blob: 6386d76f44a7d1d9939508e1e8e8be66b394033b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/time.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080015#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/fs.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090017#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/jbd.h>
19#include <linux/ext3_fs.h>
20#include <linux/ext3_jbd.h>
21#include <linux/quotaops.h>
22#include <linux/buffer_head.h>
Lukas Czernerb853b962010-11-22 12:29:17 +010023#include <linux/blkdev.h>
Lukas Czerner785c4bc2011-05-23 18:33:01 +020024#include <trace/events/ext3.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26/*
27 * balloc.c contains the blocks allocation and deallocation routines
28 */
29
30/*
31 * The free blocks are managed by bitmaps. A file system contains several
32 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
33 * block for inodes, N blocks for the inode table and data blocks.
34 *
35 * The file system contains group descriptors which are located after the
36 * super block. Each descriptor contains the number of the bitmap block and
37 * the free blocks count in the block. The descriptors are loaded in memory
Aneesh Kumar K.Ve6274322007-02-20 13:57:58 -080038 * when a file system is mounted (see ext3_fill_super).
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 */
40
41
42#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
43
Lukas Czernerb853b962010-11-22 12:29:17 +010044/*
45 * Calculate the block group number and offset, given a block number
46 */
47static void ext3_get_group_no_and_offset(struct super_block *sb,
48 ext3_fsblk_t blocknr, unsigned long *blockgrpp, ext3_grpblk_t *offsetp)
49{
50 struct ext3_super_block *es = EXT3_SB(sb)->s_es;
51
52 blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
53 if (offsetp)
54 *offsetp = blocknr % EXT3_BLOCKS_PER_GROUP(sb);
55 if (blockgrpp)
56 *blockgrpp = blocknr / EXT3_BLOCKS_PER_GROUP(sb);
57}
58
Mingming Cao36faadc142006-09-27 01:49:32 -070059/**
60 * ext3_get_group_desc() -- load group descriptor from disk
Dave Kleikampe9ad5622006-09-27 01:49:35 -070061 * @sb: super block
Mingming Cao36faadc142006-09-27 01:49:32 -070062 * @block_group: given block group
63 * @bh: pointer to the buffer head to store the block
64 * group descriptor
65 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070066struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb,
67 unsigned int block_group,
68 struct buffer_head ** bh)
69{
70 unsigned long group_desc;
71 unsigned long offset;
72 struct ext3_group_desc * desc;
73 struct ext3_sb_info *sbi = EXT3_SB(sb);
74
75 if (block_group >= sbi->s_groups_count) {
76 ext3_error (sb, "ext3_get_group_desc",
77 "block_group >= groups_count - "
78 "block_group = %d, groups_count = %lu",
79 block_group, sbi->s_groups_count);
80
81 return NULL;
82 }
83 smp_rmb();
84
85 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
86 offset = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
87 if (!sbi->s_group_desc[group_desc]) {
88 ext3_error (sb, "ext3_get_group_desc",
89 "Group descriptor not loaded - "
90 "block_group = %d, group_desc = %lu, desc = %lu",
91 block_group, group_desc, offset);
92 return NULL;
93 }
94
95 desc = (struct ext3_group_desc *) sbi->s_group_desc[group_desc]->b_data;
96 if (bh)
97 *bh = sbi->s_group_desc[group_desc];
98 return desc + offset;
99}
100
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800101static int ext3_valid_block_bitmap(struct super_block *sb,
102 struct ext3_group_desc *desc,
103 unsigned int block_group,
104 struct buffer_head *bh)
105{
106 ext3_grpblk_t offset;
107 ext3_grpblk_t next_zero_bit;
108 ext3_fsblk_t bitmap_blk;
109 ext3_fsblk_t group_first_block;
110
111 group_first_block = ext3_group_first_block_no(sb, block_group);
112
113 /* check whether block bitmap block number is set */
114 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
115 offset = bitmap_blk - group_first_block;
116 if (!ext3_test_bit(offset, bh->b_data))
117 /* bad block bitmap */
118 goto err_out;
119
120 /* check whether the inode bitmap block number is set */
121 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
122 offset = bitmap_blk - group_first_block;
123 if (!ext3_test_bit(offset, bh->b_data))
124 /* bad block bitmap */
125 goto err_out;
126
127 /* check whether the inode table block number is set */
128 bitmap_blk = le32_to_cpu(desc->bg_inode_table);
129 offset = bitmap_blk - group_first_block;
130 next_zero_bit = ext3_find_next_zero_bit(bh->b_data,
131 offset + EXT3_SB(sb)->s_itb_per_group,
132 offset);
133 if (next_zero_bit >= offset + EXT3_SB(sb)->s_itb_per_group)
134 /* good bitmap for inode tables */
135 return 1;
136
137err_out:
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700138 ext3_error(sb, __func__,
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800139 "Invalid block bitmap - "
140 "block_group = %d, block = %lu",
141 block_group, bitmap_blk);
142 return 0;
143}
144
Mingming Cao36faadc142006-09-27 01:49:32 -0700145/**
146 * read_block_bitmap()
147 * @sb: super block
148 * @block_group: given block group
149 *
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800150 * Read the bitmap for a given block_group,and validate the
151 * bits for block/inode/inode tables are set in the bitmaps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 *
153 * Return buffer_head on success or NULL in case of failure.
154 */
155static struct buffer_head *
156read_block_bitmap(struct super_block *sb, unsigned int block_group)
157{
158 struct ext3_group_desc * desc;
159 struct buffer_head * bh = NULL;
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800160 ext3_fsblk_t bitmap_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800162 desc = ext3_get_group_desc(sb, block_group, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 if (!desc)
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800164 return NULL;
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200165 trace_ext3_read_block_bitmap(sb, block_group);
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800166 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
167 bh = sb_getblk(sb, bitmap_blk);
168 if (unlikely(!bh)) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700169 ext3_error(sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 "Cannot read block bitmap - "
171 "block_group = %d, block_bitmap = %u",
172 block_group, le32_to_cpu(desc->bg_block_bitmap));
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800173 return NULL;
174 }
175 if (likely(bh_uptodate_or_lock(bh)))
176 return bh;
177
178 if (bh_submit_read(bh) < 0) {
179 brelse(bh);
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700180 ext3_error(sb, __func__,
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800181 "Cannot read block bitmap - "
182 "block_group = %d, block_bitmap = %u",
183 block_group, le32_to_cpu(desc->bg_block_bitmap));
184 return NULL;
185 }
Aneesh Kumar K.V2588ef82008-04-28 02:16:14 -0700186 ext3_valid_block_bitmap(sb, desc, block_group, bh);
187 /*
188 * file system mounted not to panic on error, continue with corrupt
189 * bitmap
190 */
Linus Torvalds0b832a42007-11-13 08:07:31 -0800191 return bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192}
193/*
194 * The reservation window structure operations
195 * --------------------------------------------
196 * Operations include:
197 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
198 *
Mingming Cao36faadc142006-09-27 01:49:32 -0700199 * We use a red-black tree to represent per-filesystem reservation
200 * windows.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 *
Mingming Cao36faadc142006-09-27 01:49:32 -0700202 */
203
204/**
205 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
206 * @rb_root: root of per-filesystem reservation rb tree
207 * @verbose: verbose mode
208 * @fn: function which wishes to dump the reservation map
209 *
210 * If verbose is turned on, it will print the whole block reservation
211 * windows(start, end). Otherwise, it will only print out the "bad" windows,
212 * those windows that overlap with their immediate neighbors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 */
Mingming Cao321fb9e2006-09-27 01:49:32 -0700214#if 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215static void __rsv_window_dump(struct rb_root *root, int verbose,
216 const char *fn)
217{
218 struct rb_node *n;
219 struct ext3_reserve_window_node *rsv, *prev;
220 int bad;
221
222restart:
223 n = rb_first(root);
224 bad = 0;
225 prev = NULL;
226
227 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
228 while (n) {
Hugh Dickinsc56d2562006-12-06 20:41:27 -0800229 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 if (verbose)
231 printk("reservation window 0x%p "
Mingming Cao321fb9e2006-09-27 01:49:32 -0700232 "start: %lu, end: %lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 rsv, rsv->rsv_start, rsv->rsv_end);
234 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
235 printk("Bad reservation %p (start >= end)\n",
236 rsv);
237 bad = 1;
238 }
239 if (prev && prev->rsv_end >= rsv->rsv_start) {
240 printk("Bad reservation %p (prev->end >= start)\n",
241 rsv);
242 bad = 1;
243 }
244 if (bad) {
245 if (!verbose) {
246 printk("Restarting reservation walk in verbose mode\n");
247 verbose = 1;
248 goto restart;
249 }
250 }
251 n = rb_next(n);
252 prev = rsv;
253 }
254 printk("Window map complete.\n");
Julia Lawall269b2612008-04-28 02:16:09 -0700255 BUG_ON(bad);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
257#define rsv_window_dump(root, verbose) \
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700258 __rsv_window_dump((root), (verbose), __func__)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259#else
260#define rsv_window_dump(root, verbose) do {} while (0)
261#endif
262
Mingming Cao36faadc142006-09-27 01:49:32 -0700263/**
264 * goal_in_my_reservation()
265 * @rsv: inode's reservation window
266 * @grp_goal: given goal block relative to the allocation block group
267 * @group: the current allocation block group
268 * @sb: filesystem super block
269 *
270 * Test if the given goal block (group relative) is within the file's
271 * own block reservation window range.
272 *
273 * If the reservation window is outside the goal allocation group, return 0;
274 * grp_goal (given goal block) could be -1, which means no specific
275 * goal block. In this case, always return 1.
276 * If the goal block is within the reservation window, return 1;
277 * otherwise, return 0;
278 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279static int
Mingming Cao1c2bf372006-06-25 05:48:06 -0700280goal_in_my_reservation(struct ext3_reserve_window *rsv, ext3_grpblk_t grp_goal,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 unsigned int group, struct super_block * sb)
282{
Mingming Cao1c2bf372006-06-25 05:48:06 -0700283 ext3_fsblk_t group_first_block, group_last_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
Mingming Cao43d23f92006-06-25 05:48:07 -0700285 group_first_block = ext3_group_first_block_no(sb, group);
Eric Sandeen32c2d2b2006-09-27 01:49:36 -0700286 group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
288 if ((rsv->_rsv_start > group_last_block) ||
289 (rsv->_rsv_end < group_first_block))
290 return 0;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700291 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
292 || (grp_goal + group_first_block > rsv->_rsv_end)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 return 0;
294 return 1;
295}
296
Mingming Cao36faadc142006-09-27 01:49:32 -0700297/**
298 * search_reserve_window()
299 * @rb_root: root of reservation tree
300 * @goal: target allocation block
301 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 * Find the reserved window which includes the goal, or the previous one
303 * if the goal is not in any window.
304 * Returns NULL if there are no windows or if all windows start after the goal.
305 */
306static struct ext3_reserve_window_node *
Mingming Cao1c2bf372006-06-25 05:48:06 -0700307search_reserve_window(struct rb_root *root, ext3_fsblk_t goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308{
309 struct rb_node *n = root->rb_node;
310 struct ext3_reserve_window_node *rsv;
311
312 if (!n)
313 return NULL;
314
315 do {
316 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
317
318 if (goal < rsv->rsv_start)
319 n = n->rb_left;
320 else if (goal > rsv->rsv_end)
321 n = n->rb_right;
322 else
323 return rsv;
324 } while (n);
325 /*
326 * We've fallen off the end of the tree: the goal wasn't inside
327 * any particular node. OK, the previous node must be to one
328 * side of the interval containing the goal. If it's the RHS,
329 * we need to back up one.
330 */
331 if (rsv->rsv_start > goal) {
332 n = rb_prev(&rsv->rsv_node);
333 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
334 }
335 return rsv;
336}
337
Mingming Cao36faadc142006-09-27 01:49:32 -0700338/**
339 * ext3_rsv_window_add() -- Insert a window to the block reservation rb tree.
340 * @sb: super block
341 * @rsv: reservation window to add
342 *
343 * Must be called with rsv_lock hold.
344 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345void ext3_rsv_window_add(struct super_block *sb,
346 struct ext3_reserve_window_node *rsv)
347{
348 struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
349 struct rb_node *node = &rsv->rsv_node;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700350 ext3_fsblk_t start = rsv->rsv_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
352 struct rb_node ** p = &root->rb_node;
353 struct rb_node * parent = NULL;
354 struct ext3_reserve_window_node *this;
355
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200356 trace_ext3_rsv_window_add(sb, rsv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 while (*p)
358 {
359 parent = *p;
360 this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
361
362 if (start < this->rsv_start)
363 p = &(*p)->rb_left;
364 else if (start > this->rsv_end)
365 p = &(*p)->rb_right;
Mingming Cao321fb9e2006-09-27 01:49:32 -0700366 else {
367 rsv_window_dump(root, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 BUG();
Mingming Cao321fb9e2006-09-27 01:49:32 -0700369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 }
371
372 rb_link_node(node, parent, p);
373 rb_insert_color(node, root);
374}
375
Mingming Cao36faadc142006-09-27 01:49:32 -0700376/**
377 * ext3_rsv_window_remove() -- unlink a window from the reservation rb tree
378 * @sb: super block
379 * @rsv: reservation window to remove
380 *
381 * Mark the block reservation window as not allocated, and unlink it
382 * from the filesystem reservation window rb tree. Must be called with
383 * rsv_lock hold.
384 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385static void rsv_window_remove(struct super_block *sb,
386 struct ext3_reserve_window_node *rsv)
387{
388 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
389 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
390 rsv->rsv_alloc_hit = 0;
391 rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
392}
393
Mingming Cao36faadc142006-09-27 01:49:32 -0700394/*
395 * rsv_is_empty() -- Check if the reservation window is allocated.
396 * @rsv: given reservation window to check
397 *
398 * returns 1 if the end block is EXT3_RESERVE_WINDOW_NOT_ALLOCATED.
399 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
401{
402 /* a valid reservation end block could not be 0 */
Mingming Cao36faadc142006-09-27 01:49:32 -0700403 return rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404}
Mingming Cao36faadc142006-09-27 01:49:32 -0700405
406/**
407 * ext3_init_block_alloc_info()
408 * @inode: file inode structure
409 *
410 * Allocate and initialize the reservation window structure, and
411 * link the window to the ext3 inode structure at last
412 *
413 * The reservation window structure is only dynamically allocated
414 * and linked to ext3 inode the first time the open file
415 * needs a new block. So, before every ext3_new_block(s) call, for
416 * regular files, we should check whether the reservation window
417 * structure exists or not. In the latter case, this function is called.
418 * Fail to do so will result in block reservation being turned off for that
419 * open file.
420 *
421 * This function is called from ext3_get_blocks_handle(), also called
422 * when setting the reservation window size through ioctl before the file
423 * is open for write (needs block allocation).
424 *
425 * Needs truncate_mutex protection prior to call this function.
426 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427void ext3_init_block_alloc_info(struct inode *inode)
428{
429 struct ext3_inode_info *ei = EXT3_I(inode);
430 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
431 struct super_block *sb = inode->i_sb;
432
433 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
434 if (block_i) {
435 struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node;
436
437 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
438 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
439
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700440 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 * if filesystem is mounted with NORESERVATION, the goal
442 * reservation window size is set to zero to indicate
443 * block reservation is off
444 */
445 if (!test_opt(sb, RESERVATION))
446 rsv->rsv_goal_size = 0;
447 else
448 rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS;
449 rsv->rsv_alloc_hit = 0;
450 block_i->last_alloc_logical_block = 0;
451 block_i->last_alloc_physical_block = 0;
452 }
453 ei->i_block_alloc_info = block_i;
454}
455
Mingming Cao36faadc142006-09-27 01:49:32 -0700456/**
457 * ext3_discard_reservation()
458 * @inode: inode
459 *
460 * Discard(free) block reservation window on last file close, or truncate
461 * or at last iput().
462 *
463 * It is being called in three cases:
464 * ext3_release_file(): last writer close the file
465 * ext3_clear_inode(): last iput(), when nobody link to this file.
466 * ext3_truncate(): when the block indirect map is about to change.
467 *
468 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469void ext3_discard_reservation(struct inode *inode)
470{
471 struct ext3_inode_info *ei = EXT3_I(inode);
472 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
473 struct ext3_reserve_window_node *rsv;
474 spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
475
476 if (!block_i)
477 return;
478
479 rsv = &block_i->rsv_window_node;
480 if (!rsv_is_empty(&rsv->rsv_window)) {
481 spin_lock(rsv_lock);
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200482 if (!rsv_is_empty(&rsv->rsv_window)) {
483 trace_ext3_discard_reservation(inode, rsv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 rsv_window_remove(inode->i_sb, rsv);
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200485 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 spin_unlock(rsv_lock);
487 }
488}
489
Mingming Cao36faadc142006-09-27 01:49:32 -0700490/**
491 * ext3_free_blocks_sb() -- Free given blocks and update quota
492 * @handle: handle to this transaction
493 * @sb: super block
494 * @block: start physcial block to free
495 * @count: number of blocks to free
496 * @pdquot_freed_blocks: pointer to quota
497 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
Mingming Cao1c2bf372006-06-25 05:48:06 -0700499 ext3_fsblk_t block, unsigned long count,
500 unsigned long *pdquot_freed_blocks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
502 struct buffer_head *bitmap_bh = NULL;
503 struct buffer_head *gd_bh;
504 unsigned long block_group;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700505 ext3_grpblk_t bit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 unsigned long i;
507 unsigned long overflow;
508 struct ext3_group_desc * desc;
509 struct ext3_super_block * es;
510 struct ext3_sb_info *sbi;
511 int err = 0, ret;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700512 ext3_grpblk_t group_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
514 *pdquot_freed_blocks = 0;
515 sbi = EXT3_SB(sb);
516 es = sbi->s_es;
517 if (block < le32_to_cpu(es->s_first_data_block) ||
518 block + count < block ||
519 block + count > le32_to_cpu(es->s_blocks_count)) {
520 ext3_error (sb, "ext3_free_blocks",
521 "Freeing blocks not in datazone - "
Mingming Cao1c2bf372006-06-25 05:48:06 -0700522 "block = "E3FSBLK", count = %lu", block, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 goto error_return;
524 }
525
526 ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
527
528do_more:
529 overflow = 0;
530 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
531 EXT3_BLOCKS_PER_GROUP(sb);
532 bit = (block - le32_to_cpu(es->s_first_data_block)) %
533 EXT3_BLOCKS_PER_GROUP(sb);
534 /*
535 * Check to see if we are freeing blocks across a group
536 * boundary.
537 */
538 if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
539 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
540 count -= overflow;
541 }
542 brelse(bitmap_bh);
543 bitmap_bh = read_block_bitmap(sb, block_group);
544 if (!bitmap_bh)
545 goto error_return;
546 desc = ext3_get_group_desc (sb, block_group, &gd_bh);
547 if (!desc)
548 goto error_return;
549
550 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
551 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
552 in_range (block, le32_to_cpu(desc->bg_inode_table),
553 sbi->s_itb_per_group) ||
554 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
Aneesh Kumar K.Vfeda58d2008-02-06 01:40:13 -0800555 sbi->s_itb_per_group)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 ext3_error (sb, "ext3_free_blocks",
557 "Freeing blocks in system zones - "
Mingming Cao1c2bf372006-06-25 05:48:06 -0700558 "Block = "E3FSBLK", count = %lu",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 block, count);
Aneesh Kumar K.Vfeda58d2008-02-06 01:40:13 -0800560 goto error_return;
561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
563 /*
564 * We are about to start releasing blocks in the bitmap,
565 * so we need undo access.
566 */
567 /* @@@ check errors */
568 BUFFER_TRACE(bitmap_bh, "getting undo access");
569 err = ext3_journal_get_undo_access(handle, bitmap_bh);
570 if (err)
571 goto error_return;
572
573 /*
574 * We are about to modify some metadata. Call the journal APIs
575 * to unshare ->b_data if a currently-committing transaction is
576 * using it
577 */
578 BUFFER_TRACE(gd_bh, "get_write_access");
579 err = ext3_journal_get_write_access(handle, gd_bh);
580 if (err)
581 goto error_return;
582
583 jbd_lock_bh_state(bitmap_bh);
584
585 for (i = 0, group_freed = 0; i < count; i++) {
586 /*
587 * An HJ special. This is expensive...
588 */
589#ifdef CONFIG_JBD_DEBUG
590 jbd_unlock_bh_state(bitmap_bh);
591 {
592 struct buffer_head *debug_bh;
593 debug_bh = sb_find_get_block(sb, block + i);
594 if (debug_bh) {
595 BUFFER_TRACE(debug_bh, "Deleted!");
596 if (!bh2jh(bitmap_bh)->b_committed_data)
597 BUFFER_TRACE(debug_bh,
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300598 "No committed data in bitmap");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
600 __brelse(debug_bh);
601 }
602 }
603 jbd_lock_bh_state(bitmap_bh);
604#endif
605 if (need_resched()) {
606 jbd_unlock_bh_state(bitmap_bh);
607 cond_resched();
608 jbd_lock_bh_state(bitmap_bh);
609 }
610 /* @@@ This prevents newly-allocated data from being
611 * freed and then reallocated within the same
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700612 * transaction.
613 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 * Ideally we would want to allow that to happen, but to
615 * do so requires making journal_forget() capable of
616 * revoking the queued write of a data block, which
617 * implies blocking on the journal lock. *forget()
618 * cannot block due to truncate races.
619 *
620 * Eventually we can fix this by making journal_forget()
621 * return a status indicating whether or not it was able
622 * to revoke the buffer. On successful revoke, it is
623 * safe not to set the allocation bit in the committed
624 * bitmap, because we know that there is no outstanding
625 * activity on the buffer any more and so it is safe to
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700626 * reallocate it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 */
628 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
629 J_ASSERT_BH(bitmap_bh,
630 bh2jh(bitmap_bh)->b_committed_data != NULL);
631 ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
632 bh2jh(bitmap_bh)->b_committed_data);
633
634 /*
635 * We clear the bit in the bitmap after setting the committed
636 * data bit, because this is the reverse order to that which
637 * the allocator uses.
638 */
639 BUFFER_TRACE(bitmap_bh, "clear bit");
640 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
641 bit + i, bitmap_bh->b_data)) {
642 jbd_unlock_bh_state(bitmap_bh);
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700643 ext3_error(sb, __func__,
Mingming Cao1c2bf372006-06-25 05:48:06 -0700644 "bit already cleared for block "E3FSBLK,
645 block + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 jbd_lock_bh_state(bitmap_bh);
647 BUFFER_TRACE(bitmap_bh, "bit already cleared");
648 } else {
649 group_freed++;
650 }
651 }
652 jbd_unlock_bh_state(bitmap_bh);
653
654 spin_lock(sb_bgl_lock(sbi, block_group));
Marcin Slusarz50e8a282008-02-08 04:20:13 -0800655 le16_add_cpu(&desc->bg_free_blocks_count, group_freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 spin_unlock(sb_bgl_lock(sbi, block_group));
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700657 percpu_counter_add(&sbi->s_freeblocks_counter, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
659 /* We dirtied the bitmap block */
660 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
661 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
662
663 /* And the group descriptor block */
664 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
665 ret = ext3_journal_dirty_metadata(handle, gd_bh);
666 if (!err) err = ret;
667 *pdquot_freed_blocks += group_freed;
668
669 if (overflow && !err) {
670 block += count;
671 count = overflow;
672 goto do_more;
673 }
Christoph Hellwigca41f7b2009-04-27 09:46:42 -0400674
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675error_return:
676 brelse(bitmap_bh);
677 ext3_std_error(sb, err);
678 return;
679}
680
Mingming Cao36faadc142006-09-27 01:49:32 -0700681/**
682 * ext3_free_blocks() -- Free given blocks and update quota
683 * @handle: handle for this transaction
684 * @inode: inode
685 * @block: start physical block to free
686 * @count: number of blocks to count
687 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688void ext3_free_blocks(handle_t *handle, struct inode *inode,
Mingming Cao1c2bf372006-06-25 05:48:06 -0700689 ext3_fsblk_t block, unsigned long count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200691 struct super_block *sb = inode->i_sb;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700692 unsigned long dquot_freed_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200694 trace_ext3_free_blocks(inode, block, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
696 if (dquot_freed_blocks)
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500697 dquot_free_block(inode, dquot_freed_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 return;
699}
700
Mingming Cao36faadc142006-09-27 01:49:32 -0700701/**
702 * ext3_test_allocatable()
703 * @nr: given allocation block group
704 * @bh: bufferhead contains the bitmap of the given block group
705 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 * For ext3 allocations, we must not reuse any blocks which are
707 * allocated in the bitmap buffer's "last committed data" copy. This
708 * prevents deletes from freeing up the page for reuse until we have
709 * committed the delete transaction.
710 *
711 * If we didn't do this, then deleting something and reallocating it as
712 * data would allow the old block to be overwritten before the
713 * transaction committed (because we force data to disk before commit).
714 * This would lead to corruption if we crashed between overwriting the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700715 * data and committing the delete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 *
717 * @@@ We may want to make this allocation behaviour conditional on
718 * data-writes at some point, and disable it for metadata allocations or
719 * sync-data inodes.
720 */
Mingming Cao1c2bf372006-06-25 05:48:06 -0700721static int ext3_test_allocatable(ext3_grpblk_t nr, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722{
723 int ret;
724 struct journal_head *jh = bh2jh(bh);
725
726 if (ext3_test_bit(nr, bh->b_data))
727 return 0;
728
729 jbd_lock_bh_state(bh);
730 if (!jh->b_committed_data)
731 ret = 1;
732 else
733 ret = !ext3_test_bit(nr, jh->b_committed_data);
734 jbd_unlock_bh_state(bh);
735 return ret;
736}
737
Mingming Cao36faadc142006-09-27 01:49:32 -0700738/**
739 * bitmap_search_next_usable_block()
740 * @start: the starting block (group relative) of the search
741 * @bh: bufferhead contains the block group bitmap
742 * @maxblocks: the ending block (group relative) of the reservation
743 *
744 * The bitmap search --- search forward alternately through the actual
745 * bitmap on disk and the last-committed copy in journal, until we find a
746 * bit free in both bitmaps.
747 */
Mingming Cao1c2bf372006-06-25 05:48:06 -0700748static ext3_grpblk_t
749bitmap_search_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
750 ext3_grpblk_t maxblocks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751{
Mingming Cao1c2bf372006-06-25 05:48:06 -0700752 ext3_grpblk_t next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 struct journal_head *jh = bh2jh(bh);
754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 while (start < maxblocks) {
756 next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
757 if (next >= maxblocks)
758 return -1;
759 if (ext3_test_allocatable(next, bh))
760 return next;
761 jbd_lock_bh_state(bh);
762 if (jh->b_committed_data)
763 start = ext3_find_next_zero_bit(jh->b_committed_data,
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700764 maxblocks, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 jbd_unlock_bh_state(bh);
766 }
767 return -1;
768}
769
Mingming Cao36faadc142006-09-27 01:49:32 -0700770/**
771 * find_next_usable_block()
772 * @start: the starting block (group relative) to find next
773 * allocatable block in bitmap.
774 * @bh: bufferhead contains the block group bitmap
775 * @maxblocks: the ending block (group relative) for the search
776 *
777 * Find an allocatable block in a bitmap. We honor both the bitmap and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 * its last-committed copy (if that exists), and perform the "most
779 * appropriate allocation" algorithm of looking for a free block near
780 * the initial goal; then for a free byte somewhere in the bitmap; then
781 * for any free bit in the bitmap.
782 */
Mingming Cao1c2bf372006-06-25 05:48:06 -0700783static ext3_grpblk_t
784find_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
785 ext3_grpblk_t maxblocks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
Mingming Cao1c2bf372006-06-25 05:48:06 -0700787 ext3_grpblk_t here, next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 char *p, *r;
789
790 if (start > 0) {
791 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700792 * The goal was occupied; search forward for a free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 * block within the next XX blocks.
794 *
795 * end_goal is more or less random, but it has to be
796 * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
797 * next 64-bit boundary is simple..
798 */
Mingming Cao1c2bf372006-06-25 05:48:06 -0700799 ext3_grpblk_t end_goal = (start + 63) & ~63;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 if (end_goal > maxblocks)
801 end_goal = maxblocks;
802 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
803 if (here < end_goal && ext3_test_allocatable(here, bh))
804 return here;
805 ext3_debug("Bit not found near goal\n");
806 }
807
808 here = start;
809 if (here < 0)
810 here = 0;
811
Namhyung Kim57e94d82010-10-11 02:10:45 +0900812 p = bh->b_data + (here >> 3);
Hugh Dickins7d1c5202006-12-06 20:41:30 -0800813 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
Namhyung Kim57e94d82010-10-11 02:10:45 +0900814 next = (r - bh->b_data) << 3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
816 if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
817 return next;
818
819 /*
820 * The bitmap search --- search forward alternately through the actual
821 * bitmap and the last-committed copy until we find a bit free in
822 * both
823 */
824 here = bitmap_search_next_usable_block(here, bh, maxblocks);
825 return here;
826}
827
Mingming Cao36faadc142006-09-27 01:49:32 -0700828/**
829 * claim_block()
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900830 * @lock: the spin lock for this block group
Mingming Cao36faadc142006-09-27 01:49:32 -0700831 * @block: the free block (group relative) to allocate
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900832 * @bh: the buffer_head contains the block group bitmap
Mingming Cao36faadc142006-09-27 01:49:32 -0700833 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 * We think we can allocate this block in this bitmap. Try to set the bit.
835 * If that succeeds then check that nobody has allocated and then freed the
836 * block since we saw that is was not marked in b_committed_data. If it _was_
837 * allocated and freed then clear the bit in the bitmap again and return
838 * zero (failure).
839 */
840static inline int
Mingming Cao1c2bf372006-06-25 05:48:06 -0700841claim_block(spinlock_t *lock, ext3_grpblk_t block, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
843 struct journal_head *jh = bh2jh(bh);
844 int ret;
845
846 if (ext3_set_bit_atomic(lock, block, bh->b_data))
847 return 0;
848 jbd_lock_bh_state(bh);
849 if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
850 ext3_clear_bit_atomic(lock, block, bh->b_data);
851 ret = 0;
852 } else {
853 ret = 1;
854 }
855 jbd_unlock_bh_state(bh);
856 return ret;
857}
858
Mingming Cao36faadc142006-09-27 01:49:32 -0700859/**
860 * ext3_try_to_allocate()
861 * @sb: superblock
862 * @handle: handle to this transaction
863 * @group: given allocation block group
864 * @bitmap_bh: bufferhead holds the block bitmap
865 * @grp_goal: given target block within the group
866 * @count: target number of blocks to allocate
867 * @my_rsv: reservation window
868 *
869 * Attempt to allocate blocks within a give range. Set the range of allocation
870 * first, then find the first free bit(s) from the bitmap (within the range),
871 * and at last, allocate the blocks by claiming the found free bit as allocated.
872 *
873 * To set the range of this allocation:
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700874 * if there is a reservation window, only try to allocate block(s) from the
Mingming Cao36faadc142006-09-27 01:49:32 -0700875 * file's own reservation window;
876 * Otherwise, the allocation range starts from the give goal block, ends at
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700877 * the block group's last block.
Mingming Cao36faadc142006-09-27 01:49:32 -0700878 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 * If we failed to allocate the desired block then we may end up crossing to a
880 * new bitmap. In that case we must release write access to the old one via
881 * ext3_journal_release_buffer(), else we'll run out of credits.
882 */
Mingming Cao1c2bf372006-06-25 05:48:06 -0700883static ext3_grpblk_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
Mingming Cao1c2bf372006-06-25 05:48:06 -0700885 struct buffer_head *bitmap_bh, ext3_grpblk_t grp_goal,
Mingming Caob54e41e2006-03-26 01:37:57 -0800886 unsigned long *count, struct ext3_reserve_window *my_rsv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887{
Mingming Cao1c2bf372006-06-25 05:48:06 -0700888 ext3_fsblk_t group_first_block;
889 ext3_grpblk_t start, end;
Mingming Caob54e41e2006-03-26 01:37:57 -0800890 unsigned long num = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
892 /* we do allocation within the reservation window if we have a window */
893 if (my_rsv) {
Mingming Cao43d23f92006-06-25 05:48:07 -0700894 group_first_block = ext3_group_first_block_no(sb, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 if (my_rsv->_rsv_start >= group_first_block)
896 start = my_rsv->_rsv_start - group_first_block;
897 else
898 /* reservation window cross group boundary */
899 start = 0;
900 end = my_rsv->_rsv_end - group_first_block + 1;
901 if (end > EXT3_BLOCKS_PER_GROUP(sb))
902 /* reservation window crosses group boundary */
903 end = EXT3_BLOCKS_PER_GROUP(sb);
Mingming Cao1c2bf372006-06-25 05:48:06 -0700904 if ((start <= grp_goal) && (grp_goal < end))
905 start = grp_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 else
Mingming Cao1c2bf372006-06-25 05:48:06 -0700907 grp_goal = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 } else {
Mingming Cao1c2bf372006-06-25 05:48:06 -0700909 if (grp_goal > 0)
910 start = grp_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 else
912 start = 0;
913 end = EXT3_BLOCKS_PER_GROUP(sb);
914 }
915
916 BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
917
918repeat:
Mingming Cao1c2bf372006-06-25 05:48:06 -0700919 if (grp_goal < 0 || !ext3_test_allocatable(grp_goal, bitmap_bh)) {
920 grp_goal = find_next_usable_block(start, bitmap_bh, end);
921 if (grp_goal < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 goto fail_access;
923 if (!my_rsv) {
924 int i;
925
Mingming Cao1c2bf372006-06-25 05:48:06 -0700926 for (i = 0; i < 7 && grp_goal > start &&
927 ext3_test_allocatable(grp_goal - 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 bitmap_bh);
Mingming Cao1c2bf372006-06-25 05:48:06 -0700929 i++, grp_goal--)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 ;
931 }
932 }
Mingming Cao1c2bf372006-06-25 05:48:06 -0700933 start = grp_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
Mingming Cao36faadc142006-09-27 01:49:32 -0700935 if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group),
936 grp_goal, bitmap_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 /*
938 * The block was allocated by another thread, or it was
939 * allocated and then freed by another thread
940 */
941 start++;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700942 grp_goal++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 if (start >= end)
944 goto fail_access;
945 goto repeat;
946 }
Mingming Caob54e41e2006-03-26 01:37:57 -0800947 num++;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700948 grp_goal++;
949 while (num < *count && grp_goal < end
950 && ext3_test_allocatable(grp_goal, bitmap_bh)
Mingming Cao36faadc142006-09-27 01:49:32 -0700951 && claim_block(sb_bgl_lock(EXT3_SB(sb), group),
952 grp_goal, bitmap_bh)) {
Mingming Caob54e41e2006-03-26 01:37:57 -0800953 num++;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700954 grp_goal++;
Mingming Caob54e41e2006-03-26 01:37:57 -0800955 }
956 *count = num;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700957 return grp_goal - num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958fail_access:
Mingming Caob54e41e2006-03-26 01:37:57 -0800959 *count = num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 return -1;
961}
962
963/**
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700964 * find_next_reservable_window():
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 * find a reservable space within the given range.
966 * It does not allocate the reservation window for now:
967 * alloc_new_reservation() will do the work later.
968 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700969 * @search_head: the head of the searching list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 * This is not necessarily the list head of the whole filesystem
971 *
972 * We have both head and start_block to assist the search
973 * for the reservable space. The list starts from head,
974 * but we will shift to the place where start_block is,
975 * then start from there, when looking for a reservable space.
976 *
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900977 * @my_rsv: the reservation window
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 *
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900979 * @sb: the super block
980 *
981 * @start_block: the first block we consider to start
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 * the real search from
983 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700984 * @last_block:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 * the maximum block number that our goal reservable space
986 * could start from. This is normally the last block in this
987 * group. The search will end when we found the start of next
988 * possible reservable space is out of this boundary.
989 * This could handle the cross boundary reservation window
990 * request.
991 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700992 * basically we search from the given range, rather than the whole
993 * reservation double linked list, (start_block, last_block)
994 * to find a free region that is of my size and has not
995 * been reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 */
Mingming Cao21fe3472005-06-28 20:45:16 -0700998static int find_next_reservable_window(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 struct ext3_reserve_window_node *search_head,
Mingming Cao21fe3472005-06-28 20:45:16 -07001000 struct ext3_reserve_window_node *my_rsv,
Mingming Cao1c2bf372006-06-25 05:48:06 -07001001 struct super_block * sb,
1002 ext3_fsblk_t start_block,
1003 ext3_fsblk_t last_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
1005 struct rb_node *next;
1006 struct ext3_reserve_window_node *rsv, *prev;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001007 ext3_fsblk_t cur;
Mingming Cao21fe3472005-06-28 20:45:16 -07001008 int size = my_rsv->rsv_goal_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010 /* TODO: make the start of the reservation window byte-aligned */
1011 /* cur = *start_block & ~7;*/
Mingming Cao21fe3472005-06-28 20:45:16 -07001012 cur = start_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 rsv = search_head;
1014 if (!rsv)
Mingming Cao21fe3472005-06-28 20:45:16 -07001015 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
1017 while (1) {
1018 if (cur <= rsv->rsv_end)
1019 cur = rsv->rsv_end + 1;
1020
1021 /* TODO?
1022 * in the case we could not find a reservable space
1023 * that is what is expected, during the re-search, we could
1024 * remember what's the largest reservable space we could have
1025 * and return that one.
1026 *
1027 * For now it will fail if we could not find the reservable
1028 * space with expected-size (or more)...
1029 */
1030 if (cur > last_block)
Mingming Cao21fe3472005-06-28 20:45:16 -07001031 return -1; /* fail */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
1033 prev = rsv;
1034 next = rb_next(&rsv->rsv_node);
Hugh Dickinsc56d2562006-12-06 20:41:27 -08001035 rsv = rb_entry(next,struct ext3_reserve_window_node,rsv_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
1037 /*
1038 * Reached the last reservation, we can just append to the
1039 * previous one.
1040 */
1041 if (!next)
1042 break;
1043
1044 if (cur + size <= rsv->rsv_start) {
1045 /*
1046 * Found a reserveable space big enough. We could
1047 * have a reservation across the group boundary here
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001048 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 break;
1050 }
1051 }
1052 /*
1053 * we come here either :
1054 * when we reach the end of the whole list,
1055 * and there is empty reservable space after last entry in the list.
1056 * append it to the end of the list.
1057 *
1058 * or we found one reservable space in the middle of the list,
1059 * return the reservation window that we could append to.
1060 * succeed.
1061 */
Mingming Cao21fe3472005-06-28 20:45:16 -07001062
1063 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
1064 rsv_window_remove(sb, my_rsv);
1065
1066 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001067 * Let's book the whole available window for now. We will check the
Mingming Cao21fe3472005-06-28 20:45:16 -07001068 * disk bitmap later and then, if there are free blocks then we adjust
1069 * the window size if it's larger than requested.
1070 * Otherwise, we will remove this node from the tree next time
1071 * call find_next_reservable_window.
1072 */
1073 my_rsv->rsv_start = cur;
1074 my_rsv->rsv_end = cur + size - 1;
1075 my_rsv->rsv_alloc_hit = 0;
1076
1077 if (prev != my_rsv)
1078 ext3_rsv_window_add(sb, my_rsv);
1079
1080 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081}
1082
1083/**
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001084 * alloc_new_reservation()--allocate a new reservation window
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 *
1086 * To make a new reservation, we search part of the filesystem
1087 * reservation list (the list that inside the group). We try to
1088 * allocate a new reservation window near the allocation goal,
1089 * or the beginning of the group, if there is no goal.
1090 *
1091 * We first find a reservable space after the goal, then from
1092 * there, we check the bitmap for the first free block after
1093 * it. If there is no free block until the end of group, then the
1094 * whole group is full, we failed. Otherwise, check if the free
1095 * block is inside the expected reservable space, if so, we
1096 * succeed.
1097 * If the first free block is outside the reservable space, then
1098 * start from the first free block, we search for next available
1099 * space, and go on.
1100 *
1101 * on succeed, a new reservation will be found and inserted into the list
1102 * It contains at least one free block, and it does not overlap with other
1103 * reservation windows.
1104 *
1105 * failed: we failed to find a reservation window in this group
1106 *
Namhyung Kima4c18ad2010-10-19 00:34:35 +09001107 * @my_rsv: the reservation window
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 *
Mingming Cao1c2bf372006-06-25 05:48:06 -07001109 * @grp_goal: The goal (group-relative). It is where the search for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 * free reservable space should start from.
Mingming Cao1c2bf372006-06-25 05:48:06 -07001111 * if we have a grp_goal(grp_goal >0 ), then start from there,
1112 * no grp_goal(grp_goal = -1), we start from the first block
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 * of the group.
1114 *
1115 * @sb: the super block
1116 * @group: the group we are trying to allocate in
1117 * @bitmap_bh: the block group block bitmap
Mingming Cao21fe3472005-06-28 20:45:16 -07001118 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 */
1120static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
Mingming Cao1c2bf372006-06-25 05:48:06 -07001121 ext3_grpblk_t grp_goal, struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 unsigned int group, struct buffer_head *bitmap_bh)
1123{
1124 struct ext3_reserve_window_node *search_head;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001125 ext3_fsblk_t group_first_block, group_end_block, start_block;
1126 ext3_grpblk_t first_free_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
1128 unsigned long size;
Mingming Cao21fe3472005-06-28 20:45:16 -07001129 int ret;
1130 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
Mingming Cao43d23f92006-06-25 05:48:07 -07001132 group_first_block = ext3_group_first_block_no(sb, group);
Eric Sandeen32c2d2b2006-09-27 01:49:36 -07001133 group_end_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
Mingming Cao1c2bf372006-06-25 05:48:06 -07001135 if (grp_goal < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 start_block = group_first_block;
1137 else
Mingming Cao1c2bf372006-06-25 05:48:06 -07001138 start_block = grp_goal + group_first_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001140 trace_ext3_alloc_new_reservation(sb, start_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 size = my_rsv->rsv_goal_size;
Mingming Cao21fe3472005-06-28 20:45:16 -07001142
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 if (!rsv_is_empty(&my_rsv->rsv_window)) {
1144 /*
1145 * if the old reservation is cross group boundary
1146 * and if the goal is inside the old reservation window,
1147 * we will come here when we just failed to allocate from
1148 * the first part of the window. We still have another part
1149 * that belongs to the next group. In this case, there is no
1150 * point to discard our window and try to allocate a new one
1151 * in this group(which will fail). we should
1152 * keep the reservation window, just simply move on.
1153 *
1154 * Maybe we could shift the start block of the reservation
1155 * window to the first block of next group.
1156 */
1157
1158 if ((my_rsv->rsv_start <= group_end_block) &&
1159 (my_rsv->rsv_end > group_end_block) &&
1160 (start_block >= my_rsv->rsv_start))
1161 return -1;
1162
1163 if ((my_rsv->rsv_alloc_hit >
1164 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
1165 /*
Mingming Cao36faadc142006-09-27 01:49:32 -07001166 * if the previously allocation hit ratio is
1167 * greater than 1/2, then we double the size of
1168 * the reservation window the next time,
1169 * otherwise we keep the same size window
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 */
1171 size = size * 2;
1172 if (size > EXT3_MAX_RESERVE_BLOCKS)
1173 size = EXT3_MAX_RESERVE_BLOCKS;
1174 my_rsv->rsv_goal_size= size;
1175 }
1176 }
Mingming Cao21fe3472005-06-28 20:45:16 -07001177
1178 spin_lock(rsv_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 /*
1180 * shift the search start to the window near the goal block
1181 */
1182 search_head = search_reserve_window(fs_rsv_root, start_block);
1183
1184 /*
1185 * find_next_reservable_window() simply finds a reservable window
1186 * inside the given range(start_block, group_end_block).
1187 *
1188 * To make sure the reservation window has a free bit inside it, we
1189 * need to check the bitmap after we found a reservable window.
1190 */
1191retry:
Mingming Cao21fe3472005-06-28 20:45:16 -07001192 ret = find_next_reservable_window(search_head, my_rsv, sb,
1193 start_block, group_end_block);
1194
1195 if (ret == -1) {
1196 if (!rsv_is_empty(&my_rsv->rsv_window))
1197 rsv_window_remove(sb, my_rsv);
1198 spin_unlock(rsv_lock);
1199 return -1;
1200 }
1201
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 /*
1203 * On success, find_next_reservable_window() returns the
1204 * reservation window where there is a reservable space after it.
1205 * Before we reserve this reservable space, we need
1206 * to make sure there is at least a free block inside this region.
1207 *
1208 * searching the first free bit on the block bitmap and copy of
1209 * last committed bitmap alternatively, until we found a allocatable
1210 * block. Search start from the start block of the reservable space
1211 * we just found.
1212 */
Mingming Cao21fe3472005-06-28 20:45:16 -07001213 spin_unlock(rsv_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 first_free_block = bitmap_search_next_usable_block(
Mingming Cao21fe3472005-06-28 20:45:16 -07001215 my_rsv->rsv_start - group_first_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 bitmap_bh, group_end_block - group_first_block + 1);
1217
1218 if (first_free_block < 0) {
1219 /*
1220 * no free block left on the bitmap, no point
1221 * to reserve the space. return failed.
1222 */
Mingming Cao21fe3472005-06-28 20:45:16 -07001223 spin_lock(rsv_lock);
1224 if (!rsv_is_empty(&my_rsv->rsv_window))
1225 rsv_window_remove(sb, my_rsv);
1226 spin_unlock(rsv_lock);
1227 return -1; /* failed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 }
Mingming Cao21fe3472005-06-28 20:45:16 -07001229
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 start_block = first_free_block + group_first_block;
1231 /*
1232 * check if the first free block is within the
Mingming Cao21fe3472005-06-28 20:45:16 -07001233 * free space we just reserved
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 */
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001235 if (start_block >= my_rsv->rsv_start &&
1236 start_block <= my_rsv->rsv_end) {
1237 trace_ext3_reserved(sb, start_block, my_rsv);
Mingming Cao21fe3472005-06-28 20:45:16 -07001238 return 0; /* success */
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001239 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 /*
1241 * if the first free bit we found is out of the reservable space
Mingming Cao21fe3472005-06-28 20:45:16 -07001242 * continue search for next reservable space,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 * start from where the free block is,
1244 * we also shift the list head to where we stopped last time
1245 */
Mingming Cao21fe3472005-06-28 20:45:16 -07001246 search_head = my_rsv;
1247 spin_lock(rsv_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249}
1250
Mingming Cao36faadc142006-09-27 01:49:32 -07001251/**
1252 * try_to_extend_reservation()
1253 * @my_rsv: given reservation window
1254 * @sb: super block
1255 * @size: the delta to extend
1256 *
1257 * Attempt to expand the reservation window large enough to have
1258 * required number of free blocks
1259 *
1260 * Since ext3_try_to_allocate() will always allocate blocks within
1261 * the reservation window range, if the window size is too small,
1262 * multiple blocks allocation has to stop at the end of the reservation
1263 * window. To make this more efficient, given the total number of
1264 * blocks needed and the current size of the window, we try to
1265 * expand the reservation window size if necessary on a best-effort
1266 * basis before ext3_new_blocks() tries to allocate blocks,
1267 */
Mingming Caod48589b2006-03-26 01:37:59 -08001268static void try_to_extend_reservation(struct ext3_reserve_window_node *my_rsv,
1269 struct super_block *sb, int size)
1270{
1271 struct ext3_reserve_window_node *next_rsv;
1272 struct rb_node *next;
1273 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1274
1275 if (!spin_trylock(rsv_lock))
1276 return;
1277
1278 next = rb_next(&my_rsv->rsv_node);
1279
1280 if (!next)
1281 my_rsv->rsv_end += size;
1282 else {
Hugh Dickinsc56d2562006-12-06 20:41:27 -08001283 next_rsv = rb_entry(next, struct ext3_reserve_window_node, rsv_node);
Mingming Caod48589b2006-03-26 01:37:59 -08001284
1285 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1286 my_rsv->rsv_end += size;
1287 else
1288 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1289 }
1290 spin_unlock(rsv_lock);
1291}
1292
Mingming Cao36faadc142006-09-27 01:49:32 -07001293/**
1294 * ext3_try_to_allocate_with_rsv()
1295 * @sb: superblock
1296 * @handle: handle to this transaction
1297 * @group: given allocation block group
1298 * @bitmap_bh: bufferhead holds the block bitmap
1299 * @grp_goal: given target block within the group
Mingming Cao36faadc142006-09-27 01:49:32 -07001300 * @my_rsv: reservation window
Namhyung Kima4c18ad2010-10-19 00:34:35 +09001301 * @count: target number of blocks to allocate
Mingming Cao36faadc142006-09-27 01:49:32 -07001302 * @errp: pointer to store the error code
1303 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 * This is the main function used to allocate a new block and its reservation
1305 * window.
1306 *
1307 * Each time when a new block allocation is need, first try to allocate from
1308 * its own reservation. If it does not have a reservation window, instead of
1309 * looking for a free bit on bitmap first, then look up the reservation list to
1310 * see if it is inside somebody else's reservation window, we try to allocate a
1311 * reservation window for it starting from the goal first. Then do the block
1312 * allocation within the reservation window.
1313 *
1314 * This will avoid keeping on searching the reservation list again and
Glauber de Oliveira Costa5b116872005-10-30 15:02:48 -08001315 * again when somebody is looking for a free block (without
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 * reservation), and there are lots of free blocks, but they are all
1317 * being reserved.
1318 *
Mingming Cao36faadc142006-09-27 01:49:32 -07001319 * We use a red-black tree for the per-filesystem reservation list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 *
1321 */
Mingming Cao1c2bf372006-06-25 05:48:06 -07001322static ext3_grpblk_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1324 unsigned int group, struct buffer_head *bitmap_bh,
Mingming Cao1c2bf372006-06-25 05:48:06 -07001325 ext3_grpblk_t grp_goal,
1326 struct ext3_reserve_window_node * my_rsv,
Mingming Caob54e41e2006-03-26 01:37:57 -08001327 unsigned long *count, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
Eric Sandeen32c2d2b2006-09-27 01:49:36 -07001329 ext3_fsblk_t group_first_block, group_last_block;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001330 ext3_grpblk_t ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 int fatal;
Mingming Caob54e41e2006-03-26 01:37:57 -08001332 unsigned long num = *count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
1334 *errp = 0;
1335
1336 /*
1337 * Make sure we use undo access for the bitmap, because it is critical
1338 * that we do the frozen_data COW on bitmap buffers in all cases even
1339 * if the buffer is in BJ_Forget state in the committing transaction.
1340 */
1341 BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1342 fatal = ext3_journal_get_undo_access(handle, bitmap_bh);
1343 if (fatal) {
1344 *errp = fatal;
1345 return -1;
1346 }
1347
1348 /*
1349 * we don't deal with reservation when
1350 * filesystem is mounted without reservation
1351 * or the file is not a regular file
1352 * or last attempt to allocate a block with reservation turned on failed
1353 */
1354 if (my_rsv == NULL ) {
Mingming Caob54e41e2006-03-26 01:37:57 -08001355 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
Mingming Cao1c2bf372006-06-25 05:48:06 -07001356 grp_goal, count, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 goto out;
1358 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 /*
Mingming Cao1c2bf372006-06-25 05:48:06 -07001360 * grp_goal is a group relative block number (if there is a goal)
Hugh Dickins16502422006-12-06 20:41:25 -08001361 * 0 <= grp_goal < EXT3_BLOCKS_PER_GROUP(sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 * first block is a filesystem wide block number
1363 * first block is the block number of the first block in this group
1364 */
Mingming Cao43d23f92006-06-25 05:48:07 -07001365 group_first_block = ext3_group_first_block_no(sb, group);
Eric Sandeen32c2d2b2006-09-27 01:49:36 -07001366 group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
1368 /*
1369 * Basically we will allocate a new block from inode's reservation
1370 * window.
1371 *
1372 * We need to allocate a new reservation window, if:
1373 * a) inode does not have a reservation window; or
1374 * b) last attempt to allocate a block from existing reservation
1375 * failed; or
1376 * c) we come here with a goal and with a reservation window
1377 *
1378 * We do not need to allocate a new reservation window if we come here
1379 * at the beginning with a goal and the goal is inside the window, or
1380 * we don't have a goal but already have a reservation window.
1381 * then we could go to allocate from the reservation window directly.
1382 */
1383 while (1) {
Mingming Cao21fe3472005-06-28 20:45:16 -07001384 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
Mingming Cao36faadc142006-09-27 01:49:32 -07001385 !goal_in_my_reservation(&my_rsv->rsv_window,
1386 grp_goal, group, sb)) {
Mingming Caod48589b2006-03-26 01:37:59 -08001387 if (my_rsv->rsv_goal_size < *count)
1388 my_rsv->rsv_goal_size = *count;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001389 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 group, bitmap_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 if (ret < 0)
1392 break; /* failed */
1393
Mingming Cao36faadc142006-09-27 01:49:32 -07001394 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1395 grp_goal, group, sb))
Mingming Cao1c2bf372006-06-25 05:48:06 -07001396 grp_goal = -1;
Hugh Dickins16502422006-12-06 20:41:25 -08001397 } else if (grp_goal >= 0) {
Mingming Cao2bd94bd2006-12-06 20:38:18 -08001398 int curr = my_rsv->rsv_end -
1399 (grp_goal + group_first_block) + 1;
1400
1401 if (curr < *count)
1402 try_to_extend_reservation(my_rsv, sb,
1403 *count - curr);
1404 }
Mingming Caod48589b2006-03-26 01:37:59 -08001405
Eric Sandeen32c2d2b2006-09-27 01:49:36 -07001406 if ((my_rsv->rsv_start > group_last_block) ||
1407 (my_rsv->rsv_end < group_first_block)) {
Mingming Cao321fb9e2006-09-27 01:49:32 -07001408 rsv_window_dump(&EXT3_SB(sb)->s_rsv_window_root, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 BUG();
Mingming Cao321fb9e2006-09-27 01:49:32 -07001410 }
Mingming Cao36faadc142006-09-27 01:49:32 -07001411 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1412 grp_goal, &num, &my_rsv->rsv_window);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 if (ret >= 0) {
Mingming Caob54e41e2006-03-26 01:37:57 -08001414 my_rsv->rsv_alloc_hit += num;
1415 *count = num;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 break; /* succeed */
1417 }
Mingming Caob54e41e2006-03-26 01:37:57 -08001418 num = *count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 }
1420out:
1421 if (ret >= 0) {
1422 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1423 "bitmap block");
1424 fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1425 if (fatal) {
1426 *errp = fatal;
1427 return -1;
1428 }
1429 return ret;
1430 }
1431
1432 BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1433 ext3_journal_release_buffer(handle, bitmap_bh);
1434 return ret;
1435}
1436
Mingming Cao36faadc142006-09-27 01:49:32 -07001437/**
1438 * ext3_has_free_blocks()
1439 * @sbi: in-core super block structure.
1440 *
1441 * Check if filesystem has at least 1 free block available for allocation.
1442 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443static int ext3_has_free_blocks(struct ext3_sb_info *sbi)
1444{
Mingming Cao1c2bf372006-06-25 05:48:06 -07001445 ext3_fsblk_t free_blocks, root_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
1447 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1448 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1449 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
David Howells6a2f90e2008-11-14 10:38:51 +11001450 sbi->s_resuid != current_fsuid() &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1452 return 0;
1453 }
1454 return 1;
1455}
1456
Mingming Cao36faadc142006-09-27 01:49:32 -07001457/**
1458 * ext3_should_retry_alloc()
1459 * @sb: super block
1460 * @retries number of attemps has been made
1461 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1463 * it is profitable to retry the operation, this function will wait
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001464 * for the current or committing transaction to complete, and then
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 * return TRUE.
Mingming Cao36faadc142006-09-27 01:49:32 -07001466 *
1467 * if the total number of retries exceed three times, return FALSE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 */
1469int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1470{
1471 if (!ext3_has_free_blocks(EXT3_SB(sb)) || (*retries)++ > 3)
1472 return 0;
1473
1474 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1475
1476 return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1477}
1478
Mingming Cao36faadc142006-09-27 01:49:32 -07001479/**
1480 * ext3_new_blocks() -- core block(s) allocation function
1481 * @handle: handle to this transaction
1482 * @inode: file inode
1483 * @goal: given target block(filesystem wide)
1484 * @count: target number of blocks to allocate
1485 * @errp: error code
1486 *
1487 * ext3_new_blocks uses a goal block to assist allocation. It tries to
1488 * allocate block(s) from the block group contains the goal block first. If that
1489 * fails, it will try to allocate block(s) from other block groups without
1490 * any specific goal block.
1491 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 */
Mingming Cao1c2bf372006-06-25 05:48:06 -07001493ext3_fsblk_t ext3_new_blocks(handle_t *handle, struct inode *inode,
1494 ext3_fsblk_t goal, unsigned long *count, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495{
1496 struct buffer_head *bitmap_bh = NULL;
1497 struct buffer_head *gdp_bh;
1498 int group_no;
1499 int goal_group;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001500 ext3_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1501 ext3_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1502 ext3_fsblk_t ret_block; /* filesyetem-wide allocated block */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 int bgi; /* blockgroup iteration index */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 int fatal = 0, err;
1505 int performed_allocation = 0;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001506 ext3_grpblk_t free_blocks; /* number of free blocks in a group */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 struct super_block *sb;
1508 struct ext3_group_desc *gdp;
1509 struct ext3_super_block *es;
1510 struct ext3_sb_info *sbi;
1511 struct ext3_reserve_window_node *my_rsv = NULL;
1512 struct ext3_block_alloc_info *block_i;
1513 unsigned short windowsz = 0;
1514#ifdef EXT3FS_DEBUG
1515 static int goal_hits, goal_attempts;
1516#endif
1517 unsigned long ngroups;
Mingming Caob54e41e2006-03-26 01:37:57 -08001518 unsigned long num = *count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
1520 *errp = -ENOSPC;
1521 sb = inode->i_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
1523 /*
1524 * Check quota for allocation of this block.
1525 */
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001526 err = dquot_alloc_block(inode, num);
1527 if (err) {
1528 *errp = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 return 0;
1530 }
1531
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001532 trace_ext3_request_blocks(inode, goal, num);
1533
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 sbi = EXT3_SB(sb);
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001535 es = sbi->s_es;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 ext3_debug("goal=%lu.\n", goal);
1537 /*
1538 * Allocate a block from reservation only when
1539 * filesystem is mounted with reservation(default,-o reservation), and
1540 * it's a regular file, and
1541 * the desired window size is greater than 0 (One could use ioctl
1542 * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1543 * reservation on that particular file)
1544 */
1545 block_i = EXT3_I(inode)->i_block_alloc_info;
1546 if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1547 my_rsv = &block_i->rsv_window_node;
1548
1549 if (!ext3_has_free_blocks(sbi)) {
1550 *errp = -ENOSPC;
1551 goto out;
1552 }
1553
1554 /*
1555 * First, test whether the goal block is free.
1556 */
1557 if (goal < le32_to_cpu(es->s_first_data_block) ||
1558 goal >= le32_to_cpu(es->s_blocks_count))
1559 goal = le32_to_cpu(es->s_first_data_block);
1560 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1561 EXT3_BLOCKS_PER_GROUP(sb);
Mingming Cao08fb306f2006-08-27 01:23:44 -07001562 goal_group = group_no;
1563retry_alloc:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1565 if (!gdp)
1566 goto io_error;
1567
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1569 /*
1570 * if there is not enough free blocks to make a new resevation
1571 * turn off reservation for this allocation
1572 */
1573 if (my_rsv && (free_blocks < windowsz)
Mingming Cao46d01a22008-10-18 20:27:56 -07001574 && (free_blocks > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 && (rsv_is_empty(&my_rsv->rsv_window)))
1576 my_rsv = NULL;
1577
1578 if (free_blocks > 0) {
Mingming Cao1c2bf372006-06-25 05:48:06 -07001579 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 EXT3_BLOCKS_PER_GROUP(sb));
1581 bitmap_bh = read_block_bitmap(sb, group_no);
1582 if (!bitmap_bh)
1583 goto io_error;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001584 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1585 group_no, bitmap_bh, grp_target_blk,
1586 my_rsv, &num, &fatal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 if (fatal)
1588 goto out;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001589 if (grp_alloc_blk >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 goto allocated;
1591 }
1592
1593 ngroups = EXT3_SB(sb)->s_groups_count;
1594 smp_rmb();
1595
1596 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001597 * Now search the rest of the groups. We assume that
Akinobu Mita144704e2008-02-06 01:40:15 -08001598 * group_no and gdp correctly point to the last group visited.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 */
1600 for (bgi = 0; bgi < ngroups; bgi++) {
1601 group_no++;
1602 if (group_no >= ngroups)
1603 group_no = 0;
1604 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
Hugh Dickins2823b552006-12-06 20:41:28 -08001605 if (!gdp)
1606 goto io_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1608 /*
Frans van de Wiel8cef1072010-03-15 19:29:34 +01001609 * skip this group (and avoid loading bitmap) if there
1610 * are no free blocks
1611 */
1612 if (!free_blocks)
1613 continue;
1614 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 * skip this group if the number of
1616 * free blocks is less than half of the reservation
1617 * window size.
1618 */
Mingming Cao46d01a22008-10-18 20:27:56 -07001619 if (my_rsv && (free_blocks <= (windowsz/2)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 continue;
1621
1622 brelse(bitmap_bh);
1623 bitmap_bh = read_block_bitmap(sb, group_no);
1624 if (!bitmap_bh)
1625 goto io_error;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001626 /*
1627 * try to allocate block(s) from this group, without a goal(-1).
1628 */
1629 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1630 group_no, bitmap_bh, -1, my_rsv,
1631 &num, &fatal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 if (fatal)
1633 goto out;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001634 if (grp_alloc_blk >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 goto allocated;
1636 }
1637 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001638 * We may end up a bogus earlier ENOSPC error due to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 * filesystem is "full" of reservations, but
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001640 * there maybe indeed free blocks available on disk
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 * In this case, we just forget about the reservations
1642 * just do block allocation as without reservations.
1643 */
1644 if (my_rsv) {
1645 my_rsv = NULL;
Hugh Dickinsef503672006-12-06 20:41:23 -08001646 windowsz = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 group_no = goal_group;
Mingming Cao08fb306f2006-08-27 01:23:44 -07001648 goto retry_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 }
1650 /* No space left on the device */
1651 *errp = -ENOSPC;
1652 goto out;
1653
1654allocated:
1655
1656 ext3_debug("using block group %d(%d)\n",
1657 group_no, gdp->bg_free_blocks_count);
1658
1659 BUFFER_TRACE(gdp_bh, "get_write_access");
1660 fatal = ext3_journal_get_write_access(handle, gdp_bh);
1661 if (fatal)
1662 goto out;
1663
Mingming Cao43d23f92006-06-25 05:48:07 -07001664 ret_block = grp_alloc_blk + ext3_group_first_block_no(sb, group_no);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
Mingming Cao1c2bf372006-06-25 05:48:06 -07001666 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1667 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1668 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
Mingming Caofaa56972006-03-26 01:37:58 -08001669 EXT3_SB(sb)->s_itb_per_group) ||
Mingming Cao1c2bf372006-06-25 05:48:06 -07001670 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
Aneesh Kumar K.Vfeda58d2008-02-06 01:40:13 -08001671 EXT3_SB(sb)->s_itb_per_group)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 ext3_error(sb, "ext3_new_block",
1673 "Allocating block in system zone - "
Mingming Cao1c2bf372006-06-25 05:48:06 -07001674 "blocks from "E3FSBLK", length %lu",
1675 ret_block, num);
Aneesh Kumar K.V2588ef82008-04-28 02:16:14 -07001676 /*
1677 * claim_block() marked the blocks we allocated as in use. So we
1678 * may want to selectively mark some of the blocks as free.
1679 */
1680 goto retry_alloc;
Aneesh Kumar K.Vfeda58d2008-02-06 01:40:13 -08001681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
1683 performed_allocation = 1;
1684
1685#ifdef CONFIG_JBD_DEBUG
1686 {
1687 struct buffer_head *debug_bh;
1688
1689 /* Record bitmap buffer state in the newly allocated block */
Mingming Cao1c2bf372006-06-25 05:48:06 -07001690 debug_bh = sb_find_get_block(sb, ret_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 if (debug_bh) {
1692 BUFFER_TRACE(debug_bh, "state when allocated");
1693 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1694 brelse(debug_bh);
1695 }
1696 }
1697 jbd_lock_bh_state(bitmap_bh);
1698 spin_lock(sb_bgl_lock(sbi, group_no));
1699 if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
Mingming Caofaa56972006-03-26 01:37:58 -08001700 int i;
1701
1702 for (i = 0; i < num; i++) {
Mingming Cao1c2bf372006-06-25 05:48:06 -07001703 if (ext3_test_bit(grp_alloc_blk+i,
Mingming Caofaa56972006-03-26 01:37:58 -08001704 bh2jh(bitmap_bh)->b_committed_data)) {
1705 printk("%s: block was unexpectedly set in "
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001706 "b_committed_data\n", __func__);
Mingming Caofaa56972006-03-26 01:37:58 -08001707 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 }
1709 }
Mingming Cao1c2bf372006-06-25 05:48:06 -07001710 ext3_debug("found bit %d\n", grp_alloc_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 spin_unlock(sb_bgl_lock(sbi, group_no));
1712 jbd_unlock_bh_state(bitmap_bh);
1713#endif
1714
Mingming Caofaa56972006-03-26 01:37:58 -08001715 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 ext3_error(sb, "ext3_new_block",
Mingming Cao1c2bf372006-06-25 05:48:06 -07001717 "block("E3FSBLK") >= blocks count(%d) - "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 "block_group = %d, es == %p ", ret_block,
1719 le32_to_cpu(es->s_blocks_count), group_no, es);
1720 goto out;
1721 }
1722
1723 /*
1724 * It is up to the caller to add the new buffer to a journal
1725 * list of some description. We don't know in advance whether
1726 * the caller wants to use it as metadata or data.
1727 */
Mingming Cao1c2bf372006-06-25 05:48:06 -07001728 ext3_debug("allocating block %lu. Goal hits %d of %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 ret_block, goal_hits, goal_attempts);
1730
1731 spin_lock(sb_bgl_lock(sbi, group_no));
Marcin Slusarz50e8a282008-02-08 04:20:13 -08001732 le16_add_cpu(&gdp->bg_free_blocks_count, -num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 spin_unlock(sb_bgl_lock(sbi, group_no));
Peter Zijlstra3cb4f9f2007-10-16 23:25:42 -07001734 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
1736 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1737 err = ext3_journal_dirty_metadata(handle, gdp_bh);
1738 if (!fatal)
1739 fatal = err;
1740
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 if (fatal)
1742 goto out;
1743
1744 *errp = 0;
1745 brelse(bitmap_bh);
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001746 dquot_free_block(inode, *count-num);
Mingming Caob54e41e2006-03-26 01:37:57 -08001747 *count = num;
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001748
1749 trace_ext3_allocate_blocks(inode, goal, num,
1750 (unsigned long long)ret_block);
1751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 return ret_block;
1753
1754io_error:
1755 *errp = -EIO;
1756out:
1757 if (fatal) {
1758 *errp = fatal;
1759 ext3_std_error(sb, fatal);
1760 }
1761 /*
1762 * Undo the block allocation
1763 */
1764 if (!performed_allocation)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001765 dquot_free_block(inode, *count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 brelse(bitmap_bh);
1767 return 0;
1768}
1769
Mingming Cao1c2bf372006-06-25 05:48:06 -07001770ext3_fsblk_t ext3_new_block(handle_t *handle, struct inode *inode,
1771 ext3_fsblk_t goal, int *errp)
Mingming Caob54e41e2006-03-26 01:37:57 -08001772{
1773 unsigned long count = 1;
1774
1775 return ext3_new_blocks(handle, inode, goal, &count, errp);
1776}
1777
Mingming Cao36faadc142006-09-27 01:49:32 -07001778/**
1779 * ext3_count_free_blocks() -- count filesystem free blocks
1780 * @sb: superblock
1781 *
1782 * Adds up the number of free blocks from each block group.
1783 */
Mingming Cao43d23f92006-06-25 05:48:07 -07001784ext3_fsblk_t ext3_count_free_blocks(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785{
Mingming Cao43d23f92006-06-25 05:48:07 -07001786 ext3_fsblk_t desc_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 struct ext3_group_desc *gdp;
1788 int i;
Glauber de Oliveira Costa8bdac5d2005-09-22 21:44:26 -07001789 unsigned long ngroups = EXT3_SB(sb)->s_groups_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790#ifdef EXT3FS_DEBUG
1791 struct ext3_super_block *es;
Mingming Cao43d23f92006-06-25 05:48:07 -07001792 ext3_fsblk_t bitmap_count;
1793 unsigned long x;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 struct buffer_head *bitmap_bh = NULL;
1795
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 es = EXT3_SB(sb)->s_es;
1797 desc_count = 0;
1798 bitmap_count = 0;
1799 gdp = NULL;
Glauber de Oliveira Costa8bdac5d2005-09-22 21:44:26 -07001800
Glauber de Oliveira Costa5b116872005-10-30 15:02:48 -08001801 smp_rmb();
Glauber de Oliveira Costa8bdac5d2005-09-22 21:44:26 -07001802 for (i = 0; i < ngroups; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 gdp = ext3_get_group_desc(sb, i, NULL);
1804 if (!gdp)
1805 continue;
1806 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1807 brelse(bitmap_bh);
1808 bitmap_bh = read_block_bitmap(sb, i);
1809 if (bitmap_bh == NULL)
1810 continue;
1811
1812 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1813 printk("group %d: stored = %d, counted = %lu\n",
1814 i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1815 bitmap_count += x;
1816 }
1817 brelse(bitmap_bh);
Mingming Cao43d23f92006-06-25 05:48:07 -07001818 printk("ext3_count_free_blocks: stored = "E3FSBLK
1819 ", computed = "E3FSBLK", "E3FSBLK"\n",
1820 le32_to_cpu(es->s_free_blocks_count),
1821 desc_count, bitmap_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 return bitmap_count;
1823#else
1824 desc_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 smp_rmb();
1826 for (i = 0; i < ngroups; i++) {
1827 gdp = ext3_get_group_desc(sb, i, NULL);
1828 if (!gdp)
1829 continue;
1830 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1831 }
1832
1833 return desc_count;
1834#endif
1835}
1836
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837static inline int test_root(int a, int b)
1838{
1839 int num = b;
1840
1841 while (a > num)
1842 num *= b;
1843 return num == a;
1844}
1845
1846static int ext3_group_sparse(int group)
1847{
1848 if (group <= 1)
1849 return 1;
1850 if (!(group & 1))
1851 return 0;
1852 return (test_root(group, 7) || test_root(group, 5) ||
1853 test_root(group, 3));
1854}
1855
1856/**
1857 * ext3_bg_has_super - number of blocks used by the superblock in group
1858 * @sb: superblock for filesystem
1859 * @group: group number to check
1860 *
1861 * Return the number of blocks used by the superblock (primary or backup)
1862 * in this group. Currently this will be only 0 or 1.
1863 */
1864int ext3_bg_has_super(struct super_block *sb, int group)
1865{
Glauber de Oliveira Costab5a7c4f2006-03-24 03:18:37 -08001866 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1867 EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1868 !ext3_group_sparse(group))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 return 0;
1870 return 1;
1871}
1872
Glauber de Oliveira Costab5a7c4f2006-03-24 03:18:37 -08001873static unsigned long ext3_bg_num_gdb_meta(struct super_block *sb, int group)
1874{
1875 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1876 unsigned long first = metagroup * EXT3_DESC_PER_BLOCK(sb);
1877 unsigned long last = first + EXT3_DESC_PER_BLOCK(sb) - 1;
1878
1879 if (group == first || group == first + 1 || group == last)
1880 return 1;
1881 return 0;
1882}
1883
1884static unsigned long ext3_bg_num_gdb_nometa(struct super_block *sb, int group)
1885{
Akinobu Mita859cb932008-02-06 01:40:17 -08001886 return ext3_bg_has_super(sb, group) ? EXT3_SB(sb)->s_gdb_count : 0;
Glauber de Oliveira Costab5a7c4f2006-03-24 03:18:37 -08001887}
1888
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889/**
1890 * ext3_bg_num_gdb - number of blocks used by the group table in group
1891 * @sb: superblock for filesystem
1892 * @group: group number to check
1893 *
1894 * Return the number of blocks used by the group descriptor table
1895 * (primary or backup) in this group. In the future there may be a
1896 * different number of descriptor blocks in each group.
1897 */
1898unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1899{
Glauber de Oliveira Costab5a7c4f2006-03-24 03:18:37 -08001900 unsigned long first_meta_bg =
1901 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_meta_bg);
1902 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
Glauber de Oliveira Costab5a7c4f2006-03-24 03:18:37 -08001904 if (!EXT3_HAS_INCOMPAT_FEATURE(sb,EXT3_FEATURE_INCOMPAT_META_BG) ||
1905 metagroup < first_meta_bg)
1906 return ext3_bg_num_gdb_nometa(sb,group);
1907
1908 return ext3_bg_num_gdb_meta(sb,group);
1909
1910}
Lukas Czernerb853b962010-11-22 12:29:17 +01001911
1912/**
1913 * ext3_trim_all_free -- function to trim all free space in alloc. group
1914 * @sb: super block for file system
1915 * @group: allocation group to trim
1916 * @start: first group block to examine
1917 * @max: last group block to examine
1918 * @gdp: allocation group description structure
1919 * @minblocks: minimum extent block count
1920 *
1921 * ext3_trim_all_free walks through group's block bitmap searching for free
1922 * blocks. When the free block is found, it tries to allocate this block and
1923 * consequent free block to get the biggest free extent possible, until it
1924 * reaches any used block. Then issue a TRIM command on this extent and free
1925 * the extent in the block bitmap. This is done until whole group is scanned.
1926 */
1927ext3_grpblk_t ext3_trim_all_free(struct super_block *sb, unsigned int group,
1928 ext3_grpblk_t start, ext3_grpblk_t max,
1929 ext3_grpblk_t minblocks)
1930{
1931 handle_t *handle;
1932 ext3_grpblk_t next, free_blocks, bit, freed, count = 0;
1933 ext3_fsblk_t discard_block;
1934 struct ext3_sb_info *sbi;
1935 struct buffer_head *gdp_bh, *bitmap_bh = NULL;
1936 struct ext3_group_desc *gdp;
1937 int err = 0, ret = 0;
1938
1939 /*
1940 * We will update one block bitmap, and one group descriptor
1941 */
1942 handle = ext3_journal_start_sb(sb, 2);
1943 if (IS_ERR(handle))
1944 return PTR_ERR(handle);
1945
1946 bitmap_bh = read_block_bitmap(sb, group);
1947 if (!bitmap_bh) {
1948 err = -EIO;
1949 goto err_out;
1950 }
1951
1952 BUFFER_TRACE(bitmap_bh, "getting undo access");
1953 err = ext3_journal_get_undo_access(handle, bitmap_bh);
1954 if (err)
1955 goto err_out;
1956
1957 gdp = ext3_get_group_desc(sb, group, &gdp_bh);
1958 if (!gdp) {
1959 err = -EIO;
1960 goto err_out;
1961 }
1962
1963 BUFFER_TRACE(gdp_bh, "get_write_access");
1964 err = ext3_journal_get_write_access(handle, gdp_bh);
1965 if (err)
1966 goto err_out;
1967
1968 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1969 sbi = EXT3_SB(sb);
1970
1971 /* Walk through the whole group */
1972 while (start < max) {
1973 start = bitmap_search_next_usable_block(start, bitmap_bh, max);
1974 if (start < 0)
1975 break;
1976 next = start;
1977
1978 /*
1979 * Allocate contiguous free extents by setting bits in the
1980 * block bitmap
1981 */
1982 while (next < max
1983 && claim_block(sb_bgl_lock(sbi, group),
1984 next, bitmap_bh)) {
1985 next++;
1986 }
1987
1988 /* We did not claim any blocks */
1989 if (next == start)
1990 continue;
1991
1992 discard_block = (ext3_fsblk_t)start +
1993 ext3_group_first_block_no(sb, group);
1994
1995 /* Update counters */
1996 spin_lock(sb_bgl_lock(sbi, group));
1997 le16_add_cpu(&gdp->bg_free_blocks_count, start - next);
1998 spin_unlock(sb_bgl_lock(sbi, group));
1999 percpu_counter_sub(&sbi->s_freeblocks_counter, next - start);
2000
Tao Mabbac7512011-01-21 22:09:20 +08002001 free_blocks -= next - start;
Lukas Czernerb853b962010-11-22 12:29:17 +01002002 /* Do not issue a TRIM on extents smaller than minblocks */
2003 if ((next - start) < minblocks)
2004 goto free_extent;
2005
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002006 trace_ext3_discard_blocks(sb, discard_block, next - start);
Lukas Czernerb853b962010-11-22 12:29:17 +01002007 /* Send the TRIM command down to the device */
2008 err = sb_issue_discard(sb, discard_block, next - start,
2009 GFP_NOFS, 0);
2010 count += (next - start);
2011free_extent:
2012 freed = 0;
2013
2014 /*
2015 * Clear bits in the bitmap
2016 */
2017 for (bit = start; bit < next; bit++) {
2018 BUFFER_TRACE(bitmap_bh, "clear bit");
2019 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, group),
2020 bit, bitmap_bh->b_data)) {
2021 ext3_error(sb, __func__,
2022 "bit already cleared for block "E3FSBLK,
2023 (unsigned long)bit);
2024 BUFFER_TRACE(bitmap_bh, "bit already cleared");
2025 } else {
2026 freed++;
2027 }
2028 }
2029
2030 /* Update couters */
2031 spin_lock(sb_bgl_lock(sbi, group));
2032 le16_add_cpu(&gdp->bg_free_blocks_count, freed);
2033 spin_unlock(sb_bgl_lock(sbi, group));
2034 percpu_counter_add(&sbi->s_freeblocks_counter, freed);
2035
2036 start = next;
2037 if (err < 0) {
2038 if (err != -EOPNOTSUPP)
2039 ext3_warning(sb, __func__, "Discard command "
2040 "returned error %d\n", err);
2041 break;
2042 }
2043
2044 if (fatal_signal_pending(current)) {
2045 err = -ERESTARTSYS;
2046 break;
2047 }
2048
2049 cond_resched();
2050
2051 /* No more suitable extents */
Tao Mabbac7512011-01-21 22:09:20 +08002052 if (free_blocks < minblocks)
Lukas Czernerb853b962010-11-22 12:29:17 +01002053 break;
2054 }
2055
2056 /* We dirtied the bitmap block */
2057 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
2058 ret = ext3_journal_dirty_metadata(handle, bitmap_bh);
2059 if (!err)
2060 err = ret;
2061
2062 /* And the group descriptor block */
2063 BUFFER_TRACE(gdp_bh, "dirtied group descriptor block");
2064 ret = ext3_journal_dirty_metadata(handle, gdp_bh);
2065 if (!err)
2066 err = ret;
2067
2068 ext3_debug("trimmed %d blocks in the group %d\n",
2069 count, group);
2070
2071err_out:
2072 if (err)
2073 count = err;
2074 ext3_journal_stop(handle);
2075 brelse(bitmap_bh);
2076
2077 return count;
2078}
2079
2080/**
2081 * ext3_trim_fs() -- trim ioctl handle function
2082 * @sb: superblock for filesystem
2083 * @start: First Byte to trim
2084 * @len: number of Bytes to trim from start
2085 * @minlen: minimum extent length in Bytes
2086 *
2087 * ext3_trim_fs goes through all allocation groups containing Bytes from
2088 * start to start+len. For each such a group ext3_trim_all_free function
2089 * is invoked to trim all free space.
2090 */
2091int ext3_trim_fs(struct super_block *sb, struct fstrim_range *range)
2092{
2093 ext3_grpblk_t last_block, first_block, free_blocks;
2094 unsigned long first_group, last_group;
2095 unsigned long group, ngroups;
2096 struct ext3_group_desc *gdp;
2097 struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2098 uint64_t start, len, minlen, trimmed;
2099 ext3_fsblk_t max_blks = le32_to_cpu(es->s_blocks_count);
2100 int ret = 0;
2101
Tao Ma4b44dd32011-01-21 10:52:56 +08002102 start = (range->start >> sb->s_blocksize_bits) +
2103 le32_to_cpu(es->s_first_data_block);
Lukas Czernerb853b962010-11-22 12:29:17 +01002104 len = range->len >> sb->s_blocksize_bits;
2105 minlen = range->minlen >> sb->s_blocksize_bits;
2106 trimmed = 0;
2107
2108 if (unlikely(minlen > EXT3_BLOCKS_PER_GROUP(sb)))
2109 return -EINVAL;
2110 if (start >= max_blks)
Lukas Czerner2c2ea942011-06-22 10:51:09 +02002111 return -EINVAL;
Lukas Czernerb853b962010-11-22 12:29:17 +01002112 if (start + len > max_blks)
2113 len = max_blks - start;
2114
2115 ngroups = EXT3_SB(sb)->s_groups_count;
2116 smp_rmb();
2117
2118 /* Determine first and last group to examine based on start and len */
2119 ext3_get_group_no_and_offset(sb, (ext3_fsblk_t) start,
2120 &first_group, &first_block);
2121 ext3_get_group_no_and_offset(sb, (ext3_fsblk_t) (start + len),
2122 &last_group, &last_block);
2123 last_group = (last_group > ngroups - 1) ? ngroups - 1 : last_group;
2124 last_block = EXT3_BLOCKS_PER_GROUP(sb);
2125
2126 if (first_group > last_group)
2127 return -EINVAL;
2128
2129 for (group = first_group; group <= last_group; group++) {
2130 gdp = ext3_get_group_desc(sb, group, NULL);
2131 if (!gdp)
2132 break;
2133
2134 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
2135 if (free_blocks < minlen)
2136 continue;
2137
Tao Ma425fa412011-03-03 22:58:37 +08002138 /*
2139 * For all the groups except the last one, last block will
2140 * always be EXT3_BLOCKS_PER_GROUP(sb), so we only need to
2141 * change it for the last group in which case first_block +
2142 * len < EXT3_BLOCKS_PER_GROUP(sb).
2143 */
2144 if (first_block + len < EXT3_BLOCKS_PER_GROUP(sb))
Lukas Czernerb853b962010-11-22 12:29:17 +01002145 last_block = first_block + len;
Tao Ma425fa412011-03-03 22:58:37 +08002146 len -= last_block - first_block;
Lukas Czernerb853b962010-11-22 12:29:17 +01002147
2148 ret = ext3_trim_all_free(sb, group, first_block,
2149 last_block, minlen);
2150 if (ret < 0)
2151 break;
2152
2153 trimmed += ret;
2154 first_block = 0;
2155 }
2156
2157 if (ret >= 0)
2158 ret = 0;
Lukas Czernerb853b962010-11-22 12:29:17 +01002159 range->len = trimmed * sb->s_blocksize;
2160
2161 return ret;
2162}