percpu_counter_mod(&sbi->s_freeblocks_counter, count);
435 
436         /* We dirtied the bitmap block */
437         BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
438         err = ext3_journal_dirty_metadata(handle, bitmap_bh);
439 
440         /* And the group descriptor block */
441         BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
442         ret = ext3_journal_dirty_metadata(handle, gd_bh);
443         if (!err) err = ret;
444 
445         if (overflow && !err) {
446                 block += count;
447                 count = overflow;
448                 goto do_more;
449         }
450         sb->s_dirt = 1;
451 error_return:
452         brelse(bitmap_bh);
453         ext3_std_error(sb, err);
454         return;
455 }
456 
457 /* Free given blocks, update quota and i_blocks field */
458 void ext3_free_blocks(handle_t *handle, struct inode *inode,
459                         unsigned long block, unsigned long count)
460 {
461         struct super_block * sb;
462         int dquot_freed_blocks;
463 
464         sb = inode->i_sb;
465         if (!sb) {
466                 printk ("ext3_free_blocks: nonexistent device");
467                 return;
468         }
469         ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
470         if (dquot_freed_blocks)
471                 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
472         return;
473 }
474 
475 /*
476  * For ext3 allocations, we must not reuse any blocks which are
477  * allocated in the bitmap buffer's "last committed data" copy.  This
478  * prevents deletes from freeing up the page for reuse until we have
479  * committed the delete transaction.
480  *
481  * If we didn't do this, then deleting something and reallocating it as
482  * data would allow the old block to be overwritten before the
483  * transaction committed (because we force data to disk before commit).
484  * This would lead to corruption if we crashed between overwriting the
485  * data and committing the delete. 
486  *
487  * @@@ We may want to make this allocation behaviour conditional on
488  * data-writes at some point, and disable it for metadata allocations or
489  * sync-data inodes.
490  */
491 static int ext3_test_allocatable(int nr, struct buffer_head *bh)
492 {
493         int ret;
494         struct journal_head *jh = bh2jh(bh);
495 
496         if (ext3_test_bit(nr, bh->b_data))
497                 return 0;
498 
499         jbd_lock_bh_state(bh);
500         if (!jh->b_committed_data)
501                 ret = 1;
502         else
503                 ret = !ext3_test_bit(nr, jh->b_committed_data);
504         jbd_unlock_bh_state(bh);
505         return ret;
506 }
507 
508 static int
509 bitmap_search_next_usable_block(int start, struct buffer_head *bh,
510                                         int maxblocks)
511 {
512         int next;
513         struct journal_head *jh = bh2jh(bh);
514 
515         /*
516          * The bitmap search --- search forward alternately through the actual
517          * bitmap and the last-committed copy until we find a bit free in
518          * both
519          */
520         while (start < maxblocks) {
521                 next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
522                 if (next >= maxblocks)
523                         return -1;
524                 if (ext3_test_allocatable(next, bh))
525                         return next;
526                 jbd_lock_bh_state(bh);
527                 if (jh->b_committed_data)
528                         start = ext3_find_next_zero_bit(jh->b_committed_data,
529                                                         maxblocks, next);
530                 jbd_unlock_bh_state(bh);
531         }
532         return -1;
533 }
534 
535 /*
536  * Find an allocatable block in a bitmap.  We honour both the bitmap and
537  * its last-committed copy (if that exists), and perform the "most
538  * appropriate allocation" algorithm of looking for a free block near
539  * the initial goal; then for a free byte somewhere in the bitmap; then
540  * for any free bit in the bitmap.
541  */
542 static int
543 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
544 {
545         int here, next;
546         char *p, *r;
547 
548         if (start > 0) {
549                 /*
550                  * The goal was occupied; search forward for a free 
551                  * block within the next XX blocks.
552                  *
553                  * end_goal is more or less random, but it has to be
554                  * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
555                  * next 64-bit boundary is simple..
556                  */
557                 int end_goal = (start + 63) & ~63;
558                 if (end_goal > maxblocks)
559                         end_goal = maxblocks;
560                 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
561                 if (here < end_goal && ext3_test_allocatable(here, bh))
562                         return here;
563                 ext3_debug("Bit not found near goal\n");
564         }
565 
566         here = start;
567         if (here < 0)
568                 here = 0;
569 
570         p = ((char *)bh->b_data) + (here >> 3);
571         r = memscan(p, 0, (maxblocks - here + 7) >> 3);
572         next = (r - ((char *)bh->b_data)) << 3;
573 
574         if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
575                 return next;
576 
577         /*
578          * The bitmap search --- search forward alternately through the actual
579          * bitmap and the last-committed copy until we find a bit free in
580          * both
581          */
582         here = bitmap_search_next_usable_block(here, bh, maxblocks);
583         return here;
584 }
585 
586 /*
587  * We think we can allocate this block in this bitmap.  Try to set the bit.
588  * If that succeeds then check that nobody has allocated and then freed the
589  * block since we saw that is was not marked in b_committed_data.  If it _was_
590  * allocated and freed then clear the bit in the bitmap again and return
591  * zero (failure).
592  */
593 static inline int
594 claim_block(spinlock_t *lock, int block, struct buffer_head *bh)
595 {
596         struct journal_head *jh = bh2jh(bh);
597         int ret;
598 
599         if (ext3_set_bit_atomic(lock, block, bh->b_data))
600                 return 0;
601         jbd_lock_bh_state(bh);
602         if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
603                 ext3_clear_bit_atomic(lock, block, bh->b_data);
604                 ret = 0;
605         } else {
606                 ret = 1;
607         }
608         jbd_unlock_bh_state(bh);
609         return ret;
610 }
611 
612 /*
613  * If we failed to allocate the desired block then we may end up crossing to a
614  * new bitmap.  In that case we must release write access to the old one via
615  * ext3_journal_release_buffer(), else we'll run out of credits.
616  */
617 static int
618 ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
619         struct buffer_head *bitmap_bh, int goal, struct ext3_reserve_window *my_rsv)
620 {
621         int group_first_block, start, end;
622 
623         /* we do allocation within the reservation window if we have a window */
624         if (my_rsv) {
625                 group_first_block =
626                         le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) +
627                         group * EXT3_BLOCKS_PER_GROUP(sb);
628                 if (my_rsv->_rsv_start >= group_first_block)
629                         start = my_rsv->_rsv_start - group_first_block;
630                 else
631                         /* reservation window cross group boundary */
632                         start = 0;
633                 end = my_rsv->_rsv_end - group_first_block + 1;
634                 if (end > EXT3_BLOCKS_PER_GROUP(sb))
635                         /* reservation window crosses group boundary */
636                         end = EXT3_BLOCKS_PER_GROUP(sb);
637                 if ((start <= goal) && (goal < end))
638                         start = goal;
639                 else
640                         goal = -1;
641         } else {
642                 if (goal > 0)
643                         start = goal;
644                 else
645                         start = 0;
646                 end = EXT3_BLOCKS_PER_GROUP(sb);
647         }
648 
649         BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
650 
651 repeat:
652         if (goal < 0 || !ext3_test_allocatable(goal, bitmap_bh)) {
653                 goal = find_next_usable_block(start, bitmap_bh, end);
654                 if (goal < 0)
655                         goto fail_access;
656                 if (!my_rsv) {
657                         int i;
658 
659                         for (i = 0; i < 7 && goal > start &&
660                                         ext3_test_allocatable(goal - 1,
661                                                                 bitmap_bh);
662                                         i++, goal--)
663                                 ;
664                 }
665         }
666         start = goal;
667 
668         if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group), goal, bitmap_bh)) {
669                 /*
670                  * The block was allocated by another thread, or it was
671                  * allocated and then freed by another thread
672                  */
673                 start++;
674                 goal++;
675                 if (start >= end)
676                         goto fail_access;
677                 goto repeat;
678         }
679         return goal;
680 fail_access:
681         return -1;
682 }
683 
684 /**
685  *      find_next_reservable_window():
686  *              find a reservable space within the given range.
687  *              It does not allocate the reservation window for now:
688  *              alloc_new_reservation() will do the work later.
689  *
690  *      @search_head: the head of the searching list;
691  *              This is not necessarily the list head of the whole filesystem
692  *
693  *              We have both head and start_block to assist the search
694  *              for the reservable space. The list starts from head,
695  *              but we will shift to the place where start_block is,
696  *              then start from there, when looking for a reservable space.
697  *
698  *      @size: the target new reservation window size
699  *
700  *      @group_first_block: the first block we consider to start
701  *                      the real search from
702  *
703  *      @last_block:
704  *              the maximum block number that our goal reservable space
705  *              could start from. This is normally the last block in this
706  *              group. The search will end when we found the start of next
707  *              possible reservable space is out of this boundary.
708  *              This could handle the cross boundary reservation window
709  *              request.
710  *
711  *      basically we search from the given range, rather than the whole
712  *      reservation double linked list, (start_block, last_block)
713  *      to find a free region that is of my size and has not
714  *      been reserved.
715  *
716  *      on succeed, it returns the reservation window to be appended to.
717  *      failed, return NULL.
718  */
719 static struct ext3_reserve_window_node *find_next_reservable_window(
720                                 struct ext3_reserve_window_node *search_head,
721                                 unsigned long size, int *start_block,
722                                 int last_block)
723 {
724         struct rb_node *next;
725         struct ext3_reserve_window_node *rsv, *prev;
726         int cur;
727 
728         /* TODO: make the start of the reservation window byte-aligned */
729         /* cur = *start_block & ~7;*/
730         cur = *start_block;
731         rsv = search_head;
732         if (!rsv)
733                 return NULL;
734 
735         while (1) {
736                 if (cur <= rsv->rsv_end)
737                         cur = rsv->rsv_end + 1;
738 
739                 /* TODO?
740                  * in the case we could not find a reservable space
741                  * that is what is expected, during the re-search, we could
742                  * remember what's the largest reservable space we could have
743                  * and return that one.
744                  *
745                  * For now it will fail if we could not find the reservable
746                  * space with expected-size (or more)...
747                  */
748                 if (cur > last_block)
749                         return NULL;            /* fail */
750 
751                 prev = rsv;
752                 next = rb_next(&rsv->rsv_node);
753                 rsv = list_entry(next, struct ext3_reserve_window_node, rsv_node);
754 
755                 /*
756                  * Reached the last reservation, we can just append to the
757                  * previous one.
758                  */
759                 if (!next)
760                         break;
761 
762                 if (cur + size <= rsv->rsv_start) {
763                         /*
764                          * Found a reserveable space big enough.  We could
765                          * have a reservation across the group boundary here
766                          */
767                         break;
768                 }
769         }
770         /*
771          * we come here either :
772          * when we reach the end of the whole list,
773          * and there is empty reservable space after last entry in the list.
774          * append it to the end of the list.
775          *
776          * or we found one reservable space in the middle of the list,
777          * return the reservation window that we could append to.
778          * succeed.
779          */
780         *start_block = cur;
781         return prev;
782 }
783 
784 /**
785  *      alloc_new_reservation()--allocate a new reservation window
786  *
787  *              To make a new reservation, we search part of the filesystem
788  *              reservation list (the list that inside the group). We try to
789  *              allocate a new reservation window near the allocation goal,
790  *              or the beginning of the group, if there is no goal.
791  *
792  *              We first find a reservable space after the goal, then from
793  *              there, we check the bitmap for the first free block after
794  *              it. If there is no free block until the end of group, then the
795  *              whole group is full, we failed. Otherwise, check if the free
796  *              block is inside the expected reservable space, if so, we
797  *              succeed.
798  *              If the first free block is outside the reservable space, then
799  *              start from the first free block, we search for next available
800  *              space, and go on.
801  *
802  *      on succeed, a new reservation will be found and inserted into the list
803  *      It contains at least one free block, and it does not overlap with other
804  *      reservation windows.
805  *
806  *      failed: we failed to find a reservation window in this group
807  *
808  *      @rsv: the reservation
809  *
810  *      @goal: The goal (group-relative).  It is where the search for a
811  *              free reservable space should start from.
812  *              if we have a goal(goal >0 ), then start from there,
813  *              no goal(goal = -1), we start from the first block
814  *              of the group.
815  *
816  *      @sb: the super block
817  *      @group: the group we are trying to allocate in
818  *      @bitmap_bh: the block group block bitmap
819  */
820 static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
821                 int goal, struct super_block *sb,
822                 unsigned int group, struct buffer_head *bitmap_bh)
823 {
824         struct ext3_reserve_window_node *search_head;
825         int group_first_block, group_end_block, start_block;
826         int first_free_block;
827         int reservable_space_start;
828         struct ext3_reserve_window_node *prev_rsv;
829         struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
830         unsigned long size;
831 
832         group_first_block = le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) +
833                                 group * EXT3_BLOCKS_PER_GROUP(sb);
834         group_end_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1;
835 
836         if (goal < 0)
837                 start_block = group_first_block;
838         else
839                 start_block = goal + group_first_block;
840 
841         size = atomic_read(&my_rsv->rsv_goal_size);
842         if (!rsv_is_empty(&my_rsv->rsv_window)) {
843                 /*
844                  * if the old reservation is cross group boundary
845                  * and if the goal is inside the old reservation window,
846                  * we will come here when we just failed to allocate from
847                  * the first part of the window. We still have another part
848                  * that belongs to the next group. In this case, there is no
849                  * point to discard our window and try to allocate a new one
850                  * in this group(which will fail). we should
851                  * keep the reservation window, just simply move on.
852                  *
853                  * Maybe we could shift the start block of the reservation
854                  * window to the first block of next group.
855                  */
856 
857                 if ((my_rsv->rsv_start <= group_end_block) &&
858                                 (my_rsv->rsv_end > group_end_block) &&
859                                 (start_block >= my_rsv->rsv_start))
860                         return -1;
861 
862                 if ((atomic_read(&my_rsv->rsv_alloc_hit) >
863                      (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
864                         /*
865                          * if we previously allocation hit ration is greater than half
866                          * we double the size of reservation window next time
867                          * otherwise keep the same
868                          */
869                         size = size * 2;
870                         if (size > EXT3_MAX_RESERVE_BLOCKS)
871                                 size = EXT3_MAX_RESERVE_BLOCKS;
872                         atomic_set(&my_rsv->rsv_goal_size, size);
873                 }
874         }
875         /*
876          * shift the search start to the window near the goal block
877          */
878         search_head = search_reserve_window(fs_rsv_root, start_block);
879 
880         /*
881          * find_next_reservable_window() simply finds a reservable window
882          * inside the given range(start_block, group_end_block).
883          *
884          * To make sure the reservation window has a free bit inside it, we
885          * need to check the bitmap after we found a reservable window.
886          */
887 retry:
888         prev_rsv = find_next_reservable_window(search_head, size,
889                                                 &start_block, group_end_block);
890         if (prev_rsv == NULL)
891                 goto failed;
892         reservable_space_start = start_block;
893         /*
894          * On success, find_next_reservable_window() returns the
895          * reservation window where there is a reservable space after it.
896          * Before we reserve this reservable space, we need
897          * to make sure there is at least a free block inside this region.
898          *
899          * searching the first free bit on the block bitmap and copy of
900          * last committed bitmap alternatively, until we found a allocatable
901          * block. Search start from the start block of the reservable space
902          * we just found.
903          */
904         first_free_block = bitmap_search_next_usable_block(
905                         reservable_space_start - group_first_block,
906                         bitmap_bh, group_end_block - group_first_block + 1);
907 
908         if (first_free_block < 0) {
909                 /*
910                  * no free block left on the bitmap, no point
911                  * to reserve the space. return failed.
912                  */
913                 goto failed;
914         }
915         start_block = first_free_block + group_first_block;
916         /*
917          * check if the first free block is within the
918          * free space we just found
919          */
920         if ((start_block >= reservable_space_start) &&
921           (start_block < reservable_space_start + size))
922                 goto found_rsv_window;
923         /*
924          * if the first free bit we found is out of the reservable space
925          * this means there is no free block on the reservable space
926          * we should continue search for next reservable space,
927          * start from where the free block is,
928          * we also shift the list head to where we stopped last time
929          */
930         search_head = prev_rsv;
931         goto retry;
932 
933 found_rsv_window:
934         /*
935          * great! the reservable space contains some free blocks.
936          * if the search returns that we should add the new
937          * window just next to where the old window, we don't
938          * need to remove the old window first then add it to the
939          * same place, just update the new start and new end.
940          */
941         if (my_rsv != prev_rsv)  {
942                 if (!rsv_is_empty(&my_rsv->rsv_window))
943                         rsv_window_remove(sb, my_rsv);
944         }
945         my_rsv->rsv_start = reservable_space_start;
946         my_rsv->rsv_end = my_rsv->rsv_start + size - 1;
947         atomic_set(&my_rsv->rsv_alloc_hit, 0);
948         if (my_rsv != prev_rsv)  {
949                 ext3_rsv_window_add(sb, my_rsv);
950         }
951         return 0;               /* succeed */
952 failed:
953         /*
954          * failed to find a new reservation window in the current
955          * group, remove the current(stale) reservation window
956          * if there is any
957          */
958         if (!rsv_is_empty(&my_rsv->rsv_window))
959                 rsv_window_remove(sb, my_rsv);
960         return -1;              /* failed */
961 }
962 
963 /*
964  * This is the main function used to allocate a new block and its reservation
965  * window.
966  *
967  * Each time when a new block allocation is need, first try to allocate from
968  * its own reservation.  If it does not have a reservation window, instead of
969  * looking for a free bit on bitmap first, then look up the reservation list to
970  * see if it is inside somebody else's reservation window, we try to allocate a
971  * reservation window for it starting from the goal first. Then do the block
972  * allocation within the reservation window.
973  *
974  * This will avoid keeping on searching the reservation list again and
975  * again when someboday is looking for a free block (without
976  * reservation), and there are lots of free blocks, but they are all
977  * being reserved.
978  *
979  * We use a sorted double linked list for the per-filesystem reservation list.
980  * The insert, remove and find a free space(non-reserved) operations for the
981  * sorted double linked list should be fast.
982  *
983  */
984 static int
985 ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
986                         unsigned int group, struct buffer_head *bitmap_bh,
987                         int goal, struct ext3_reserve_window_node * my_rsv,
988                         int *errp)
989 {
990         spinlock_t *rsv_lock;
991         unsigned long group_first_block;
992         int ret = 0;
993         int fatal;
994         int credits = 0;
995 
996         *errp = 0;
997 
998         /*
999          * Make sure we use undo access for the bitmap, because it is critical
1000          * that we do the frozen_data COW on bitmap buffers in all cases even
1001          * if the buffer is in BJ_Forget state in the committing transaction.
1002          */
1003         BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1004         fatal = ext3_journal_get_undo_access(handle, bitmap_bh, &credits);
1005         if (fatal) {
1006                 *errp = fatal;
1007                 return -1;
1008         }
1009 
1010         /*
1011          * we don't deal with reservation when
1012          * filesystem is mounted without reservation
1013          * or the file is not a regular file
1014          * or last attempt to allocate a block with reservation turned on failed
1015          */
1016         if (my_rsv == NULL ) {
1017                 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh, goal, NULL);
1018                 goto out;
1019         }
1020         rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1021         /*
1022          * goal is a group relative block number (if there is a goal)
1023          * 0 < goal < EXT3_BLOCKS_PER_GROUP(sb)
1024          * first block is a filesystem wide block number
1025          * first block is the block number of the first block in this group
1026          */
1027         group_first_block = le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) +
1028                         group * EXT3_BLOCKS_PER_GROUP(sb);
1029 
1030         /*
1031          * Basically we will allocate a new block from inode's reservation
1032          * window.
1033          *
1034          * We need to allocate a new reservation window, if:
1035          * a) inode does not have a reservation window; or
1036          * b) last attempt to allocate a block from existing reservation
1037          *    failed; or
1038          * c) we come here with a goal and with a reservation window
1039          *
1040          * We do not need to allocate a new reservation window if we come here
1041          * at the beginning with a goal and the goal is inside the window, or
1042          * we don't have a goal but already have a reservation window.
1043          * then we could go to allocate from the reservation window directly.
1044          */
1045         while (1) {
1046                 struct ext3_reserve_window rsv_copy;
1047                 unsigned int seq;
1048 
1049                 do {
1050                         seq = read_seqbegin(&my_rsv->rsv_seqlock);
1051                         rsv_copy._rsv_start = my_rsv->rsv_start;
1052                         rsv_copy._rsv_end = my_rsv->rsv_end;
1053                 } while (read_seqretry(&my_rsv->rsv_seqlock, seq));
1054 
1055                 if (rsv_is_empty(&rsv_copy) || (ret < 0) ||
1056                         !goal_in_my_reservation(&rsv_copy, goal, group, sb)) {
1057                         spin_lock(rsv_lock);
1058                         write_seqlock(&my_rsv->rsv_seqlock);
1059                         ret = alloc_new_reservation(my_rsv, goal, sb,
1060                                                         group, bitmap_bh);
1061                         rsv_copy._rsv_start = my_rsv->rsv_start;
1062                         rsv_copy._rsv_end = my_rsv->rsv_end;
1063                         write_sequnlock(&my_rsv->rsv_seqlock);
1064                         spin_unlock(rsv_lock);
1065                         if (ret < 0)
1066                                 break;                  /* failed */
1067 
1068                         if (!goal_in_my_reservation(&rsv_copy, goal, group, sb))
1069                                 goal = -1;
1070                 }
1071                 if ((rsv_copy._rsv_start >= group_first_block + EXT3_BLOCKS_PER_GROUP(sb))
1072                     || (rsv_copy._rsv_end < group_first_block))
1073                         BUG();
1074                 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh, goal,
1075                                            &rsv_copy);
1076                 if (ret >= 0) {
1077                         if (!read_seqretry(&my_rsv->rsv_seqlock, seq))
1078                                 atomic_inc(&my_rsv->rsv_alloc_hit);
1079                         break;                          /* succeed */
1080                 }
1081         }
1082 out:
1083         if (ret >= 0) {
1084                 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1085                                         "bitmap block");
1086                 fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1087                 if (fatal) {
1088                         *errp = fatal;
1089                         return -1;
1090                 }
1091                 return ret;
1092         }
1093 
1094         BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1095         ext3_journal_release_buffer(handle, bitmap_bh, credits);
1096         return ret;
1097 }
1098 
1099 static int ext3_has_free_blocks(struct ext3_sb_info *sbi)
1100 {
1101         int free_blocks, root_blocks;
1102 
1103         free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1104         root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1105         if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1106                 sbi->s_resuid != current->fsuid &&
1107                 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1108                 return 0;
1109         }
1110         return 1;
1111 }
1112 
1113 /*
1114  * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1115  * it is profitable to retry the operation, this function will wait
1116  * for the current or commiting transaction to complete, and then
1117  * return TRUE.
1118  */
1119 int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1120 {
1121         if (!ext3_has_free_blocks(EXT3_SB(sb)) || (*retries)++ > 3)
1122                 return 0;
1123 
1124         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1125 
1126         return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1127 }
1128 
1129 /*
1130  * ext3_new_block uses a goal block to assist allocation.  If the goal is
1131  * free, or there is a free block within 32 blocks of the goal, that block
1132  * is allocated.  Otherwise a forward search is made for a free block; within 
1133  * each block group the search first looks for an entire free byte in the block
1134  * bitmap, and then for any free bit if that fails.
1135  * This function also updates quota and i_blocks field.
1136  */
1137 int ext3_new_block(handle_t *handle, struct inode *inode,
1138                         unsigned long goal, int *errp)
1139 {
1140         struct buffer_head *bitmap_bh = NULL;
1141         struct buffer_head *gdp_bh;
1142         int group_no;
1143         int goal_group;
1144         int ret_block;
1145         int bgi;                        /* blockgroup iteration index */
1146         int target_block;
1147         int fatal = 0, err;
1148         int performed_allocation = 0;
1149         int free_blocks;
1150         struct super_block *sb;
1151         struct ext3_group_desc *gdp;
1152         struct ext3_super_block *es;
1153         struct ext3_sb_info *sbi;
1154         struct ext3_reserve_window_node *my_rsv = NULL;
1155         struct ext3_reserve_window_node *rsv = &EXT3_I(inode)->i_rsv_window;
1156         unsigned short windowsz = 0;
1157 #ifdef EXT3FS_DEBUG
1158         static int goal_hits, goal_attempts;
1159 #endif
1160         unsigned long ngroups;
1161 
1162         *errp = -ENOSPC;
1163         sb = inode->i_sb;
1164         if (!sb) {
1165                 printk("ext3_new_block: nonexistent device");
1166                 return 0;
1167         }
1168 
1169         /*
1170          * Check quota for allocation of this block.
1171          */
1172         if (DQUOT_ALLOC_BLOCK(inode, 1)) {
1173                 *errp = -EDQUOT;
1174                 return 0;
1175         }
1176 
1177         sbi = EXT3_SB(sb);
1178         es = EXT3_SB(sb)->s_es;
1179         ext3_debug("goal=%lu.\n", goal);
1180         /*
1181          * Allocate a block from reservation only when
1182          * filesystem is mounted with reservation(default,-o reservation), and
1183          * it's a regular file, and
1184          * the desired window size is greater than 0 (One could use ioctl
1185          * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1186          * reservation on that particular file)
1187          */
1188         windowsz = atomic_read(&rsv->rsv_goal_size);
1189         if (test_opt(sb, RESERVATION) &&
1190                 S_ISREG(inode->i_mode) && (windowsz > 0))
1191                 my_rsv = rsv;
1192         if (!ext3_has_free_blocks(sbi)) {
1193                 *errp = -ENOSPC;
1194                 goto out;
1195         }
1196 
1197         /*
1198          * First, test whether the goal block is free.
1199          */
1200         if (goal < le32_to_cpu(es->s_first_data_block) ||
1201             goal >= le32_to_cpu(es->s_blocks_count))
1202                 goal = le32_to_cpu(es->s_first_data_block);
1203         group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1204                         EXT3_BLOCKS_PER_GROUP(sb);
1205         gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1206         if (!gdp)
1207                 goto io_error;
1208 
1209         goal_group = group_no;
1210 retry:
1211         free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1212         if (free_blocks > 0) {
1213                 ret_block = ((goal - le32_to_cpu(es->s_first_data_block)) %
1214                                 EXT3_BLOCKS_PER_GROUP(sb));
1215                 bitmap_bh = read_block_bitmap(sb, group_no);
1216                 if (!bitmap_bh)
1217                         goto io_error;
1218                 ret_block = ext3_try_to_allocate_with_rsv(sb, handle, group_no,
1219                                         bitmap_bh, ret_block, my_rsv, &fatal);
1220                 if (fatal)
1221                         goto out;
1222                 if (ret_block >= 0)
1223                         goto allocated;
1224         }
1225 
1226         ngroups = EXT3_SB(sb)->s_groups_count;
1227         smp_rmb();
1228 
1229         /*
1230          * Now search the rest of the groups.  We assume that 
1231          * i and gdp correctly point to the last group visited.
1232          */
1233         for (bgi = 0; bgi < ngroups; bgi++) {
1234                 group_no++;
1235                 if (group_no >= ngroups)
1236                         group_no = 0;
1237                 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1238                 if (!gdp) {
1239                         *errp = -EIO;
1240                         goto out;
1241                 }
1242                 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1243                 /*
1244                  * skip this group if the number of
1245                  * free blocks is less than half of the reservation
1246                  * window size.
1247                  */
1248                 if (free_blocks <= (windowsz/2))
1249                         continue;
1250 
1251                 brelse(bitmap_bh);
1252                 bitmap_bh = read_block_bitmap(sb, group_no);
1253                 if (!bitmap_bh)
1254                         goto io_error;
1255                 ret_block = ext3_try_to_allocate_with_rsv(sb, handle, group_no,
1256                                         bitmap_bh, -1, my_rsv, &fatal);
1257                 if (fatal)
1258                         goto out;
1259                 if (ret_block >= 0) 
1260                         goto allocated;
1261         }
1262         /*
1263          * We may end up a bogus ealier ENOSPC error due to
1264          * filesystem is "full" of reservations, but
1265          * there maybe indeed free blocks avaliable on disk
1266          * In this case, we just forget about the reservations
1267          * just do block allocation as without reservations.
1268          */
1269         if (my_rsv) {
1270                 my_rsv = NULL;
1271                 group_no = goal_group;
1272                 goto retry;
1273         }
1274         /* No space left on the device */
1275         *errp = -ENOSPC;
1276         goto out;
1277 
1278 allocated:
1279 
1280         ext3_debug("using block group %d(%d)\n",
1281                         group_no, gdp->bg_free_blocks_count);
1282 
1283         BUFFER_TRACE(gdp_bh, "get_write_access");
1284         fatal = ext3_journal_get_write_access(handle, gdp_bh);
1285         if (fatal)
1286                 goto out;
1287 
1288         target_block = ret_block + group_no * EXT3_BLOCKS_PER_GROUP(sb)
1289                                 + le32_to_cpu(es->s_first_data_block);
1290 
1291         if (target_block == le32_to_cpu(gdp->bg_block_bitmap) ||
1292             target_block == le32_to_cpu(gdp->bg_inode_bitmap) ||
1293             in_range(target_block, le32_to_cpu(gdp->bg_inode_table),
1294                       EXT3_SB(sb)->s_itb_per_group))
1295                 ext3_error(sb, "ext3_new_block",
1296                             "Allocating block in system zone - "
1297                             "block = %u", target_block);
1298 
1299         performed_allocation = 1;
1300 
1301 #ifdef CONFIG_JBD_DEBUG
1302         {
1303                 struct buffer_head *debug_bh;
1304 
1305                 /* Record bitmap buffer state in the newly allocated block */
1306                 debug_bh = sb_find_get_block(sb, target_block);
1307                 if (debug_bh) {
1308                         BUFFER_TRACE(debug_bh, "state when allocated");
1309                         BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1310                         brelse(debug_bh);
1311                 }
1312         }
1313         jbd_lock_bh_state(bitmap_bh);
1314         spin_lock(sb_bgl_lock(sbi, group_no));
1315         if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1316                 if (ext3_test_bit(ret_block,
1317                                 bh2jh(bitmap_bh)->b_committed_data)) {
1318                         printk("%s: block was unexpectedly set in "
1319                                 "b_committed_data\n", __FUNCTION__);
1320                 }
1321         }
1322         ext3_debug("found bit %d\n", ret_block);
1323         spin_unlock(sb_bgl_lock(sbi, group_no));
1324         jbd_unlock_bh_state(bitmap_bh);
1325 #endif
1326 
1327         /* ret_block was blockgroup-relative.  Now it becomes fs-relative */
1328         ret_block = target_block;
1329 
1330         if (ret_block >= le32_to_cpu(es->s_blocks_count)) {
1331                 ext3_error(sb, "ext3_new_block",
1332                             "block(%d) >= blocks count(%d) - "
1333                             "block_group = %d, es == %p ", ret_block,
1334                         le32_to_cpu(es->s_blocks_count), group_no, es);
1335                 goto out;
1336         }
1337 
1338         /*
1339          * It is up to the caller to add the new buffer to a journal
1340          * list of some description.  We don't know in advance whether
1341          * the caller wants to use it as metadata or data.
1342          */
1343         ext3_debug("allocating block %d. Goal hits %d of %d.\n",
1344                         ret_block, goal_hits, goal_attempts);
1345 
1346         spin_lock(sb_bgl_lock(sbi, group_no));
1347         gdp->bg_free_blocks_count =
1348                         cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count) - 1);
1349         spin_unlock(sb_bgl_lock(sbi, group_no));
1350         percpu_counter_mod(&sbi->s_freeblocks_counter, -1);
1351 
1352         BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1353         err = ext3_journal_dirty_metadata(handle, gdp_bh);
1354         if (!fatal)
1355                 fatal = err;
1356 
1357         sb->s_dirt = 1;
1358         if (fatal)
1359                 goto out;
1360 
1361         *errp = 0;
1362         brelse(bitmap_bh);
1363         return ret_block;
1364 
1365 io_error:
1366         *errp = -EIO;
1367 out:
1368         if (fatal) {
1369                 *errp = fatal;
1370                 ext3_std_error(sb, fatal);
1371         }
1372         /*
1373          * Undo the block allocation
1374          */
1375         if (!performed_allocation)
1376                 DQUOT_FREE_BLOCK(inode, 1);
1377         brelse(bitmap_bh);
1378         return 0;
1379 }
1380 
1381 unsigned long ext3_count_free_blocks(struct super_block *sb)
1382 {
1383         unsigned long desc_count;
1384         struct ext3_group_desc *gdp;
1385         int i;
1386         unsigned long ngroups;
1387 #ifdef EXT3FS_DEBUG
1388         struct ext3_super_block *es;
1389         unsigned long bitmap_count, x;
1390         struct buffer_head *bitmap_bh = NULL;
1391 
1392         lock_super(sb);
1393         es = EXT3_SB(sb)->s_es;
1394         desc_count = 0;
1395         bitmap_count = 0;
1396         gdp = NULL;
1397         for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
1398                 gdp = ext3_get_group_desc(sb, i, NULL);
1399                 if (!gdp)
1400                         continue;
1401                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1402                 brelse(bitmap_bh);
1403                 bitmap_bh = read_block_bitmap(sb, i);
1404                 if (bitmap_bh == NULL)
1405                         continue;
1406 
1407                 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1408                 printk("group %d: stored = %d, counted = %lu\n",
1409                         i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1410                 bitmap_count += x;
1411         }
1412         brelse(bitmap_bh);
1413         printk("ext3_count_free_blocks: stored = %u, computed = %lu, %lu\n",
1414                le32_to_cpu(es->s_free_blocks_count), desc_count, bitmap_count);
1415         unlock_super(sb);
1416         return bitmap_count;
1417 #else
1418         desc_count = 0;
1419         ngroups = EXT3_SB(sb)->s_groups_count;
1420         smp_rmb();
1421         for (i = 0; i < ngroups; i++) {
1422                 gdp = ext3_get_group_desc(sb, i, NULL);
1423                 if (!gdp)
1424                         continue;
1425                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1426         }
1427 
1428         return desc_count;
1429 #endif
1430 }
1431 
1432 static inline int block_in_use(unsigned long block,
1433                                 struct super_block * sb,
1434                                 unsigned char * map)
1435 {
1436         return ext3_test_bit ((block -
1437                 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block)) %
1438                          EXT3_BLOCKS_PER_GROUP(sb), map);
1439 }
1440 
1441 static inline int test_root(int a, int b)
1442 {
1443         int num = b;
1444 
1445         while (a > num)
1446                 num *= b;
1447         return num == a;
1448 }
1449 
1450 static int ext3_group_sparse(int group)
1451 {
1452         if (group <= 1)
1453                 return 1;
1454         return (test_root(group, 3) || test_root(group, 5) ||
1455                 test_root(group, 7));
1456 }
1457 
1458 /**
1459  *      ext3_bg_has_super - number of blocks used by the superblock in group
1460  *      @sb: superblock for filesystem
1461  *      @group: group number to check
1462  *
1463  *      Return the number of blocks used by the superblock (primary or backup)
1464  *      in this group.  Currently this will be only 0 or 1.
1465  */
1466 int ext3_bg_has_super(struct super_block *sb, int group)
1467 {
1468         if (EXT3_HAS_RO_COMPAT_FEATURE(sb,EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1469             !ext3_group_sparse(group))
1470                 return 0;
1471         return 1;
1472 }
1473 
1474 /**
1475  *      ext3_bg_num_gdb - number of blocks used by the group table in group
1476  *      @sb: superblock for filesystem
1477  *      @group: group number to check
1478  *
1479  *      Return the number of blocks used by the group descriptor table
1480  *      (primary or backup) in this group.  In the future there may be a
1481  *      different number of descriptor blocks in each group.
1482  */
1483 unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1484 {
1485         if (EXT3_HAS_RO_COMPAT_FEATURE(sb,EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1486             !ext3_group_sparse(group))
1487                 return 0;
1488         return EXT3_SB(sb)->s_gdb_count;
1489 }
1490 
1491 #ifdef CONFIG_EXT3_CHECK
1492 /* Called at mount-time, super-block is locked */
1493 void ext3_check_blocks_bitmap (struct super_block * sb)
1494 {
1495         struct ext3_super_block *es;
1496         unsigned long desc_count, bitmap_count, x, j;
1497         unsigned long desc_blocks;
1498         struct buffer_head *bitmap_bh = NULL;
1499         struct ext3_group_desc *gdp;
1500         int i;
1501 
1502         es = EXT3_SB(sb)->s_es;
1503         desc_count = 0;
1504         bitmap_count = 0;
1505         gdp = NULL;
1506         for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
1507                 gdp = ext3_get_group_desc (sb, i, NULL);
1508                 if (!gdp)
1509                         continue;
1510                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1511                 brelse(bitmap_bh);
1512                 bitmap_bh = read_block_bitmap(sb, i);
1513                 if (bitmap_bh == NULL)
1514                         continue;
1515 
1516                 if (ext3_bg_has_super(sb, i) &&
1517                                 !ext3_test_bit(0, bitmap_bh->b_data))
1518                         ext3_error(sb, __FUNCTION__,
1519                                    "Superblock in group %d is marked free", i);
1520 
1521                 desc_blocks = ext3_bg_num_gdb(sb, i);
1522                 for (j = 0; j < desc_blocks; j++)
1523                         if (!ext3_test_bit(j + 1, bitmap_bh->b_data))
1524                                 ext3_error(sb, __FUNCTION__,
1525                                            "Descriptor block #%ld in group "
1526                                            "%d is marked free", j, i);
1527 
1528                 if (!block_in_use (le32_to_cpu(gdp->bg_block_bitmap),
1529                                                 sb, bitmap_bh->b_data))
1530                         ext3_error (sb, "ext3_check_blocks_bitmap",
1531                                     "Block bitmap for group %d is marked free",
1532                                     i);
1533 
1534                 if (!block_in_use (le32_to_cpu(gdp->bg_inode_bitmap),
1535                                                 sb, bitmap_bh->b_data))
1536                         ext3_error (sb, "ext3_check_blocks_bitmap",
1537                                     "Inode bitmap for group %d is marked free",
1538                                     i);
1539 
1540                 for (j = 0; j < EXT3_SB(sb)->s_itb_per_group; j++)
1541                         if (!block_in_use (le32_to_cpu(gdp->bg_inode_table) + j,
1542                                                         sb, bitmap_bh->b_data))
1543                                 ext3_error (sb, "ext3_check_blocks_bitmap",
1544                                             "Block #%d of the inode table in "
1545                                             "group %d is marked free", j, i);
1546 
1547                 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1548                 if (le16_to_cpu(gdp->bg_free_blocks_count) != x)
1549                         ext3_error (sb, "ext3_check_blocks_bitmap",
1550                                     "Wrong free blocks count for group %d, "
1551                                     "stored = %d, counted = %lu", i,
1552                                     le16_to_cpu(gdp->bg_free_blocks_count), x);
1553                 bitmap_count += x;
1554         }
1555         brelse(bitmap_bh);
1556         if (le32_to_cpu(es->s_free_blocks_count) != bitmap_count)
1557                 ext3_error (sb, "ext3_check_blocks_bitmap",
1558                         "Wrong free blocks count in super block, "
1559                         "stored = %lu, counted = %lu",
1560                         (unsigned long)le32_to_cpu(es->s_free_blocks_count),
1561                         bitmap_count);
1562 }
1563 #endif
1564 
  This page was automatically generated by the LXR engine.