Merge git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw
[linux-2.6] / fs / ext3 / balloc.c
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/time.h>
15 #include <linux/capability.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_fill_super).
36  */
37
38
39 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
40
41 /**
42  * ext3_get_group_desc() -- load group descriptor from disk
43  * @sb:                 super block
44  * @block_group:        given block group
45  * @bh:                 pointer to the buffer head to store the block
46  *                      group descriptor
47  */
48 struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb,
49                                              unsigned int block_group,
50                                              struct buffer_head ** bh)
51 {
52         unsigned long group_desc;
53         unsigned long offset;
54         struct ext3_group_desc * desc;
55         struct ext3_sb_info *sbi = EXT3_SB(sb);
56
57         if (block_group >= sbi->s_groups_count) {
58                 ext3_error (sb, "ext3_get_group_desc",
59                             "block_group >= groups_count - "
60                             "block_group = %d, groups_count = %lu",
61                             block_group, sbi->s_groups_count);
62
63                 return NULL;
64         }
65         smp_rmb();
66
67         group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
68         offset = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
69         if (!sbi->s_group_desc[group_desc]) {
70                 ext3_error (sb, "ext3_get_group_desc",
71                             "Group descriptor not loaded - "
72                             "block_group = %d, group_desc = %lu, desc = %lu",
73                              block_group, group_desc, offset);
74                 return NULL;
75         }
76
77         desc = (struct ext3_group_desc *) sbi->s_group_desc[group_desc]->b_data;
78         if (bh)
79                 *bh = sbi->s_group_desc[group_desc];
80         return desc + offset;
81 }
82
83 static int ext3_valid_block_bitmap(struct super_block *sb,
84                                         struct ext3_group_desc *desc,
85                                         unsigned int block_group,
86                                         struct buffer_head *bh)
87 {
88         ext3_grpblk_t offset;
89         ext3_grpblk_t next_zero_bit;
90         ext3_fsblk_t bitmap_blk;
91         ext3_fsblk_t group_first_block;
92
93         group_first_block = ext3_group_first_block_no(sb, block_group);
94
95         /* check whether block bitmap block number is set */
96         bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
97         offset = bitmap_blk - group_first_block;
98         if (!ext3_test_bit(offset, bh->b_data))
99                 /* bad block bitmap */
100                 goto err_out;
101
102         /* check whether the inode bitmap block number is set */
103         bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
104         offset = bitmap_blk - group_first_block;
105         if (!ext3_test_bit(offset, bh->b_data))
106                 /* bad block bitmap */
107                 goto err_out;
108
109         /* check whether the inode table block number is set */
110         bitmap_blk = le32_to_cpu(desc->bg_inode_table);
111         offset = bitmap_blk - group_first_block;
112         next_zero_bit = ext3_find_next_zero_bit(bh->b_data,
113                                 offset + EXT3_SB(sb)->s_itb_per_group,
114                                 offset);
115         if (next_zero_bit >= offset + EXT3_SB(sb)->s_itb_per_group)
116                 /* good bitmap for inode tables */
117                 return 1;
118
119 err_out:
120         ext3_error(sb, __func__,
121                         "Invalid block bitmap - "
122                         "block_group = %d, block = %lu",
123                         block_group, bitmap_blk);
124         return 0;
125 }
126
127 /**
128  * read_block_bitmap()
129  * @sb:                 super block
130  * @block_group:        given block group
131  *
132  * Read the bitmap for a given block_group,and validate the
133  * bits for block/inode/inode tables are set in the bitmaps
134  *
135  * Return buffer_head on success or NULL in case of failure.
136  */
137 static struct buffer_head *
138 read_block_bitmap(struct super_block *sb, unsigned int block_group)
139 {
140         struct ext3_group_desc * desc;
141         struct buffer_head * bh = NULL;
142         ext3_fsblk_t bitmap_blk;
143
144         desc = ext3_get_group_desc(sb, block_group, NULL);
145         if (!desc)
146                 return NULL;
147         bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
148         bh = sb_getblk(sb, bitmap_blk);
149         if (unlikely(!bh)) {
150                 ext3_error(sb, __func__,
151                             "Cannot read block bitmap - "
152                             "block_group = %d, block_bitmap = %u",
153                             block_group, le32_to_cpu(desc->bg_block_bitmap));
154                 return NULL;
155         }
156         if (likely(bh_uptodate_or_lock(bh)))
157                 return bh;
158
159         if (bh_submit_read(bh) < 0) {
160                 brelse(bh);
161                 ext3_error(sb, __func__,
162                             "Cannot read block bitmap - "
163                             "block_group = %d, block_bitmap = %u",
164                             block_group, le32_to_cpu(desc->bg_block_bitmap));
165                 return NULL;
166         }
167         ext3_valid_block_bitmap(sb, desc, block_group, bh);
168         /*
169          * file system mounted not to panic on error, continue with corrupt
170          * bitmap
171          */
172         return bh;
173 }
174 /*
175  * The reservation window structure operations
176  * --------------------------------------------
177  * Operations include:
178  * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
179  *
180  * We use a red-black tree to represent per-filesystem reservation
181  * windows.
182  *
183  */
184
185 /**
186  * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
187  * @rb_root:            root of per-filesystem reservation rb tree
188  * @verbose:            verbose mode
189  * @fn:                 function which wishes to dump the reservation map
190  *
191  * If verbose is turned on, it will print the whole block reservation
192  * windows(start, end). Otherwise, it will only print out the "bad" windows,
193  * those windows that overlap with their immediate neighbors.
194  */
195 #if 1
196 static void __rsv_window_dump(struct rb_root *root, int verbose,
197                               const char *fn)
198 {
199         struct rb_node *n;
200         struct ext3_reserve_window_node *rsv, *prev;
201         int bad;
202
203 restart:
204         n = rb_first(root);
205         bad = 0;
206         prev = NULL;
207
208         printk("Block Allocation Reservation Windows Map (%s):\n", fn);
209         while (n) {
210                 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
211                 if (verbose)
212                         printk("reservation window 0x%p "
213                                "start:  %lu, end:  %lu\n",
214                                rsv, rsv->rsv_start, rsv->rsv_end);
215                 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
216                         printk("Bad reservation %p (start >= end)\n",
217                                rsv);
218                         bad = 1;
219                 }
220                 if (prev && prev->rsv_end >= rsv->rsv_start) {
221                         printk("Bad reservation %p (prev->end >= start)\n",
222                                rsv);
223                         bad = 1;
224                 }
225                 if (bad) {
226                         if (!verbose) {
227                                 printk("Restarting reservation walk in verbose mode\n");
228                                 verbose = 1;
229                                 goto restart;
230                         }
231                 }
232                 n = rb_next(n);
233                 prev = rsv;
234         }
235         printk("Window map complete.\n");
236         BUG_ON(bad);
237 }
238 #define rsv_window_dump(root, verbose) \
239         __rsv_window_dump((root), (verbose), __func__)
240 #else
241 #define rsv_window_dump(root, verbose) do {} while (0)
242 #endif
243
244 /**
245  * goal_in_my_reservation()
246  * @rsv:                inode's reservation window
247  * @grp_goal:           given goal block relative to the allocation block group
248  * @group:              the current allocation block group
249  * @sb:                 filesystem super block
250  *
251  * Test if the given goal block (group relative) is within the file's
252  * own block reservation window range.
253  *
254  * If the reservation window is outside the goal allocation group, return 0;
255  * grp_goal (given goal block) could be -1, which means no specific
256  * goal block. In this case, always return 1.
257  * If the goal block is within the reservation window, return 1;
258  * otherwise, return 0;
259  */
260 static int
261 goal_in_my_reservation(struct ext3_reserve_window *rsv, ext3_grpblk_t grp_goal,
262                         unsigned int group, struct super_block * sb)
263 {
264         ext3_fsblk_t group_first_block, group_last_block;
265
266         group_first_block = ext3_group_first_block_no(sb, group);
267         group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
268
269         if ((rsv->_rsv_start > group_last_block) ||
270             (rsv->_rsv_end < group_first_block))
271                 return 0;
272         if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
273                 || (grp_goal + group_first_block > rsv->_rsv_end)))
274                 return 0;
275         return 1;
276 }
277
278 /**
279  * search_reserve_window()
280  * @rb_root:            root of reservation tree
281  * @goal:               target allocation block
282  *
283  * Find the reserved window which includes the goal, or the previous one
284  * if the goal is not in any window.
285  * Returns NULL if there are no windows or if all windows start after the goal.
286  */
287 static struct ext3_reserve_window_node *
288 search_reserve_window(struct rb_root *root, ext3_fsblk_t goal)
289 {
290         struct rb_node *n = root->rb_node;
291         struct ext3_reserve_window_node *rsv;
292
293         if (!n)
294                 return NULL;
295
296         do {
297                 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
298
299                 if (goal < rsv->rsv_start)
300                         n = n->rb_left;
301                 else if (goal > rsv->rsv_end)
302                         n = n->rb_right;
303                 else
304                         return rsv;
305         } while (n);
306         /*
307          * We've fallen off the end of the tree: the goal wasn't inside
308          * any particular node.  OK, the previous node must be to one
309          * side of the interval containing the goal.  If it's the RHS,
310          * we need to back up one.
311          */
312         if (rsv->rsv_start > goal) {
313                 n = rb_prev(&rsv->rsv_node);
314                 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
315         }
316         return rsv;
317 }
318
319 /**
320  * ext3_rsv_window_add() -- Insert a window to the block reservation rb tree.
321  * @sb:                 super block
322  * @rsv:                reservation window to add
323  *
324  * Must be called with rsv_lock hold.
325  */
326 void ext3_rsv_window_add(struct super_block *sb,
327                     struct ext3_reserve_window_node *rsv)
328 {
329         struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
330         struct rb_node *node = &rsv->rsv_node;
331         ext3_fsblk_t start = rsv->rsv_start;
332
333         struct rb_node ** p = &root->rb_node;
334         struct rb_node * parent = NULL;
335         struct ext3_reserve_window_node *this;
336
337         while (*p)
338         {
339                 parent = *p;
340                 this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
341
342                 if (start < this->rsv_start)
343                         p = &(*p)->rb_left;
344                 else if (start > this->rsv_end)
345                         p = &(*p)->rb_right;
346                 else {
347                         rsv_window_dump(root, 1);
348                         BUG();
349                 }
350         }
351
352         rb_link_node(node, parent, p);
353         rb_insert_color(node, root);
354 }
355
356 /**
357  * ext3_rsv_window_remove() -- unlink a window from the reservation rb tree
358  * @sb:                 super block
359  * @rsv:                reservation window to remove
360  *
361  * Mark the block reservation window as not allocated, and unlink it
362  * from the filesystem reservation window rb tree. Must be called with
363  * rsv_lock hold.
364  */
365 static void rsv_window_remove(struct super_block *sb,
366                               struct ext3_reserve_window_node *rsv)
367 {
368         rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
369         rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
370         rsv->rsv_alloc_hit = 0;
371         rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
372 }
373
374 /*
375  * rsv_is_empty() -- Check if the reservation window is allocated.
376  * @rsv:                given reservation window to check
377  *
378  * returns 1 if the end block is EXT3_RESERVE_WINDOW_NOT_ALLOCATED.
379  */
380 static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
381 {
382         /* a valid reservation end block could not be 0 */
383         return rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
384 }
385
386 /**
387  * ext3_init_block_alloc_info()
388  * @inode:              file inode structure
389  *
390  * Allocate and initialize the  reservation window structure, and
391  * link the window to the ext3 inode structure at last
392  *
393  * The reservation window structure is only dynamically allocated
394  * and linked to ext3 inode the first time the open file
395  * needs a new block. So, before every ext3_new_block(s) call, for
396  * regular files, we should check whether the reservation window
397  * structure exists or not. In the latter case, this function is called.
398  * Fail to do so will result in block reservation being turned off for that
399  * open file.
400  *
401  * This function is called from ext3_get_blocks_handle(), also called
402  * when setting the reservation window size through ioctl before the file
403  * is open for write (needs block allocation).
404  *
405  * Needs truncate_mutex protection prior to call this function.
406  */
407 void ext3_init_block_alloc_info(struct inode *inode)
408 {
409         struct ext3_inode_info *ei = EXT3_I(inode);
410         struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
411         struct super_block *sb = inode->i_sb;
412
413         block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
414         if (block_i) {
415                 struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node;
416
417                 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
418                 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
419
420                 /*
421                  * if filesystem is mounted with NORESERVATION, the goal
422                  * reservation window size is set to zero to indicate
423                  * block reservation is off
424                  */
425                 if (!test_opt(sb, RESERVATION))
426                         rsv->rsv_goal_size = 0;
427                 else
428                         rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS;
429                 rsv->rsv_alloc_hit = 0;
430                 block_i->last_alloc_logical_block = 0;
431                 block_i->last_alloc_physical_block = 0;
432         }
433         ei->i_block_alloc_info = block_i;
434 }
435
436 /**
437  * ext3_discard_reservation()
438  * @inode:              inode
439  *
440  * Discard(free) block reservation window on last file close, or truncate
441  * or at last iput().
442  *
443  * It is being called in three cases:
444  *      ext3_release_file(): last writer close the file
445  *      ext3_clear_inode(): last iput(), when nobody link to this file.
446  *      ext3_truncate(): when the block indirect map is about to change.
447  *
448  */
449 void ext3_discard_reservation(struct inode *inode)
450 {
451         struct ext3_inode_info *ei = EXT3_I(inode);
452         struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
453         struct ext3_reserve_window_node *rsv;
454         spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
455
456         if (!block_i)
457                 return;
458
459         rsv = &block_i->rsv_window_node;
460         if (!rsv_is_empty(&rsv->rsv_window)) {
461                 spin_lock(rsv_lock);
462                 if (!rsv_is_empty(&rsv->rsv_window))
463                         rsv_window_remove(inode->i_sb, rsv);
464                 spin_unlock(rsv_lock);
465         }
466 }
467
468 /**
469  * ext3_free_blocks_sb() -- Free given blocks and update quota
470  * @handle:                     handle to this transaction
471  * @sb:                         super block
472  * @block:                      start physcial block to free
473  * @count:                      number of blocks to free
474  * @pdquot_freed_blocks:        pointer to quota
475  */
476 void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
477                          ext3_fsblk_t block, unsigned long count,
478                          unsigned long *pdquot_freed_blocks)
479 {
480         struct buffer_head *bitmap_bh = NULL;
481         struct buffer_head *gd_bh;
482         unsigned long block_group;
483         ext3_grpblk_t bit;
484         unsigned long i;
485         unsigned long overflow;
486         struct ext3_group_desc * desc;
487         struct ext3_super_block * es;
488         struct ext3_sb_info *sbi;
489         int err = 0, ret;
490         ext3_grpblk_t group_freed;
491
492         *pdquot_freed_blocks = 0;
493         sbi = EXT3_SB(sb);
494         es = sbi->s_es;
495         if (block < le32_to_cpu(es->s_first_data_block) ||
496             block + count < block ||
497             block + count > le32_to_cpu(es->s_blocks_count)) {
498                 ext3_error (sb, "ext3_free_blocks",
499                             "Freeing blocks not in datazone - "
500                             "block = "E3FSBLK", count = %lu", block, count);
501                 goto error_return;
502         }
503
504         ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
505
506 do_more:
507         overflow = 0;
508         block_group = (block - le32_to_cpu(es->s_first_data_block)) /
509                       EXT3_BLOCKS_PER_GROUP(sb);
510         bit = (block - le32_to_cpu(es->s_first_data_block)) %
511                       EXT3_BLOCKS_PER_GROUP(sb);
512         /*
513          * Check to see if we are freeing blocks across a group
514          * boundary.
515          */
516         if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
517                 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
518                 count -= overflow;
519         }
520         brelse(bitmap_bh);
521         bitmap_bh = read_block_bitmap(sb, block_group);
522         if (!bitmap_bh)
523                 goto error_return;
524         desc = ext3_get_group_desc (sb, block_group, &gd_bh);
525         if (!desc)
526                 goto error_return;
527
528         if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
529             in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
530             in_range (block, le32_to_cpu(desc->bg_inode_table),
531                       sbi->s_itb_per_group) ||
532             in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
533                       sbi->s_itb_per_group)) {
534                 ext3_error (sb, "ext3_free_blocks",
535                             "Freeing blocks in system zones - "
536                             "Block = "E3FSBLK", count = %lu",
537                             block, count);
538                 goto error_return;
539         }
540
541         /*
542          * We are about to start releasing blocks in the bitmap,
543          * so we need undo access.
544          */
545         /* @@@ check errors */
546         BUFFER_TRACE(bitmap_bh, "getting undo access");
547         err = ext3_journal_get_undo_access(handle, bitmap_bh);
548         if (err)
549                 goto error_return;
550
551         /*
552          * We are about to modify some metadata.  Call the journal APIs
553          * to unshare ->b_data if a currently-committing transaction is
554          * using it
555          */
556         BUFFER_TRACE(gd_bh, "get_write_access");
557         err = ext3_journal_get_write_access(handle, gd_bh);
558         if (err)
559                 goto error_return;
560
561         jbd_lock_bh_state(bitmap_bh);
562
563         for (i = 0, group_freed = 0; i < count; i++) {
564                 /*
565                  * An HJ special.  This is expensive...
566                  */
567 #ifdef CONFIG_JBD_DEBUG
568                 jbd_unlock_bh_state(bitmap_bh);
569                 {
570                         struct buffer_head *debug_bh;
571                         debug_bh = sb_find_get_block(sb, block + i);
572                         if (debug_bh) {
573                                 BUFFER_TRACE(debug_bh, "Deleted!");
574                                 if (!bh2jh(bitmap_bh)->b_committed_data)
575                                         BUFFER_TRACE(debug_bh,
576                                                 "No commited data in bitmap");
577                                 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
578                                 __brelse(debug_bh);
579                         }
580                 }
581                 jbd_lock_bh_state(bitmap_bh);
582 #endif
583                 if (need_resched()) {
584                         jbd_unlock_bh_state(bitmap_bh);
585                         cond_resched();
586                         jbd_lock_bh_state(bitmap_bh);
587                 }
588                 /* @@@ This prevents newly-allocated data from being
589                  * freed and then reallocated within the same
590                  * transaction.
591                  *
592                  * Ideally we would want to allow that to happen, but to
593                  * do so requires making journal_forget() capable of
594                  * revoking the queued write of a data block, which
595                  * implies blocking on the journal lock.  *forget()
596                  * cannot block due to truncate races.
597                  *
598                  * Eventually we can fix this by making journal_forget()
599                  * return a status indicating whether or not it was able
600                  * to revoke the buffer.  On successful revoke, it is
601                  * safe not to set the allocation bit in the committed
602                  * bitmap, because we know that there is no outstanding
603                  * activity on the buffer any more and so it is safe to
604                  * reallocate it.
605                  */
606                 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
607                 J_ASSERT_BH(bitmap_bh,
608                                 bh2jh(bitmap_bh)->b_committed_data != NULL);
609                 ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
610                                 bh2jh(bitmap_bh)->b_committed_data);
611
612                 /*
613                  * We clear the bit in the bitmap after setting the committed
614                  * data bit, because this is the reverse order to that which
615                  * the allocator uses.
616                  */
617                 BUFFER_TRACE(bitmap_bh, "clear bit");
618                 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
619                                                 bit + i, bitmap_bh->b_data)) {
620                         jbd_unlock_bh_state(bitmap_bh);
621                         ext3_error(sb, __func__,
622                                 "bit already cleared for block "E3FSBLK,
623                                  block + i);
624                         jbd_lock_bh_state(bitmap_bh);
625                         BUFFER_TRACE(bitmap_bh, "bit already cleared");
626                 } else {
627                         group_freed++;
628                 }
629         }
630         jbd_unlock_bh_state(bitmap_bh);
631
632         spin_lock(sb_bgl_lock(sbi, block_group));
633         le16_add_cpu(&desc->bg_free_blocks_count, group_freed);
634         spin_unlock(sb_bgl_lock(sbi, block_group));
635         percpu_counter_add(&sbi->s_freeblocks_counter, count);
636
637         /* We dirtied the bitmap block */
638         BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
639         err = ext3_journal_dirty_metadata(handle, bitmap_bh);
640
641         /* And the group descriptor block */
642         BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
643         ret = ext3_journal_dirty_metadata(handle, gd_bh);
644         if (!err) err = ret;
645         *pdquot_freed_blocks += group_freed;
646
647         if (overflow && !err) {
648                 block += count;
649                 count = overflow;
650                 goto do_more;
651         }
652         sb->s_dirt = 1;
653 error_return:
654         brelse(bitmap_bh);
655         ext3_std_error(sb, err);
656         return;
657 }
658
659 /**
660  * ext3_free_blocks() -- Free given blocks and update quota
661  * @handle:             handle for this transaction
662  * @inode:              inode
663  * @block:              start physical block to free
664  * @count:              number of blocks to count
665  */
666 void ext3_free_blocks(handle_t *handle, struct inode *inode,
667                         ext3_fsblk_t block, unsigned long count)
668 {
669         struct super_block * sb;
670         unsigned long dquot_freed_blocks;
671
672         sb = inode->i_sb;
673         if (!sb) {
674                 printk ("ext3_free_blocks: nonexistent device");
675                 return;
676         }
677         ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
678         if (dquot_freed_blocks)
679                 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
680         return;
681 }
682
683 /**
684  * ext3_test_allocatable()
685  * @nr:                 given allocation block group
686  * @bh:                 bufferhead contains the bitmap of the given block group
687  *
688  * For ext3 allocations, we must not reuse any blocks which are
689  * allocated in the bitmap buffer's "last committed data" copy.  This
690  * prevents deletes from freeing up the page for reuse until we have
691  * committed the delete transaction.
692  *
693  * If we didn't do this, then deleting something and reallocating it as
694  * data would allow the old block to be overwritten before the
695  * transaction committed (because we force data to disk before commit).
696  * This would lead to corruption if we crashed between overwriting the
697  * data and committing the delete.
698  *
699  * @@@ We may want to make this allocation behaviour conditional on
700  * data-writes at some point, and disable it for metadata allocations or
701  * sync-data inodes.
702  */
703 static int ext3_test_allocatable(ext3_grpblk_t nr, struct buffer_head *bh)
704 {
705         int ret;
706         struct journal_head *jh = bh2jh(bh);
707
708         if (ext3_test_bit(nr, bh->b_data))
709                 return 0;
710
711         jbd_lock_bh_state(bh);
712         if (!jh->b_committed_data)
713                 ret = 1;
714         else
715                 ret = !ext3_test_bit(nr, jh->b_committed_data);
716         jbd_unlock_bh_state(bh);
717         return ret;
718 }
719
720 /**
721  * bitmap_search_next_usable_block()
722  * @start:              the starting block (group relative) of the search
723  * @bh:                 bufferhead contains the block group bitmap
724  * @maxblocks:          the ending block (group relative) of the reservation
725  *
726  * The bitmap search --- search forward alternately through the actual
727  * bitmap on disk and the last-committed copy in journal, until we find a
728  * bit free in both bitmaps.
729  */
730 static ext3_grpblk_t
731 bitmap_search_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
732                                         ext3_grpblk_t maxblocks)
733 {
734         ext3_grpblk_t next;
735         struct journal_head *jh = bh2jh(bh);
736
737         while (start < maxblocks) {
738                 next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
739                 if (next >= maxblocks)
740                         return -1;
741                 if (ext3_test_allocatable(next, bh))
742                         return next;
743                 jbd_lock_bh_state(bh);
744                 if (jh->b_committed_data)
745                         start = ext3_find_next_zero_bit(jh->b_committed_data,
746                                                         maxblocks, next);
747                 jbd_unlock_bh_state(bh);
748         }
749         return -1;
750 }
751
752 /**
753  * find_next_usable_block()
754  * @start:              the starting block (group relative) to find next
755  *                      allocatable block in bitmap.
756  * @bh:                 bufferhead contains the block group bitmap
757  * @maxblocks:          the ending block (group relative) for the search
758  *
759  * Find an allocatable block in a bitmap.  We honor both the bitmap and
760  * its last-committed copy (if that exists), and perform the "most
761  * appropriate allocation" algorithm of looking for a free block near
762  * the initial goal; then for a free byte somewhere in the bitmap; then
763  * for any free bit in the bitmap.
764  */
765 static ext3_grpblk_t
766 find_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
767                         ext3_grpblk_t maxblocks)
768 {
769         ext3_grpblk_t here, next;
770         char *p, *r;
771
772         if (start > 0) {
773                 /*
774                  * The goal was occupied; search forward for a free
775                  * block within the next XX blocks.
776                  *
777                  * end_goal is more or less random, but it has to be
778                  * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
779                  * next 64-bit boundary is simple..
780                  */
781                 ext3_grpblk_t end_goal = (start + 63) & ~63;
782                 if (end_goal > maxblocks)
783                         end_goal = maxblocks;
784                 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
785                 if (here < end_goal && ext3_test_allocatable(here, bh))
786                         return here;
787                 ext3_debug("Bit not found near goal\n");
788         }
789
790         here = start;
791         if (here < 0)
792                 here = 0;
793
794         p = ((char *)bh->b_data) + (here >> 3);
795         r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
796         next = (r - ((char *)bh->b_data)) << 3;
797
798         if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
799                 return next;
800
801         /*
802          * The bitmap search --- search forward alternately through the actual
803          * bitmap and the last-committed copy until we find a bit free in
804          * both
805          */
806         here = bitmap_search_next_usable_block(here, bh, maxblocks);
807         return here;
808 }
809
810 /**
811  * claim_block()
812  * @block:              the free block (group relative) to allocate
813  * @bh:                 the bufferhead containts the block group bitmap
814  *
815  * We think we can allocate this block in this bitmap.  Try to set the bit.
816  * If that succeeds then check that nobody has allocated and then freed the
817  * block since we saw that is was not marked in b_committed_data.  If it _was_
818  * allocated and freed then clear the bit in the bitmap again and return
819  * zero (failure).
820  */
821 static inline int
822 claim_block(spinlock_t *lock, ext3_grpblk_t block, struct buffer_head *bh)
823 {
824         struct journal_head *jh = bh2jh(bh);
825         int ret;
826
827         if (ext3_set_bit_atomic(lock, block, bh->b_data))
828                 return 0;
829         jbd_lock_bh_state(bh);
830         if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
831                 ext3_clear_bit_atomic(lock, block, bh->b_data);
832                 ret = 0;
833         } else {
834                 ret = 1;
835         }
836         jbd_unlock_bh_state(bh);
837         return ret;
838 }
839
840 /**
841  * ext3_try_to_allocate()
842  * @sb:                 superblock
843  * @handle:             handle to this transaction
844  * @group:              given allocation block group
845  * @bitmap_bh:          bufferhead holds the block bitmap
846  * @grp_goal:           given target block within the group
847  * @count:              target number of blocks to allocate
848  * @my_rsv:             reservation window
849  *
850  * Attempt to allocate blocks within a give range. Set the range of allocation
851  * first, then find the first free bit(s) from the bitmap (within the range),
852  * and at last, allocate the blocks by claiming the found free bit as allocated.
853  *
854  * To set the range of this allocation:
855  *      if there is a reservation window, only try to allocate block(s) from the
856  *      file's own reservation window;
857  *      Otherwise, the allocation range starts from the give goal block, ends at
858  *      the block group's last block.
859  *
860  * If we failed to allocate the desired block then we may end up crossing to a
861  * new bitmap.  In that case we must release write access to the old one via
862  * ext3_journal_release_buffer(), else we'll run out of credits.
863  */
864 static ext3_grpblk_t
865 ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
866                         struct buffer_head *bitmap_bh, ext3_grpblk_t grp_goal,
867                         unsigned long *count, struct ext3_reserve_window *my_rsv)
868 {
869         ext3_fsblk_t group_first_block;
870         ext3_grpblk_t start, end;
871         unsigned long num = 0;
872
873         /* we do allocation within the reservation window if we have a window */
874         if (my_rsv) {
875                 group_first_block = ext3_group_first_block_no(sb, group);
876                 if (my_rsv->_rsv_start >= group_first_block)
877                         start = my_rsv->_rsv_start - group_first_block;
878                 else
879                         /* reservation window cross group boundary */
880                         start = 0;
881                 end = my_rsv->_rsv_end - group_first_block + 1;
882                 if (end > EXT3_BLOCKS_PER_GROUP(sb))
883                         /* reservation window crosses group boundary */
884                         end = EXT3_BLOCKS_PER_GROUP(sb);
885                 if ((start <= grp_goal) && (grp_goal < end))
886                         start = grp_goal;
887                 else
888                         grp_goal = -1;
889         } else {
890                 if (grp_goal > 0)
891                         start = grp_goal;
892                 else
893                         start = 0;
894                 end = EXT3_BLOCKS_PER_GROUP(sb);
895         }
896
897         BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
898
899 repeat:
900         if (grp_goal < 0 || !ext3_test_allocatable(grp_goal, bitmap_bh)) {
901                 grp_goal = find_next_usable_block(start, bitmap_bh, end);
902                 if (grp_goal < 0)
903                         goto fail_access;
904                 if (!my_rsv) {
905                         int i;
906
907                         for (i = 0; i < 7 && grp_goal > start &&
908                                         ext3_test_allocatable(grp_goal - 1,
909                                                                 bitmap_bh);
910                                         i++, grp_goal--)
911                                 ;
912                 }
913         }
914         start = grp_goal;
915
916         if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group),
917                 grp_goal, bitmap_bh)) {
918                 /*
919                  * The block was allocated by another thread, or it was
920                  * allocated and then freed by another thread
921                  */
922                 start++;
923                 grp_goal++;
924                 if (start >= end)
925                         goto fail_access;
926                 goto repeat;
927         }
928         num++;
929         grp_goal++;
930         while (num < *count && grp_goal < end
931                 && ext3_test_allocatable(grp_goal, bitmap_bh)
932                 && claim_block(sb_bgl_lock(EXT3_SB(sb), group),
933                                 grp_goal, bitmap_bh)) {
934                 num++;
935                 grp_goal++;
936         }
937         *count = num;
938         return grp_goal - num;
939 fail_access:
940         *count = num;
941         return -1;
942 }
943
944 /**
945  *      find_next_reservable_window():
946  *              find a reservable space within the given range.
947  *              It does not allocate the reservation window for now:
948  *              alloc_new_reservation() will do the work later.
949  *
950  *      @search_head: the head of the searching list;
951  *              This is not necessarily the list head of the whole filesystem
952  *
953  *              We have both head and start_block to assist the search
954  *              for the reservable space. The list starts from head,
955  *              but we will shift to the place where start_block is,
956  *              then start from there, when looking for a reservable space.
957  *
958  *      @size: the target new reservation window size
959  *
960  *      @group_first_block: the first block we consider to start
961  *                      the real search from
962  *
963  *      @last_block:
964  *              the maximum block number that our goal reservable space
965  *              could start from. This is normally the last block in this
966  *              group. The search will end when we found the start of next
967  *              possible reservable space is out of this boundary.
968  *              This could handle the cross boundary reservation window
969  *              request.
970  *
971  *      basically we search from the given range, rather than the whole
972  *      reservation double linked list, (start_block, last_block)
973  *      to find a free region that is of my size and has not
974  *      been reserved.
975  *
976  */
977 static int find_next_reservable_window(
978                                 struct ext3_reserve_window_node *search_head,
979                                 struct ext3_reserve_window_node *my_rsv,
980                                 struct super_block * sb,
981                                 ext3_fsblk_t start_block,
982                                 ext3_fsblk_t last_block)
983 {
984         struct rb_node *next;
985         struct ext3_reserve_window_node *rsv, *prev;
986         ext3_fsblk_t cur;
987         int size = my_rsv->rsv_goal_size;
988
989         /* TODO: make the start of the reservation window byte-aligned */
990         /* cur = *start_block & ~7;*/
991         cur = start_block;
992         rsv = search_head;
993         if (!rsv)
994                 return -1;
995
996         while (1) {
997                 if (cur <= rsv->rsv_end)
998                         cur = rsv->rsv_end + 1;
999
1000                 /* TODO?
1001                  * in the case we could not find a reservable space
1002                  * that is what is expected, during the re-search, we could
1003                  * remember what's the largest reservable space we could have
1004                  * and return that one.
1005                  *
1006                  * For now it will fail if we could not find the reservable
1007                  * space with expected-size (or more)...
1008                  */
1009                 if (cur > last_block)
1010                         return -1;              /* fail */
1011
1012                 prev = rsv;
1013                 next = rb_next(&rsv->rsv_node);
1014                 rsv = rb_entry(next,struct ext3_reserve_window_node,rsv_node);
1015
1016                 /*
1017                  * Reached the last reservation, we can just append to the
1018                  * previous one.
1019                  */
1020                 if (!next)
1021                         break;
1022
1023                 if (cur + size <= rsv->rsv_start) {
1024                         /*
1025                          * Found a reserveable space big enough.  We could
1026                          * have a reservation across the group boundary here
1027                          */
1028                         break;
1029                 }
1030         }
1031         /*
1032          * we come here either :
1033          * when we reach the end of the whole list,
1034          * and there is empty reservable space after last entry in the list.
1035          * append it to the end of the list.
1036          *
1037          * or we found one reservable space in the middle of the list,
1038          * return the reservation window that we could append to.
1039          * succeed.
1040          */
1041
1042         if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
1043                 rsv_window_remove(sb, my_rsv);
1044
1045         /*
1046          * Let's book the whole avaliable window for now.  We will check the
1047          * disk bitmap later and then, if there are free blocks then we adjust
1048          * the window size if it's larger than requested.
1049          * Otherwise, we will remove this node from the tree next time
1050          * call find_next_reservable_window.
1051          */
1052         my_rsv->rsv_start = cur;
1053         my_rsv->rsv_end = cur + size - 1;
1054         my_rsv->rsv_alloc_hit = 0;
1055
1056         if (prev != my_rsv)
1057                 ext3_rsv_window_add(sb, my_rsv);
1058
1059         return 0;
1060 }
1061
1062 /**
1063  *      alloc_new_reservation()--allocate a new reservation window
1064  *
1065  *              To make a new reservation, we search part of the filesystem
1066  *              reservation list (the list that inside the group). We try to
1067  *              allocate a new reservation window near the allocation goal,
1068  *              or the beginning of the group, if there is no goal.
1069  *
1070  *              We first find a reservable space after the goal, then from
1071  *              there, we check the bitmap for the first free block after
1072  *              it. If there is no free block until the end of group, then the
1073  *              whole group is full, we failed. Otherwise, check if the free
1074  *              block is inside the expected reservable space, if so, we
1075  *              succeed.
1076  *              If the first free block is outside the reservable space, then
1077  *              start from the first free block, we search for next available
1078  *              space, and go on.
1079  *
1080  *      on succeed, a new reservation will be found and inserted into the list
1081  *      It contains at least one free block, and it does not overlap with other
1082  *      reservation windows.
1083  *
1084  *      failed: we failed to find a reservation window in this group
1085  *
1086  *      @rsv: the reservation
1087  *
1088  *      @grp_goal: The goal (group-relative).  It is where the search for a
1089  *              free reservable space should start from.
1090  *              if we have a grp_goal(grp_goal >0 ), then start from there,
1091  *              no grp_goal(grp_goal = -1), we start from the first block
1092  *              of the group.
1093  *
1094  *      @sb: the super block
1095  *      @group: the group we are trying to allocate in
1096  *      @bitmap_bh: the block group block bitmap
1097  *
1098  */
1099 static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
1100                 ext3_grpblk_t grp_goal, struct super_block *sb,
1101                 unsigned int group, struct buffer_head *bitmap_bh)
1102 {
1103         struct ext3_reserve_window_node *search_head;
1104         ext3_fsblk_t group_first_block, group_end_block, start_block;
1105         ext3_grpblk_t first_free_block;
1106         struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
1107         unsigned long size;
1108         int ret;
1109         spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1110
1111         group_first_block = ext3_group_first_block_no(sb, group);
1112         group_end_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1113
1114         if (grp_goal < 0)
1115                 start_block = group_first_block;
1116         else
1117                 start_block = grp_goal + group_first_block;
1118
1119         size = my_rsv->rsv_goal_size;
1120
1121         if (!rsv_is_empty(&my_rsv->rsv_window)) {
1122                 /*
1123                  * if the old reservation is cross group boundary
1124                  * and if the goal is inside the old reservation window,
1125                  * we will come here when we just failed to allocate from
1126                  * the first part of the window. We still have another part
1127                  * that belongs to the next group. In this case, there is no
1128                  * point to discard our window and try to allocate a new one
1129                  * in this group(which will fail). we should
1130                  * keep the reservation window, just simply move on.
1131                  *
1132                  * Maybe we could shift the start block of the reservation
1133                  * window to the first block of next group.
1134                  */
1135
1136                 if ((my_rsv->rsv_start <= group_end_block) &&
1137                                 (my_rsv->rsv_end > group_end_block) &&
1138                                 (start_block >= my_rsv->rsv_start))
1139                         return -1;
1140
1141                 if ((my_rsv->rsv_alloc_hit >
1142                      (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
1143                         /*
1144                          * if the previously allocation hit ratio is
1145                          * greater than 1/2, then we double the size of
1146                          * the reservation window the next time,
1147                          * otherwise we keep the same size window
1148                          */
1149                         size = size * 2;
1150                         if (size > EXT3_MAX_RESERVE_BLOCKS)
1151                                 size = EXT3_MAX_RESERVE_BLOCKS;
1152                         my_rsv->rsv_goal_size= size;
1153                 }
1154         }
1155
1156         spin_lock(rsv_lock);
1157         /*
1158          * shift the search start to the window near the goal block
1159          */
1160         search_head = search_reserve_window(fs_rsv_root, start_block);
1161
1162         /*
1163          * find_next_reservable_window() simply finds a reservable window
1164          * inside the given range(start_block, group_end_block).
1165          *
1166          * To make sure the reservation window has a free bit inside it, we
1167          * need to check the bitmap after we found a reservable window.
1168          */
1169 retry:
1170         ret = find_next_reservable_window(search_head, my_rsv, sb,
1171                                                 start_block, group_end_block);
1172
1173         if (ret == -1) {
1174                 if (!rsv_is_empty(&my_rsv->rsv_window))
1175                         rsv_window_remove(sb, my_rsv);
1176                 spin_unlock(rsv_lock);
1177                 return -1;
1178         }
1179
1180         /*
1181          * On success, find_next_reservable_window() returns the
1182          * reservation window where there is a reservable space after it.
1183          * Before we reserve this reservable space, we need
1184          * to make sure there is at least a free block inside this region.
1185          *
1186          * searching the first free bit on the block bitmap and copy of
1187          * last committed bitmap alternatively, until we found a allocatable
1188          * block. Search start from the start block of the reservable space
1189          * we just found.
1190          */
1191         spin_unlock(rsv_lock);
1192         first_free_block = bitmap_search_next_usable_block(
1193                         my_rsv->rsv_start - group_first_block,
1194                         bitmap_bh, group_end_block - group_first_block + 1);
1195
1196         if (first_free_block < 0) {
1197                 /*
1198                  * no free block left on the bitmap, no point
1199                  * to reserve the space. return failed.
1200                  */
1201                 spin_lock(rsv_lock);
1202                 if (!rsv_is_empty(&my_rsv->rsv_window))
1203                         rsv_window_remove(sb, my_rsv);
1204                 spin_unlock(rsv_lock);
1205                 return -1;              /* failed */
1206         }
1207
1208         start_block = first_free_block + group_first_block;
1209         /*
1210          * check if the first free block is within the
1211          * free space we just reserved
1212          */
1213         if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1214                 return 0;               /* success */
1215         /*
1216          * if the first free bit we found is out of the reservable space
1217          * continue search for next reservable space,
1218          * start from where the free block is,
1219          * we also shift the list head to where we stopped last time
1220          */
1221         search_head = my_rsv;
1222         spin_lock(rsv_lock);
1223         goto retry;
1224 }
1225
1226 /**
1227  * try_to_extend_reservation()
1228  * @my_rsv:             given reservation window
1229  * @sb:                 super block
1230  * @size:               the delta to extend
1231  *
1232  * Attempt to expand the reservation window large enough to have
1233  * required number of free blocks
1234  *
1235  * Since ext3_try_to_allocate() will always allocate blocks within
1236  * the reservation window range, if the window size is too small,
1237  * multiple blocks allocation has to stop at the end of the reservation
1238  * window. To make this more efficient, given the total number of
1239  * blocks needed and the current size of the window, we try to
1240  * expand the reservation window size if necessary on a best-effort
1241  * basis before ext3_new_blocks() tries to allocate blocks,
1242  */
1243 static void try_to_extend_reservation(struct ext3_reserve_window_node *my_rsv,
1244                         struct super_block *sb, int size)
1245 {
1246         struct ext3_reserve_window_node *next_rsv;
1247         struct rb_node *next;
1248         spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1249
1250         if (!spin_trylock(rsv_lock))
1251                 return;
1252
1253         next = rb_next(&my_rsv->rsv_node);
1254
1255         if (!next)
1256                 my_rsv->rsv_end += size;
1257         else {
1258                 next_rsv = rb_entry(next, struct ext3_reserve_window_node, rsv_node);
1259
1260                 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1261                         my_rsv->rsv_end += size;
1262                 else
1263                         my_rsv->rsv_end = next_rsv->rsv_start - 1;
1264         }
1265         spin_unlock(rsv_lock);
1266 }
1267
1268 /**
1269  * ext3_try_to_allocate_with_rsv()
1270  * @sb:                 superblock
1271  * @handle:             handle to this transaction
1272  * @group:              given allocation block group
1273  * @bitmap_bh:          bufferhead holds the block bitmap
1274  * @grp_goal:           given target block within the group
1275  * @count:              target number of blocks to allocate
1276  * @my_rsv:             reservation window
1277  * @errp:               pointer to store the error code
1278  *
1279  * This is the main function used to allocate a new block and its reservation
1280  * window.
1281  *
1282  * Each time when a new block allocation is need, first try to allocate from
1283  * its own reservation.  If it does not have a reservation window, instead of
1284  * looking for a free bit on bitmap first, then look up the reservation list to
1285  * see if it is inside somebody else's reservation window, we try to allocate a
1286  * reservation window for it starting from the goal first. Then do the block
1287  * allocation within the reservation window.
1288  *
1289  * This will avoid keeping on searching the reservation list again and
1290  * again when somebody is looking for a free block (without
1291  * reservation), and there are lots of free blocks, but they are all
1292  * being reserved.
1293  *
1294  * We use a red-black tree for the per-filesystem reservation list.
1295  *
1296  */
1297 static ext3_grpblk_t
1298 ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1299                         unsigned int group, struct buffer_head *bitmap_bh,
1300                         ext3_grpblk_t grp_goal,
1301                         struct ext3_reserve_window_node * my_rsv,
1302                         unsigned long *count, int *errp)
1303 {
1304         ext3_fsblk_t group_first_block, group_last_block;
1305         ext3_grpblk_t ret = 0;
1306         int fatal;
1307         unsigned long num = *count;
1308
1309         *errp = 0;
1310
1311         /*
1312          * Make sure we use undo access for the bitmap, because it is critical
1313          * that we do the frozen_data COW on bitmap buffers in all cases even
1314          * if the buffer is in BJ_Forget state in the committing transaction.
1315          */
1316         BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1317         fatal = ext3_journal_get_undo_access(handle, bitmap_bh);
1318         if (fatal) {
1319                 *errp = fatal;
1320                 return -1;
1321         }
1322
1323         /*
1324          * we don't deal with reservation when
1325          * filesystem is mounted without reservation
1326          * or the file is not a regular file
1327          * or last attempt to allocate a block with reservation turned on failed
1328          */
1329         if (my_rsv == NULL ) {
1330                 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1331                                                 grp_goal, count, NULL);
1332                 goto out;
1333         }
1334         /*
1335          * grp_goal is a group relative block number (if there is a goal)
1336          * 0 <= grp_goal < EXT3_BLOCKS_PER_GROUP(sb)
1337          * first block is a filesystem wide block number
1338          * first block is the block number of the first block in this group
1339          */
1340         group_first_block = ext3_group_first_block_no(sb, group);
1341         group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1342
1343         /*
1344          * Basically we will allocate a new block from inode's reservation
1345          * window.
1346          *
1347          * We need to allocate a new reservation window, if:
1348          * a) inode does not have a reservation window; or
1349          * b) last attempt to allocate a block from existing reservation
1350          *    failed; or
1351          * c) we come here with a goal and with a reservation window
1352          *
1353          * We do not need to allocate a new reservation window if we come here
1354          * at the beginning with a goal and the goal is inside the window, or
1355          * we don't have a goal but already have a reservation window.
1356          * then we could go to allocate from the reservation window directly.
1357          */
1358         while (1) {
1359                 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1360                         !goal_in_my_reservation(&my_rsv->rsv_window,
1361                                                 grp_goal, group, sb)) {
1362                         if (my_rsv->rsv_goal_size < *count)
1363                                 my_rsv->rsv_goal_size = *count;
1364                         ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1365                                                         group, bitmap_bh);
1366                         if (ret < 0)
1367                                 break;                  /* failed */
1368
1369                         if (!goal_in_my_reservation(&my_rsv->rsv_window,
1370                                                         grp_goal, group, sb))
1371                                 grp_goal = -1;
1372                 } else if (grp_goal >= 0) {
1373                         int curr = my_rsv->rsv_end -
1374                                         (grp_goal + group_first_block) + 1;
1375
1376                         if (curr < *count)
1377                                 try_to_extend_reservation(my_rsv, sb,
1378                                                         *count - curr);
1379                 }
1380
1381                 if ((my_rsv->rsv_start > group_last_block) ||
1382                                 (my_rsv->rsv_end < group_first_block)) {
1383                         rsv_window_dump(&EXT3_SB(sb)->s_rsv_window_root, 1);
1384                         BUG();
1385                 }
1386                 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1387                                            grp_goal, &num, &my_rsv->rsv_window);
1388                 if (ret >= 0) {
1389                         my_rsv->rsv_alloc_hit += num;
1390                         *count = num;
1391                         break;                          /* succeed */
1392                 }
1393                 num = *count;
1394         }
1395 out:
1396         if (ret >= 0) {
1397                 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1398                                         "bitmap block");
1399                 fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1400                 if (fatal) {
1401                         *errp = fatal;
1402                         return -1;
1403                 }
1404                 return ret;
1405         }
1406
1407         BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1408         ext3_journal_release_buffer(handle, bitmap_bh);
1409         return ret;
1410 }
1411
1412 /**
1413  * ext3_has_free_blocks()
1414  * @sbi:                in-core super block structure.
1415  *
1416  * Check if filesystem has at least 1 free block available for allocation.
1417  */
1418 static int ext3_has_free_blocks(struct ext3_sb_info *sbi)
1419 {
1420         ext3_fsblk_t free_blocks, root_blocks;
1421
1422         free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1423         root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1424         if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1425                 sbi->s_resuid != current_fsuid() &&
1426                 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1427                 return 0;
1428         }
1429         return 1;
1430 }
1431
1432 /**
1433  * ext3_should_retry_alloc()
1434  * @sb:                 super block
1435  * @retries             number of attemps has been made
1436  *
1437  * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1438  * it is profitable to retry the operation, this function will wait
1439  * for the current or commiting transaction to complete, and then
1440  * return TRUE.
1441  *
1442  * if the total number of retries exceed three times, return FALSE.
1443  */
1444 int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1445 {
1446         if (!ext3_has_free_blocks(EXT3_SB(sb)) || (*retries)++ > 3)
1447                 return 0;
1448
1449         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1450
1451         return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1452 }
1453
1454 /**
1455  * ext3_new_blocks() -- core block(s) allocation function
1456  * @handle:             handle to this transaction
1457  * @inode:              file inode
1458  * @goal:               given target block(filesystem wide)
1459  * @count:              target number of blocks to allocate
1460  * @errp:               error code
1461  *
1462  * ext3_new_blocks uses a goal block to assist allocation.  It tries to
1463  * allocate block(s) from the block group contains the goal block first. If that
1464  * fails, it will try to allocate block(s) from other block groups without
1465  * any specific goal block.
1466  *
1467  */
1468 ext3_fsblk_t ext3_new_blocks(handle_t *handle, struct inode *inode,
1469                         ext3_fsblk_t goal, unsigned long *count, int *errp)
1470 {
1471         struct buffer_head *bitmap_bh = NULL;
1472         struct buffer_head *gdp_bh;
1473         int group_no;
1474         int goal_group;
1475         ext3_grpblk_t grp_target_blk;   /* blockgroup relative goal block */
1476         ext3_grpblk_t grp_alloc_blk;    /* blockgroup-relative allocated block*/
1477         ext3_fsblk_t ret_block;         /* filesyetem-wide allocated block */
1478         int bgi;                        /* blockgroup iteration index */
1479         int fatal = 0, err;
1480         int performed_allocation = 0;
1481         ext3_grpblk_t free_blocks;      /* number of free blocks in a group */
1482         struct super_block *sb;
1483         struct ext3_group_desc *gdp;
1484         struct ext3_super_block *es;
1485         struct ext3_sb_info *sbi;
1486         struct ext3_reserve_window_node *my_rsv = NULL;
1487         struct ext3_block_alloc_info *block_i;
1488         unsigned short windowsz = 0;
1489 #ifdef EXT3FS_DEBUG
1490         static int goal_hits, goal_attempts;
1491 #endif
1492         unsigned long ngroups;
1493         unsigned long num = *count;
1494
1495         *errp = -ENOSPC;
1496         sb = inode->i_sb;
1497         if (!sb) {
1498                 printk("ext3_new_block: nonexistent device");
1499                 return 0;
1500         }
1501
1502         /*
1503          * Check quota for allocation of this block.
1504          */
1505         if (DQUOT_ALLOC_BLOCK(inode, num)) {
1506                 *errp = -EDQUOT;
1507                 return 0;
1508         }
1509
1510         sbi = EXT3_SB(sb);
1511         es = EXT3_SB(sb)->s_es;
1512         ext3_debug("goal=%lu.\n", goal);
1513         /*
1514          * Allocate a block from reservation only when
1515          * filesystem is mounted with reservation(default,-o reservation), and
1516          * it's a regular file, and
1517          * the desired window size is greater than 0 (One could use ioctl
1518          * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1519          * reservation on that particular file)
1520          */
1521         block_i = EXT3_I(inode)->i_block_alloc_info;
1522         if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1523                 my_rsv = &block_i->rsv_window_node;
1524
1525         if (!ext3_has_free_blocks(sbi)) {
1526                 *errp = -ENOSPC;
1527                 goto out;
1528         }
1529
1530         /*
1531          * First, test whether the goal block is free.
1532          */
1533         if (goal < le32_to_cpu(es->s_first_data_block) ||
1534             goal >= le32_to_cpu(es->s_blocks_count))
1535                 goal = le32_to_cpu(es->s_first_data_block);
1536         group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1537                         EXT3_BLOCKS_PER_GROUP(sb);
1538         goal_group = group_no;
1539 retry_alloc:
1540         gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1541         if (!gdp)
1542                 goto io_error;
1543
1544         free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1545         /*
1546          * if there is not enough free blocks to make a new resevation
1547          * turn off reservation for this allocation
1548          */
1549         if (my_rsv && (free_blocks < windowsz)
1550                 && (free_blocks > 0)
1551                 && (rsv_is_empty(&my_rsv->rsv_window)))
1552                 my_rsv = NULL;
1553
1554         if (free_blocks > 0) {
1555                 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1556                                 EXT3_BLOCKS_PER_GROUP(sb));
1557                 bitmap_bh = read_block_bitmap(sb, group_no);
1558                 if (!bitmap_bh)
1559                         goto io_error;
1560                 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1561                                         group_no, bitmap_bh, grp_target_blk,
1562                                         my_rsv, &num, &fatal);
1563                 if (fatal)
1564                         goto out;
1565                 if (grp_alloc_blk >= 0)
1566                         goto allocated;
1567         }
1568
1569         ngroups = EXT3_SB(sb)->s_groups_count;
1570         smp_rmb();
1571
1572         /*
1573          * Now search the rest of the groups.  We assume that
1574          * group_no and gdp correctly point to the last group visited.
1575          */
1576         for (bgi = 0; bgi < ngroups; bgi++) {
1577                 group_no++;
1578                 if (group_no >= ngroups)
1579                         group_no = 0;
1580                 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1581                 if (!gdp)
1582                         goto io_error;
1583                 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1584                 /*
1585                  * skip this group if the number of
1586                  * free blocks is less than half of the reservation
1587                  * window size.
1588                  */
1589                 if (my_rsv && (free_blocks <= (windowsz/2)))
1590                         continue;
1591
1592                 brelse(bitmap_bh);
1593                 bitmap_bh = read_block_bitmap(sb, group_no);
1594                 if (!bitmap_bh)
1595                         goto io_error;
1596                 /*
1597                  * try to allocate block(s) from this group, without a goal(-1).
1598                  */
1599                 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1600                                         group_no, bitmap_bh, -1, my_rsv,
1601                                         &num, &fatal);
1602                 if (fatal)
1603                         goto out;
1604                 if (grp_alloc_blk >= 0)
1605                         goto allocated;
1606         }
1607         /*
1608          * We may end up a bogus ealier ENOSPC error due to
1609          * filesystem is "full" of reservations, but
1610          * there maybe indeed free blocks avaliable on disk
1611          * In this case, we just forget about the reservations
1612          * just do block allocation as without reservations.
1613          */
1614         if (my_rsv) {
1615                 my_rsv = NULL;
1616                 windowsz = 0;
1617                 group_no = goal_group;
1618                 goto retry_alloc;
1619         }
1620         /* No space left on the device */
1621         *errp = -ENOSPC;
1622         goto out;
1623
1624 allocated:
1625
1626         ext3_debug("using block group %d(%d)\n",
1627                         group_no, gdp->bg_free_blocks_count);
1628
1629         BUFFER_TRACE(gdp_bh, "get_write_access");
1630         fatal = ext3_journal_get_write_access(handle, gdp_bh);
1631         if (fatal)
1632                 goto out;
1633
1634         ret_block = grp_alloc_blk + ext3_group_first_block_no(sb, group_no);
1635
1636         if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1637             in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1638             in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1639                       EXT3_SB(sb)->s_itb_per_group) ||
1640             in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1641                       EXT3_SB(sb)->s_itb_per_group)) {
1642                 ext3_error(sb, "ext3_new_block",
1643                             "Allocating block in system zone - "
1644                             "blocks from "E3FSBLK", length %lu",
1645                              ret_block, num);
1646                 /*
1647                  * claim_block() marked the blocks we allocated as in use. So we
1648                  * may want to selectively mark some of the blocks as free.
1649                  */
1650                 goto retry_alloc;
1651         }
1652
1653         performed_allocation = 1;
1654
1655 #ifdef CONFIG_JBD_DEBUG
1656         {
1657                 struct buffer_head *debug_bh;
1658
1659                 /* Record bitmap buffer state in the newly allocated block */
1660                 debug_bh = sb_find_get_block(sb, ret_block);
1661                 if (debug_bh) {
1662                         BUFFER_TRACE(debug_bh, "state when allocated");
1663                         BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1664                         brelse(debug_bh);
1665                 }
1666         }
1667         jbd_lock_bh_state(bitmap_bh);
1668         spin_lock(sb_bgl_lock(sbi, group_no));
1669         if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1670                 int i;
1671
1672                 for (i = 0; i < num; i++) {
1673                         if (ext3_test_bit(grp_alloc_blk+i,
1674                                         bh2jh(bitmap_bh)->b_committed_data)) {
1675                                 printk("%s: block was unexpectedly set in "
1676                                         "b_committed_data\n", __func__);
1677                         }
1678                 }
1679         }
1680         ext3_debug("found bit %d\n", grp_alloc_blk);
1681         spin_unlock(sb_bgl_lock(sbi, group_no));
1682         jbd_unlock_bh_state(bitmap_bh);
1683 #endif
1684
1685         if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1686                 ext3_error(sb, "ext3_new_block",
1687                             "block("E3FSBLK") >= blocks count(%d) - "
1688                             "block_group = %d, es == %p ", ret_block,
1689                         le32_to_cpu(es->s_blocks_count), group_no, es);
1690                 goto out;
1691         }
1692
1693         /*
1694          * It is up to the caller to add the new buffer to a journal
1695          * list of some description.  We don't know in advance whether
1696          * the caller wants to use it as metadata or data.
1697          */
1698         ext3_debug("allocating block %lu. Goal hits %d of %d.\n",
1699                         ret_block, goal_hits, goal_attempts);
1700
1701         spin_lock(sb_bgl_lock(sbi, group_no));
1702         le16_add_cpu(&gdp->bg_free_blocks_count, -num);
1703         spin_unlock(sb_bgl_lock(sbi, group_no));
1704         percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1705
1706         BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1707         err = ext3_journal_dirty_metadata(handle, gdp_bh);
1708         if (!fatal)
1709                 fatal = err;
1710
1711         sb->s_dirt = 1;
1712         if (fatal)
1713                 goto out;
1714
1715         *errp = 0;
1716         brelse(bitmap_bh);
1717         DQUOT_FREE_BLOCK(inode, *count-num);
1718         *count = num;
1719         return ret_block;
1720
1721 io_error:
1722         *errp = -EIO;
1723 out:
1724         if (fatal) {
1725                 *errp = fatal;
1726                 ext3_std_error(sb, fatal);
1727         }
1728         /*
1729          * Undo the block allocation
1730          */
1731         if (!performed_allocation)
1732                 DQUOT_FREE_BLOCK(inode, *count);
1733         brelse(bitmap_bh);
1734         return 0;
1735 }
1736
1737 ext3_fsblk_t ext3_new_block(handle_t *handle, struct inode *inode,
1738                         ext3_fsblk_t goal, int *errp)
1739 {
1740         unsigned long count = 1;
1741
1742         return ext3_new_blocks(handle, inode, goal, &count, errp);
1743 }
1744
1745 /**
1746  * ext3_count_free_blocks() -- count filesystem free blocks
1747  * @sb:         superblock
1748  *
1749  * Adds up the number of free blocks from each block group.
1750  */
1751 ext3_fsblk_t ext3_count_free_blocks(struct super_block *sb)
1752 {
1753         ext3_fsblk_t desc_count;
1754         struct ext3_group_desc *gdp;
1755         int i;
1756         unsigned long ngroups = EXT3_SB(sb)->s_groups_count;
1757 #ifdef EXT3FS_DEBUG
1758         struct ext3_super_block *es;
1759         ext3_fsblk_t bitmap_count;
1760         unsigned long x;
1761         struct buffer_head *bitmap_bh = NULL;
1762
1763         es = EXT3_SB(sb)->s_es;
1764         desc_count = 0;
1765         bitmap_count = 0;
1766         gdp = NULL;
1767
1768         smp_rmb();
1769         for (i = 0; i < ngroups; i++) {
1770                 gdp = ext3_get_group_desc(sb, i, NULL);
1771                 if (!gdp)
1772                         continue;
1773                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1774                 brelse(bitmap_bh);
1775                 bitmap_bh = read_block_bitmap(sb, i);
1776                 if (bitmap_bh == NULL)
1777                         continue;
1778
1779                 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1780                 printk("group %d: stored = %d, counted = %lu\n",
1781                         i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1782                 bitmap_count += x;
1783         }
1784         brelse(bitmap_bh);
1785         printk("ext3_count_free_blocks: stored = "E3FSBLK
1786                 ", computed = "E3FSBLK", "E3FSBLK"\n",
1787                le32_to_cpu(es->s_free_blocks_count),
1788                 desc_count, bitmap_count);
1789         return bitmap_count;
1790 #else
1791         desc_count = 0;
1792         smp_rmb();
1793         for (i = 0; i < ngroups; i++) {
1794                 gdp = ext3_get_group_desc(sb, i, NULL);
1795                 if (!gdp)
1796                         continue;
1797                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1798         }
1799
1800         return desc_count;
1801 #endif
1802 }
1803
1804 static inline int test_root(int a, int b)
1805 {
1806         int num = b;
1807
1808         while (a > num)
1809                 num *= b;
1810         return num == a;
1811 }
1812
1813 static int ext3_group_sparse(int group)
1814 {
1815         if (group <= 1)
1816                 return 1;
1817         if (!(group & 1))
1818                 return 0;
1819         return (test_root(group, 7) || test_root(group, 5) ||
1820                 test_root(group, 3));
1821 }
1822
1823 /**
1824  *      ext3_bg_has_super - number of blocks used by the superblock in group
1825  *      @sb: superblock for filesystem
1826  *      @group: group number to check
1827  *
1828  *      Return the number of blocks used by the superblock (primary or backup)
1829  *      in this group.  Currently this will be only 0 or 1.
1830  */
1831 int ext3_bg_has_super(struct super_block *sb, int group)
1832 {
1833         if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1834                                 EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1835                         !ext3_group_sparse(group))
1836                 return 0;
1837         return 1;
1838 }
1839
1840 static unsigned long ext3_bg_num_gdb_meta(struct super_block *sb, int group)
1841 {
1842         unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1843         unsigned long first = metagroup * EXT3_DESC_PER_BLOCK(sb);
1844         unsigned long last = first + EXT3_DESC_PER_BLOCK(sb) - 1;
1845
1846         if (group == first || group == first + 1 || group == last)
1847                 return 1;
1848         return 0;
1849 }
1850
1851 static unsigned long ext3_bg_num_gdb_nometa(struct super_block *sb, int group)
1852 {
1853         return ext3_bg_has_super(sb, group) ? EXT3_SB(sb)->s_gdb_count : 0;
1854 }
1855
1856 /**
1857  *      ext3_bg_num_gdb - number of blocks used by the group table in group
1858  *      @sb: superblock for filesystem
1859  *      @group: group number to check
1860  *
1861  *      Return the number of blocks used by the group descriptor table
1862  *      (primary or backup) in this group.  In the future there may be a
1863  *      different number of descriptor blocks in each group.
1864  */
1865 unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1866 {
1867         unsigned long first_meta_bg =
1868                         le32_to_cpu(EXT3_SB(sb)->s_es->s_first_meta_bg);
1869         unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1870
1871         if (!EXT3_HAS_INCOMPAT_FEATURE(sb,EXT3_FEATURE_INCOMPAT_META_BG) ||
1872                         metagroup < first_meta_bg)
1873                 return ext3_bg_num_gdb_nometa(sb,group);
1874
1875         return ext3_bg_num_gdb_meta(sb,group);
1876
1877 }