blob: 22548f56197b52cf06b343285b7e7beb7d74cb10 [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/quotaops.h>
Lukas Czernerb853b962010-11-22 12:29:17 +010015#include <linux/blkdev.h>
Al Viro4613ad12012-03-29 22:30:07 -040016#include "ext3.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070017
18/*
19 * balloc.c contains the blocks allocation and deallocation routines
20 */
21
22/*
23 * The free blocks are managed by bitmaps. A file system contains several
24 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
25 * block for inodes, N blocks for the inode table and data blocks.
26 *
27 * The file system contains group descriptors which are located after the
28 * super block. Each descriptor contains the number of the bitmap block and
29 * the free blocks count in the block. The descriptors are loaded in memory
Aneesh Kumar K.Ve6274322007-02-20 13:57:58 -080030 * when a file system is mounted (see ext3_fill_super).
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 */
32
33
34#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
35
Lukas Czernerb853b962010-11-22 12:29:17 +010036/*
37 * Calculate the block group number and offset, given a block number
38 */
39static void ext3_get_group_no_and_offset(struct super_block *sb,
40 ext3_fsblk_t blocknr, unsigned long *blockgrpp, ext3_grpblk_t *offsetp)
41{
42 struct ext3_super_block *es = EXT3_SB(sb)->s_es;
43
44 blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
45 if (offsetp)
46 *offsetp = blocknr % EXT3_BLOCKS_PER_GROUP(sb);
47 if (blockgrpp)
48 *blockgrpp = blocknr / EXT3_BLOCKS_PER_GROUP(sb);
49}
50
Mingming Cao36faadc142006-09-27 01:49:32 -070051/**
52 * ext3_get_group_desc() -- load group descriptor from disk
Dave Kleikampe9ad5622006-09-27 01:49:35 -070053 * @sb: super block
Mingming Cao36faadc142006-09-27 01:49:32 -070054 * @block_group: given block group
55 * @bh: pointer to the buffer head to store the block
56 * group descriptor
57 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070058struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb,
59 unsigned int block_group,
60 struct buffer_head ** bh)
61{
62 unsigned long group_desc;
63 unsigned long offset;
64 struct ext3_group_desc * desc;
65 struct ext3_sb_info *sbi = EXT3_SB(sb);
66
67 if (block_group >= sbi->s_groups_count) {
68 ext3_error (sb, "ext3_get_group_desc",
69 "block_group >= groups_count - "
70 "block_group = %d, groups_count = %lu",
71 block_group, sbi->s_groups_count);
72
73 return NULL;
74 }
75 smp_rmb();
76
77 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
78 offset = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
79 if (!sbi->s_group_desc[group_desc]) {
80 ext3_error (sb, "ext3_get_group_desc",
81 "Group descriptor not loaded - "
82 "block_group = %d, group_desc = %lu, desc = %lu",
83 block_group, group_desc, offset);
84 return NULL;
85 }
86
87 desc = (struct ext3_group_desc *) sbi->s_group_desc[group_desc]->b_data;
88 if (bh)
89 *bh = sbi->s_group_desc[group_desc];
90 return desc + offset;
91}
92
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -080093static int ext3_valid_block_bitmap(struct super_block *sb,
94 struct ext3_group_desc *desc,
95 unsigned int block_group,
96 struct buffer_head *bh)
97{
98 ext3_grpblk_t offset;
99 ext3_grpblk_t next_zero_bit;
100 ext3_fsblk_t bitmap_blk;
101 ext3_fsblk_t group_first_block;
102
103 group_first_block = ext3_group_first_block_no(sb, block_group);
104
105 /* check whether block bitmap block number is set */
106 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
107 offset = bitmap_blk - group_first_block;
108 if (!ext3_test_bit(offset, bh->b_data))
109 /* bad block bitmap */
110 goto err_out;
111
112 /* check whether the inode bitmap block number is set */
113 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
114 offset = bitmap_blk - group_first_block;
115 if (!ext3_test_bit(offset, bh->b_data))
116 /* bad block bitmap */
117 goto err_out;
118
119 /* check whether the inode table block number is set */
120 bitmap_blk = le32_to_cpu(desc->bg_inode_table);
121 offset = bitmap_blk - group_first_block;
122 next_zero_bit = ext3_find_next_zero_bit(bh->b_data,
123 offset + EXT3_SB(sb)->s_itb_per_group,
124 offset);
125 if (next_zero_bit >= offset + EXT3_SB(sb)->s_itb_per_group)
126 /* good bitmap for inode tables */
127 return 1;
128
129err_out:
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700130 ext3_error(sb, __func__,
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800131 "Invalid block bitmap - "
132 "block_group = %d, block = %lu",
133 block_group, bitmap_blk);
134 return 0;
135}
136
Mingming Cao36faadc142006-09-27 01:49:32 -0700137/**
138 * read_block_bitmap()
139 * @sb: super block
140 * @block_group: given block group
141 *
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800142 * Read the bitmap for a given block_group,and validate the
143 * bits for block/inode/inode tables are set in the bitmaps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 *
145 * Return buffer_head on success or NULL in case of failure.
146 */
147static struct buffer_head *
148read_block_bitmap(struct super_block *sb, unsigned int block_group)
149{
150 struct ext3_group_desc * desc;
151 struct buffer_head * bh = NULL;
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800152 ext3_fsblk_t bitmap_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800154 desc = ext3_get_group_desc(sb, block_group, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (!desc)
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800156 return NULL;
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200157 trace_ext3_read_block_bitmap(sb, block_group);
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800158 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
159 bh = sb_getblk(sb, bitmap_blk);
160 if (unlikely(!bh)) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700161 ext3_error(sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 "Cannot read block bitmap - "
163 "block_group = %d, block_bitmap = %u",
164 block_group, le32_to_cpu(desc->bg_block_bitmap));
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800165 return NULL;
166 }
167 if (likely(bh_uptodate_or_lock(bh)))
168 return bh;
169
170 if (bh_submit_read(bh) < 0) {
171 brelse(bh);
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700172 ext3_error(sb, __func__,
Aneesh Kumar K.Vf762e902008-02-06 01:40:09 -0800173 "Cannot read block bitmap - "
174 "block_group = %d, block_bitmap = %u",
175 block_group, le32_to_cpu(desc->bg_block_bitmap));
176 return NULL;
177 }
Aneesh Kumar K.V2588ef82008-04-28 02:16:14 -0700178 ext3_valid_block_bitmap(sb, desc, block_group, bh);
179 /*
180 * file system mounted not to panic on error, continue with corrupt
181 * bitmap
182 */
Linus Torvalds0b832a42007-11-13 08:07:31 -0800183 return bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184}
185/*
186 * The reservation window structure operations
187 * --------------------------------------------
188 * Operations include:
189 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
190 *
Mingming Cao36faadc142006-09-27 01:49:32 -0700191 * We use a red-black tree to represent per-filesystem reservation
192 * windows.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 *
Mingming Cao36faadc142006-09-27 01:49:32 -0700194 */
195
196/**
197 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
198 * @rb_root: root of per-filesystem reservation rb tree
199 * @verbose: verbose mode
200 * @fn: function which wishes to dump the reservation map
201 *
202 * If verbose is turned on, it will print the whole block reservation
203 * windows(start, end). Otherwise, it will only print out the "bad" windows,
204 * those windows that overlap with their immediate neighbors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 */
Mingming Cao321fb9e2006-09-27 01:49:32 -0700206#if 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207static void __rsv_window_dump(struct rb_root *root, int verbose,
208 const char *fn)
209{
210 struct rb_node *n;
211 struct ext3_reserve_window_node *rsv, *prev;
212 int bad;
213
214restart:
215 n = rb_first(root);
216 bad = 0;
217 prev = NULL;
218
219 printk("Block Allocation Reservation Windows Map (%s):\n", fn);
220 while (n) {
Hugh Dickinsc56d2562006-12-06 20:41:27 -0800221 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 if (verbose)
223 printk("reservation window 0x%p "
Mingming Cao321fb9e2006-09-27 01:49:32 -0700224 "start: %lu, end: %lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 rsv, rsv->rsv_start, rsv->rsv_end);
226 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
227 printk("Bad reservation %p (start >= end)\n",
228 rsv);
229 bad = 1;
230 }
231 if (prev && prev->rsv_end >= rsv->rsv_start) {
232 printk("Bad reservation %p (prev->end >= start)\n",
233 rsv);
234 bad = 1;
235 }
236 if (bad) {
237 if (!verbose) {
238 printk("Restarting reservation walk in verbose mode\n");
239 verbose = 1;
240 goto restart;
241 }
242 }
243 n = rb_next(n);
244 prev = rsv;
245 }
246 printk("Window map complete.\n");
Julia Lawall269b2612008-04-28 02:16:09 -0700247 BUG_ON(bad);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
249#define rsv_window_dump(root, verbose) \
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700250 __rsv_window_dump((root), (verbose), __func__)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251#else
252#define rsv_window_dump(root, verbose) do {} while (0)
253#endif
254
Mingming Cao36faadc142006-09-27 01:49:32 -0700255/**
256 * goal_in_my_reservation()
257 * @rsv: inode's reservation window
258 * @grp_goal: given goal block relative to the allocation block group
259 * @group: the current allocation block group
260 * @sb: filesystem super block
261 *
262 * Test if the given goal block (group relative) is within the file's
263 * own block reservation window range.
264 *
265 * If the reservation window is outside the goal allocation group, return 0;
266 * grp_goal (given goal block) could be -1, which means no specific
267 * goal block. In this case, always return 1.
268 * If the goal block is within the reservation window, return 1;
269 * otherwise, return 0;
270 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271static int
Mingming Cao1c2bf372006-06-25 05:48:06 -0700272goal_in_my_reservation(struct ext3_reserve_window *rsv, ext3_grpblk_t grp_goal,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 unsigned int group, struct super_block * sb)
274{
Mingming Cao1c2bf372006-06-25 05:48:06 -0700275 ext3_fsblk_t group_first_block, group_last_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
Mingming Cao43d23f92006-06-25 05:48:07 -0700277 group_first_block = ext3_group_first_block_no(sb, group);
Eric Sandeen32c2d2b2006-09-27 01:49:36 -0700278 group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
280 if ((rsv->_rsv_start > group_last_block) ||
281 (rsv->_rsv_end < group_first_block))
282 return 0;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700283 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
284 || (grp_goal + group_first_block > rsv->_rsv_end)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 return 0;
286 return 1;
287}
288
Mingming Cao36faadc142006-09-27 01:49:32 -0700289/**
290 * search_reserve_window()
291 * @rb_root: root of reservation tree
292 * @goal: target allocation block
293 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 * Find the reserved window which includes the goal, or the previous one
295 * if the goal is not in any window.
296 * Returns NULL if there are no windows or if all windows start after the goal.
297 */
298static struct ext3_reserve_window_node *
Mingming Cao1c2bf372006-06-25 05:48:06 -0700299search_reserve_window(struct rb_root *root, ext3_fsblk_t goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300{
301 struct rb_node *n = root->rb_node;
302 struct ext3_reserve_window_node *rsv;
303
304 if (!n)
305 return NULL;
306
307 do {
308 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
309
310 if (goal < rsv->rsv_start)
311 n = n->rb_left;
312 else if (goal > rsv->rsv_end)
313 n = n->rb_right;
314 else
315 return rsv;
316 } while (n);
317 /*
318 * We've fallen off the end of the tree: the goal wasn't inside
319 * any particular node. OK, the previous node must be to one
320 * side of the interval containing the goal. If it's the RHS,
321 * we need to back up one.
322 */
323 if (rsv->rsv_start > goal) {
324 n = rb_prev(&rsv->rsv_node);
325 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
326 }
327 return rsv;
328}
329
Mingming Cao36faadc142006-09-27 01:49:32 -0700330/**
331 * ext3_rsv_window_add() -- Insert a window to the block reservation rb tree.
332 * @sb: super block
333 * @rsv: reservation window to add
334 *
335 * Must be called with rsv_lock hold.
336 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337void ext3_rsv_window_add(struct super_block *sb,
338 struct ext3_reserve_window_node *rsv)
339{
340 struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
341 struct rb_node *node = &rsv->rsv_node;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700342 ext3_fsblk_t start = rsv->rsv_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
344 struct rb_node ** p = &root->rb_node;
345 struct rb_node * parent = NULL;
346 struct ext3_reserve_window_node *this;
347
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200348 trace_ext3_rsv_window_add(sb, rsv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 while (*p)
350 {
351 parent = *p;
352 this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
353
354 if (start < this->rsv_start)
355 p = &(*p)->rb_left;
356 else if (start > this->rsv_end)
357 p = &(*p)->rb_right;
Mingming Cao321fb9e2006-09-27 01:49:32 -0700358 else {
359 rsv_window_dump(root, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 BUG();
Mingming Cao321fb9e2006-09-27 01:49:32 -0700361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 }
363
364 rb_link_node(node, parent, p);
365 rb_insert_color(node, root);
366}
367
Mingming Cao36faadc142006-09-27 01:49:32 -0700368/**
369 * ext3_rsv_window_remove() -- unlink a window from the reservation rb tree
370 * @sb: super block
371 * @rsv: reservation window to remove
372 *
373 * Mark the block reservation window as not allocated, and unlink it
374 * from the filesystem reservation window rb tree. Must be called with
375 * rsv_lock hold.
376 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377static void rsv_window_remove(struct super_block *sb,
378 struct ext3_reserve_window_node *rsv)
379{
380 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
381 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
382 rsv->rsv_alloc_hit = 0;
383 rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
384}
385
Mingming Cao36faadc142006-09-27 01:49:32 -0700386/*
387 * rsv_is_empty() -- Check if the reservation window is allocated.
388 * @rsv: given reservation window to check
389 *
390 * returns 1 if the end block is EXT3_RESERVE_WINDOW_NOT_ALLOCATED.
391 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
393{
394 /* a valid reservation end block could not be 0 */
Mingming Cao36faadc142006-09-27 01:49:32 -0700395 return rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
Mingming Cao36faadc142006-09-27 01:49:32 -0700397
398/**
399 * ext3_init_block_alloc_info()
400 * @inode: file inode structure
401 *
402 * Allocate and initialize the reservation window structure, and
403 * link the window to the ext3 inode structure at last
404 *
405 * The reservation window structure is only dynamically allocated
406 * and linked to ext3 inode the first time the open file
407 * needs a new block. So, before every ext3_new_block(s) call, for
408 * regular files, we should check whether the reservation window
409 * structure exists or not. In the latter case, this function is called.
410 * Fail to do so will result in block reservation being turned off for that
411 * open file.
412 *
413 * This function is called from ext3_get_blocks_handle(), also called
414 * when setting the reservation window size through ioctl before the file
415 * is open for write (needs block allocation).
416 *
417 * Needs truncate_mutex protection prior to call this function.
418 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419void ext3_init_block_alloc_info(struct inode *inode)
420{
421 struct ext3_inode_info *ei = EXT3_I(inode);
Julia Lawall46130222011-08-04 12:29:31 +0200422 struct ext3_block_alloc_info *block_i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 struct super_block *sb = inode->i_sb;
424
425 block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
426 if (block_i) {
427 struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node;
428
429 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
430 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
431
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700432 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 * if filesystem is mounted with NORESERVATION, the goal
434 * reservation window size is set to zero to indicate
435 * block reservation is off
436 */
437 if (!test_opt(sb, RESERVATION))
438 rsv->rsv_goal_size = 0;
439 else
440 rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS;
441 rsv->rsv_alloc_hit = 0;
442 block_i->last_alloc_logical_block = 0;
443 block_i->last_alloc_physical_block = 0;
444 }
445 ei->i_block_alloc_info = block_i;
446}
447
Mingming Cao36faadc142006-09-27 01:49:32 -0700448/**
449 * ext3_discard_reservation()
450 * @inode: inode
451 *
452 * Discard(free) block reservation window on last file close, or truncate
453 * or at last iput().
454 *
455 * It is being called in three cases:
456 * ext3_release_file(): last writer close the file
457 * ext3_clear_inode(): last iput(), when nobody link to this file.
458 * ext3_truncate(): when the block indirect map is about to change.
459 *
460 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461void ext3_discard_reservation(struct inode *inode)
462{
463 struct ext3_inode_info *ei = EXT3_I(inode);
464 struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
465 struct ext3_reserve_window_node *rsv;
466 spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
467
468 if (!block_i)
469 return;
470
471 rsv = &block_i->rsv_window_node;
472 if (!rsv_is_empty(&rsv->rsv_window)) {
473 spin_lock(rsv_lock);
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200474 if (!rsv_is_empty(&rsv->rsv_window)) {
475 trace_ext3_discard_reservation(inode, rsv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 rsv_window_remove(inode->i_sb, rsv);
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200477 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 spin_unlock(rsv_lock);
479 }
480}
481
Mingming Cao36faadc142006-09-27 01:49:32 -0700482/**
483 * ext3_free_blocks_sb() -- Free given blocks and update quota
484 * @handle: handle to this transaction
485 * @sb: super block
Anatol Pomozov4907cb72012-09-01 10:31:09 -0700486 * @block: start physical block to free
Mingming Cao36faadc142006-09-27 01:49:32 -0700487 * @count: number of blocks to free
488 * @pdquot_freed_blocks: pointer to quota
489 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
Mingming Cao1c2bf372006-06-25 05:48:06 -0700491 ext3_fsblk_t block, unsigned long count,
492 unsigned long *pdquot_freed_blocks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
494 struct buffer_head *bitmap_bh = NULL;
495 struct buffer_head *gd_bh;
496 unsigned long block_group;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700497 ext3_grpblk_t bit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 unsigned long i;
499 unsigned long overflow;
500 struct ext3_group_desc * desc;
501 struct ext3_super_block * es;
502 struct ext3_sb_info *sbi;
503 int err = 0, ret;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700504 ext3_grpblk_t group_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
506 *pdquot_freed_blocks = 0;
507 sbi = EXT3_SB(sb);
508 es = sbi->s_es;
509 if (block < le32_to_cpu(es->s_first_data_block) ||
510 block + count < block ||
511 block + count > le32_to_cpu(es->s_blocks_count)) {
512 ext3_error (sb, "ext3_free_blocks",
513 "Freeing blocks not in datazone - "
Mingming Cao1c2bf372006-06-25 05:48:06 -0700514 "block = "E3FSBLK", count = %lu", block, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 goto error_return;
516 }
517
518 ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
519
520do_more:
521 overflow = 0;
522 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
523 EXT3_BLOCKS_PER_GROUP(sb);
524 bit = (block - le32_to_cpu(es->s_first_data_block)) %
525 EXT3_BLOCKS_PER_GROUP(sb);
526 /*
527 * Check to see if we are freeing blocks across a group
528 * boundary.
529 */
530 if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
531 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
532 count -= overflow;
533 }
534 brelse(bitmap_bh);
535 bitmap_bh = read_block_bitmap(sb, block_group);
536 if (!bitmap_bh)
537 goto error_return;
538 desc = ext3_get_group_desc (sb, block_group, &gd_bh);
539 if (!desc)
540 goto error_return;
541
542 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
543 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
544 in_range (block, le32_to_cpu(desc->bg_inode_table),
545 sbi->s_itb_per_group) ||
546 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
Aneesh Kumar K.Vfeda58d2008-02-06 01:40:13 -0800547 sbi->s_itb_per_group)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 ext3_error (sb, "ext3_free_blocks",
549 "Freeing blocks in system zones - "
Mingming Cao1c2bf372006-06-25 05:48:06 -0700550 "Block = "E3FSBLK", count = %lu",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 block, count);
Aneesh Kumar K.Vfeda58d2008-02-06 01:40:13 -0800552 goto error_return;
553 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555 /*
556 * We are about to start releasing blocks in the bitmap,
557 * so we need undo access.
558 */
559 /* @@@ check errors */
560 BUFFER_TRACE(bitmap_bh, "getting undo access");
561 err = ext3_journal_get_undo_access(handle, bitmap_bh);
562 if (err)
563 goto error_return;
564
565 /*
566 * We are about to modify some metadata. Call the journal APIs
567 * to unshare ->b_data if a currently-committing transaction is
568 * using it
569 */
570 BUFFER_TRACE(gd_bh, "get_write_access");
571 err = ext3_journal_get_write_access(handle, gd_bh);
572 if (err)
573 goto error_return;
574
575 jbd_lock_bh_state(bitmap_bh);
576
577 for (i = 0, group_freed = 0; i < count; i++) {
578 /*
579 * An HJ special. This is expensive...
580 */
581#ifdef CONFIG_JBD_DEBUG
582 jbd_unlock_bh_state(bitmap_bh);
583 {
584 struct buffer_head *debug_bh;
585 debug_bh = sb_find_get_block(sb, block + i);
586 if (debug_bh) {
587 BUFFER_TRACE(debug_bh, "Deleted!");
588 if (!bh2jh(bitmap_bh)->b_committed_data)
589 BUFFER_TRACE(debug_bh,
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300590 "No committed data in bitmap");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
592 __brelse(debug_bh);
593 }
594 }
595 jbd_lock_bh_state(bitmap_bh);
596#endif
597 if (need_resched()) {
598 jbd_unlock_bh_state(bitmap_bh);
599 cond_resched();
600 jbd_lock_bh_state(bitmap_bh);
601 }
602 /* @@@ This prevents newly-allocated data from being
603 * freed and then reallocated within the same
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700604 * transaction.
605 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 * Ideally we would want to allow that to happen, but to
607 * do so requires making journal_forget() capable of
608 * revoking the queued write of a data block, which
609 * implies blocking on the journal lock. *forget()
610 * cannot block due to truncate races.
611 *
612 * Eventually we can fix this by making journal_forget()
613 * return a status indicating whether or not it was able
614 * to revoke the buffer. On successful revoke, it is
615 * safe not to set the allocation bit in the committed
616 * bitmap, because we know that there is no outstanding
617 * activity on the buffer any more and so it is safe to
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700618 * reallocate it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 */
620 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
621 J_ASSERT_BH(bitmap_bh,
622 bh2jh(bitmap_bh)->b_committed_data != NULL);
623 ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
624 bh2jh(bitmap_bh)->b_committed_data);
625
626 /*
627 * We clear the bit in the bitmap after setting the committed
628 * data bit, because this is the reverse order to that which
629 * the allocator uses.
630 */
631 BUFFER_TRACE(bitmap_bh, "clear bit");
632 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
633 bit + i, bitmap_bh->b_data)) {
634 jbd_unlock_bh_state(bitmap_bh);
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700635 ext3_error(sb, __func__,
Mingming Cao1c2bf372006-06-25 05:48:06 -0700636 "bit already cleared for block "E3FSBLK,
637 block + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 jbd_lock_bh_state(bitmap_bh);
639 BUFFER_TRACE(bitmap_bh, "bit already cleared");
640 } else {
641 group_freed++;
642 }
643 }
644 jbd_unlock_bh_state(bitmap_bh);
645
646 spin_lock(sb_bgl_lock(sbi, block_group));
Marcin Slusarz50e8a282008-02-08 04:20:13 -0800647 le16_add_cpu(&desc->bg_free_blocks_count, group_freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 spin_unlock(sb_bgl_lock(sbi, block_group));
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700649 percpu_counter_add(&sbi->s_freeblocks_counter, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
651 /* We dirtied the bitmap block */
652 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
653 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
654
655 /* And the group descriptor block */
656 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
657 ret = ext3_journal_dirty_metadata(handle, gd_bh);
658 if (!err) err = ret;
659 *pdquot_freed_blocks += group_freed;
660
661 if (overflow && !err) {
662 block += count;
663 count = overflow;
664 goto do_more;
665 }
Christoph Hellwigca41f7b2009-04-27 09:46:42 -0400666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667error_return:
668 brelse(bitmap_bh);
669 ext3_std_error(sb, err);
670 return;
671}
672
Mingming Cao36faadc142006-09-27 01:49:32 -0700673/**
674 * ext3_free_blocks() -- Free given blocks and update quota
675 * @handle: handle for this transaction
676 * @inode: inode
677 * @block: start physical block to free
678 * @count: number of blocks to count
679 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680void ext3_free_blocks(handle_t *handle, struct inode *inode,
Mingming Cao1c2bf372006-06-25 05:48:06 -0700681 ext3_fsblk_t block, unsigned long count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200683 struct super_block *sb = inode->i_sb;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700684 unsigned long dquot_freed_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200686 trace_ext3_free_blocks(inode, block, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
688 if (dquot_freed_blocks)
Christoph Hellwig5dd40562010-03-03 09:05:00 -0500689 dquot_free_block(inode, dquot_freed_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 return;
691}
692
Mingming Cao36faadc142006-09-27 01:49:32 -0700693/**
694 * ext3_test_allocatable()
695 * @nr: given allocation block group
696 * @bh: bufferhead contains the bitmap of the given block group
697 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 * For ext3 allocations, we must not reuse any blocks which are
699 * allocated in the bitmap buffer's "last committed data" copy. This
700 * prevents deletes from freeing up the page for reuse until we have
701 * committed the delete transaction.
702 *
703 * If we didn't do this, then deleting something and reallocating it as
704 * data would allow the old block to be overwritten before the
705 * transaction committed (because we force data to disk before commit).
706 * This would lead to corruption if we crashed between overwriting the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700707 * data and committing the delete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 *
709 * @@@ We may want to make this allocation behaviour conditional on
710 * data-writes at some point, and disable it for metadata allocations or
711 * sync-data inodes.
712 */
Mingming Cao1c2bf372006-06-25 05:48:06 -0700713static int ext3_test_allocatable(ext3_grpblk_t nr, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714{
715 int ret;
716 struct journal_head *jh = bh2jh(bh);
717
718 if (ext3_test_bit(nr, bh->b_data))
719 return 0;
720
721 jbd_lock_bh_state(bh);
722 if (!jh->b_committed_data)
723 ret = 1;
724 else
725 ret = !ext3_test_bit(nr, jh->b_committed_data);
726 jbd_unlock_bh_state(bh);
727 return ret;
728}
729
Mingming Cao36faadc142006-09-27 01:49:32 -0700730/**
731 * bitmap_search_next_usable_block()
732 * @start: the starting block (group relative) of the search
733 * @bh: bufferhead contains the block group bitmap
734 * @maxblocks: the ending block (group relative) of the reservation
735 *
736 * The bitmap search --- search forward alternately through the actual
737 * bitmap on disk and the last-committed copy in journal, until we find a
738 * bit free in both bitmaps.
739 */
Mingming Cao1c2bf372006-06-25 05:48:06 -0700740static ext3_grpblk_t
741bitmap_search_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
742 ext3_grpblk_t maxblocks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743{
Mingming Cao1c2bf372006-06-25 05:48:06 -0700744 ext3_grpblk_t next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 struct journal_head *jh = bh2jh(bh);
746
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 while (start < maxblocks) {
748 next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
749 if (next >= maxblocks)
750 return -1;
751 if (ext3_test_allocatable(next, bh))
752 return next;
753 jbd_lock_bh_state(bh);
754 if (jh->b_committed_data)
755 start = ext3_find_next_zero_bit(jh->b_committed_data,
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700756 maxblocks, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 jbd_unlock_bh_state(bh);
758 }
759 return -1;
760}
761
Mingming Cao36faadc142006-09-27 01:49:32 -0700762/**
763 * find_next_usable_block()
764 * @start: the starting block (group relative) to find next
765 * allocatable block in bitmap.
766 * @bh: bufferhead contains the block group bitmap
767 * @maxblocks: the ending block (group relative) for the search
768 *
769 * Find an allocatable block in a bitmap. We honor both the bitmap and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 * its last-committed copy (if that exists), and perform the "most
771 * appropriate allocation" algorithm of looking for a free block near
772 * the initial goal; then for a free byte somewhere in the bitmap; then
773 * for any free bit in the bitmap.
774 */
Mingming Cao1c2bf372006-06-25 05:48:06 -0700775static ext3_grpblk_t
776find_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
777 ext3_grpblk_t maxblocks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778{
Mingming Cao1c2bf372006-06-25 05:48:06 -0700779 ext3_grpblk_t here, next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 char *p, *r;
781
782 if (start > 0) {
783 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700784 * The goal was occupied; search forward for a free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 * block within the next XX blocks.
786 *
787 * end_goal is more or less random, but it has to be
788 * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
789 * next 64-bit boundary is simple..
790 */
Mingming Cao1c2bf372006-06-25 05:48:06 -0700791 ext3_grpblk_t end_goal = (start + 63) & ~63;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if (end_goal > maxblocks)
793 end_goal = maxblocks;
794 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
795 if (here < end_goal && ext3_test_allocatable(here, bh))
796 return here;
797 ext3_debug("Bit not found near goal\n");
798 }
799
800 here = start;
801 if (here < 0)
802 here = 0;
803
Namhyung Kim57e94d82010-10-11 02:10:45 +0900804 p = bh->b_data + (here >> 3);
Hugh Dickins7d1c5202006-12-06 20:41:30 -0800805 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
Namhyung Kim57e94d82010-10-11 02:10:45 +0900806 next = (r - bh->b_data) << 3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
808 if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
809 return next;
810
811 /*
812 * The bitmap search --- search forward alternately through the actual
813 * bitmap and the last-committed copy until we find a bit free in
814 * both
815 */
816 here = bitmap_search_next_usable_block(here, bh, maxblocks);
817 return here;
818}
819
Mingming Cao36faadc142006-09-27 01:49:32 -0700820/**
821 * claim_block()
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900822 * @lock: the spin lock for this block group
Mingming Cao36faadc142006-09-27 01:49:32 -0700823 * @block: the free block (group relative) to allocate
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900824 * @bh: the buffer_head contains the block group bitmap
Mingming Cao36faadc142006-09-27 01:49:32 -0700825 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 * We think we can allocate this block in this bitmap. Try to set the bit.
827 * If that succeeds then check that nobody has allocated and then freed the
828 * block since we saw that is was not marked in b_committed_data. If it _was_
829 * allocated and freed then clear the bit in the bitmap again and return
830 * zero (failure).
831 */
832static inline int
Mingming Cao1c2bf372006-06-25 05:48:06 -0700833claim_block(spinlock_t *lock, ext3_grpblk_t block, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834{
835 struct journal_head *jh = bh2jh(bh);
836 int ret;
837
838 if (ext3_set_bit_atomic(lock, block, bh->b_data))
839 return 0;
840 jbd_lock_bh_state(bh);
841 if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
842 ext3_clear_bit_atomic(lock, block, bh->b_data);
843 ret = 0;
844 } else {
845 ret = 1;
846 }
847 jbd_unlock_bh_state(bh);
848 return ret;
849}
850
Mingming Cao36faadc142006-09-27 01:49:32 -0700851/**
852 * ext3_try_to_allocate()
853 * @sb: superblock
854 * @handle: handle to this transaction
855 * @group: given allocation block group
856 * @bitmap_bh: bufferhead holds the block bitmap
857 * @grp_goal: given target block within the group
858 * @count: target number of blocks to allocate
859 * @my_rsv: reservation window
860 *
861 * Attempt to allocate blocks within a give range. Set the range of allocation
862 * first, then find the first free bit(s) from the bitmap (within the range),
863 * and at last, allocate the blocks by claiming the found free bit as allocated.
864 *
865 * To set the range of this allocation:
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700866 * if there is a reservation window, only try to allocate block(s) from the
Mingming Cao36faadc142006-09-27 01:49:32 -0700867 * file's own reservation window;
868 * Otherwise, the allocation range starts from the give goal block, ends at
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700869 * the block group's last block.
Mingming Cao36faadc142006-09-27 01:49:32 -0700870 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 * If we failed to allocate the desired block then we may end up crossing to a
872 * new bitmap. In that case we must release write access to the old one via
873 * ext3_journal_release_buffer(), else we'll run out of credits.
874 */
Mingming Cao1c2bf372006-06-25 05:48:06 -0700875static ext3_grpblk_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
Mingming Cao1c2bf372006-06-25 05:48:06 -0700877 struct buffer_head *bitmap_bh, ext3_grpblk_t grp_goal,
Mingming Caob54e41e2006-03-26 01:37:57 -0800878 unsigned long *count, struct ext3_reserve_window *my_rsv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879{
Mingming Cao1c2bf372006-06-25 05:48:06 -0700880 ext3_fsblk_t group_first_block;
881 ext3_grpblk_t start, end;
Mingming Caob54e41e2006-03-26 01:37:57 -0800882 unsigned long num = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
884 /* we do allocation within the reservation window if we have a window */
885 if (my_rsv) {
Mingming Cao43d23f92006-06-25 05:48:07 -0700886 group_first_block = ext3_group_first_block_no(sb, group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 if (my_rsv->_rsv_start >= group_first_block)
888 start = my_rsv->_rsv_start - group_first_block;
889 else
890 /* reservation window cross group boundary */
891 start = 0;
892 end = my_rsv->_rsv_end - group_first_block + 1;
893 if (end > EXT3_BLOCKS_PER_GROUP(sb))
894 /* reservation window crosses group boundary */
895 end = EXT3_BLOCKS_PER_GROUP(sb);
Mingming Cao1c2bf372006-06-25 05:48:06 -0700896 if ((start <= grp_goal) && (grp_goal < end))
897 start = grp_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 else
Mingming Cao1c2bf372006-06-25 05:48:06 -0700899 grp_goal = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 } else {
Mingming Cao1c2bf372006-06-25 05:48:06 -0700901 if (grp_goal > 0)
902 start = grp_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 else
904 start = 0;
905 end = EXT3_BLOCKS_PER_GROUP(sb);
906 }
907
908 BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
909
910repeat:
Mingming Cao1c2bf372006-06-25 05:48:06 -0700911 if (grp_goal < 0 || !ext3_test_allocatable(grp_goal, bitmap_bh)) {
912 grp_goal = find_next_usable_block(start, bitmap_bh, end);
913 if (grp_goal < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 goto fail_access;
915 if (!my_rsv) {
916 int i;
917
Mingming Cao1c2bf372006-06-25 05:48:06 -0700918 for (i = 0; i < 7 && grp_goal > start &&
919 ext3_test_allocatable(grp_goal - 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 bitmap_bh);
Mingming Cao1c2bf372006-06-25 05:48:06 -0700921 i++, grp_goal--)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 ;
923 }
924 }
Mingming Cao1c2bf372006-06-25 05:48:06 -0700925 start = grp_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
Mingming Cao36faadc142006-09-27 01:49:32 -0700927 if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group),
928 grp_goal, bitmap_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 /*
930 * The block was allocated by another thread, or it was
931 * allocated and then freed by another thread
932 */
933 start++;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700934 grp_goal++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 if (start >= end)
936 goto fail_access;
937 goto repeat;
938 }
Mingming Caob54e41e2006-03-26 01:37:57 -0800939 num++;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700940 grp_goal++;
941 while (num < *count && grp_goal < end
942 && ext3_test_allocatable(grp_goal, bitmap_bh)
Mingming Cao36faadc142006-09-27 01:49:32 -0700943 && claim_block(sb_bgl_lock(EXT3_SB(sb), group),
944 grp_goal, bitmap_bh)) {
Mingming Caob54e41e2006-03-26 01:37:57 -0800945 num++;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700946 grp_goal++;
Mingming Caob54e41e2006-03-26 01:37:57 -0800947 }
948 *count = num;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700949 return grp_goal - num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950fail_access:
Mingming Caob54e41e2006-03-26 01:37:57 -0800951 *count = num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 return -1;
953}
954
955/**
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700956 * find_next_reservable_window():
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 * find a reservable space within the given range.
958 * It does not allocate the reservation window for now:
959 * alloc_new_reservation() will do the work later.
960 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700961 * @search_head: the head of the searching list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 * This is not necessarily the list head of the whole filesystem
963 *
964 * We have both head and start_block to assist the search
965 * for the reservable space. The list starts from head,
966 * but we will shift to the place where start_block is,
967 * then start from there, when looking for a reservable space.
968 *
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900969 * @my_rsv: the reservation window
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 *
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900971 * @sb: the super block
972 *
973 * @start_block: the first block we consider to start
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 * the real search from
975 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700976 * @last_block:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 * the maximum block number that our goal reservable space
978 * could start from. This is normally the last block in this
979 * group. The search will end when we found the start of next
980 * possible reservable space is out of this boundary.
981 * This could handle the cross boundary reservation window
982 * request.
983 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700984 * basically we search from the given range, rather than the whole
985 * reservation double linked list, (start_block, last_block)
986 * to find a free region that is of my size and has not
987 * been reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 */
Mingming Cao21fe3472005-06-28 20:45:16 -0700990static int find_next_reservable_window(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 struct ext3_reserve_window_node *search_head,
Mingming Cao21fe3472005-06-28 20:45:16 -0700992 struct ext3_reserve_window_node *my_rsv,
Mingming Cao1c2bf372006-06-25 05:48:06 -0700993 struct super_block * sb,
994 ext3_fsblk_t start_block,
995 ext3_fsblk_t last_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996{
997 struct rb_node *next;
998 struct ext3_reserve_window_node *rsv, *prev;
Mingming Cao1c2bf372006-06-25 05:48:06 -0700999 ext3_fsblk_t cur;
Mingming Cao21fe3472005-06-28 20:45:16 -07001000 int size = my_rsv->rsv_goal_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
1002 /* TODO: make the start of the reservation window byte-aligned */
1003 /* cur = *start_block & ~7;*/
Mingming Cao21fe3472005-06-28 20:45:16 -07001004 cur = start_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 rsv = search_head;
1006 if (!rsv)
Mingming Cao21fe3472005-06-28 20:45:16 -07001007 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 while (1) {
1010 if (cur <= rsv->rsv_end)
1011 cur = rsv->rsv_end + 1;
1012
1013 /* TODO?
1014 * in the case we could not find a reservable space
1015 * that is what is expected, during the re-search, we could
1016 * remember what's the largest reservable space we could have
1017 * and return that one.
1018 *
1019 * For now it will fail if we could not find the reservable
1020 * space with expected-size (or more)...
1021 */
1022 if (cur > last_block)
Mingming Cao21fe3472005-06-28 20:45:16 -07001023 return -1; /* fail */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025 prev = rsv;
1026 next = rb_next(&rsv->rsv_node);
Hugh Dickinsc56d2562006-12-06 20:41:27 -08001027 rsv = rb_entry(next,struct ext3_reserve_window_node,rsv_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029 /*
1030 * Reached the last reservation, we can just append to the
1031 * previous one.
1032 */
1033 if (!next)
1034 break;
1035
1036 if (cur + size <= rsv->rsv_start) {
1037 /*
1038 * Found a reserveable space big enough. We could
1039 * have a reservation across the group boundary here
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001040 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 break;
1042 }
1043 }
1044 /*
1045 * we come here either :
1046 * when we reach the end of the whole list,
1047 * and there is empty reservable space after last entry in the list.
1048 * append it to the end of the list.
1049 *
1050 * or we found one reservable space in the middle of the list,
1051 * return the reservation window that we could append to.
1052 * succeed.
1053 */
Mingming Cao21fe3472005-06-28 20:45:16 -07001054
1055 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
1056 rsv_window_remove(sb, my_rsv);
1057
1058 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001059 * Let's book the whole available window for now. We will check the
Mingming Cao21fe3472005-06-28 20:45:16 -07001060 * disk bitmap later and then, if there are free blocks then we adjust
1061 * the window size if it's larger than requested.
1062 * Otherwise, we will remove this node from the tree next time
1063 * call find_next_reservable_window.
1064 */
1065 my_rsv->rsv_start = cur;
1066 my_rsv->rsv_end = cur + size - 1;
1067 my_rsv->rsv_alloc_hit = 0;
1068
1069 if (prev != my_rsv)
1070 ext3_rsv_window_add(sb, my_rsv);
1071
1072 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
1074
1075/**
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001076 * alloc_new_reservation()--allocate a new reservation window
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 *
1078 * To make a new reservation, we search part of the filesystem
1079 * reservation list (the list that inside the group). We try to
1080 * allocate a new reservation window near the allocation goal,
1081 * or the beginning of the group, if there is no goal.
1082 *
1083 * We first find a reservable space after the goal, then from
1084 * there, we check the bitmap for the first free block after
1085 * it. If there is no free block until the end of group, then the
1086 * whole group is full, we failed. Otherwise, check if the free
1087 * block is inside the expected reservable space, if so, we
1088 * succeed.
1089 * If the first free block is outside the reservable space, then
1090 * start from the first free block, we search for next available
1091 * space, and go on.
1092 *
1093 * on succeed, a new reservation will be found and inserted into the list
1094 * It contains at least one free block, and it does not overlap with other
1095 * reservation windows.
1096 *
1097 * failed: we failed to find a reservation window in this group
1098 *
Namhyung Kima4c18ad2010-10-19 00:34:35 +09001099 * @my_rsv: the reservation window
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 *
Mingming Cao1c2bf372006-06-25 05:48:06 -07001101 * @grp_goal: The goal (group-relative). It is where the search for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 * free reservable space should start from.
Mingming Cao1c2bf372006-06-25 05:48:06 -07001103 * if we have a grp_goal(grp_goal >0 ), then start from there,
1104 * no grp_goal(grp_goal = -1), we start from the first block
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 * of the group.
1106 *
1107 * @sb: the super block
1108 * @group: the group we are trying to allocate in
1109 * @bitmap_bh: the block group block bitmap
Mingming Cao21fe3472005-06-28 20:45:16 -07001110 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 */
1112static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
Mingming Cao1c2bf372006-06-25 05:48:06 -07001113 ext3_grpblk_t grp_goal, struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 unsigned int group, struct buffer_head *bitmap_bh)
1115{
1116 struct ext3_reserve_window_node *search_head;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001117 ext3_fsblk_t group_first_block, group_end_block, start_block;
1118 ext3_grpblk_t first_free_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
1120 unsigned long size;
Mingming Cao21fe3472005-06-28 20:45:16 -07001121 int ret;
1122 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Mingming Cao43d23f92006-06-25 05:48:07 -07001124 group_first_block = ext3_group_first_block_no(sb, group);
Eric Sandeen32c2d2b2006-09-27 01:49:36 -07001125 group_end_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
Mingming Cao1c2bf372006-06-25 05:48:06 -07001127 if (grp_goal < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 start_block = group_first_block;
1129 else
Mingming Cao1c2bf372006-06-25 05:48:06 -07001130 start_block = grp_goal + group_first_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001132 trace_ext3_alloc_new_reservation(sb, start_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 size = my_rsv->rsv_goal_size;
Mingming Cao21fe3472005-06-28 20:45:16 -07001134
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 if (!rsv_is_empty(&my_rsv->rsv_window)) {
1136 /*
1137 * if the old reservation is cross group boundary
1138 * and if the goal is inside the old reservation window,
1139 * we will come here when we just failed to allocate from
1140 * the first part of the window. We still have another part
1141 * that belongs to the next group. In this case, there is no
1142 * point to discard our window and try to allocate a new one
1143 * in this group(which will fail). we should
1144 * keep the reservation window, just simply move on.
1145 *
1146 * Maybe we could shift the start block of the reservation
1147 * window to the first block of next group.
1148 */
1149
1150 if ((my_rsv->rsv_start <= group_end_block) &&
1151 (my_rsv->rsv_end > group_end_block) &&
1152 (start_block >= my_rsv->rsv_start))
1153 return -1;
1154
1155 if ((my_rsv->rsv_alloc_hit >
1156 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
1157 /*
Mingming Cao36faadc142006-09-27 01:49:32 -07001158 * if the previously allocation hit ratio is
1159 * greater than 1/2, then we double the size of
1160 * the reservation window the next time,
1161 * otherwise we keep the same size window
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 */
1163 size = size * 2;
1164 if (size > EXT3_MAX_RESERVE_BLOCKS)
1165 size = EXT3_MAX_RESERVE_BLOCKS;
1166 my_rsv->rsv_goal_size= size;
1167 }
1168 }
Mingming Cao21fe3472005-06-28 20:45:16 -07001169
1170 spin_lock(rsv_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 /*
1172 * shift the search start to the window near the goal block
1173 */
1174 search_head = search_reserve_window(fs_rsv_root, start_block);
1175
1176 /*
1177 * find_next_reservable_window() simply finds a reservable window
1178 * inside the given range(start_block, group_end_block).
1179 *
1180 * To make sure the reservation window has a free bit inside it, we
1181 * need to check the bitmap after we found a reservable window.
1182 */
1183retry:
Mingming Cao21fe3472005-06-28 20:45:16 -07001184 ret = find_next_reservable_window(search_head, my_rsv, sb,
1185 start_block, group_end_block);
1186
1187 if (ret == -1) {
1188 if (!rsv_is_empty(&my_rsv->rsv_window))
1189 rsv_window_remove(sb, my_rsv);
1190 spin_unlock(rsv_lock);
1191 return -1;
1192 }
1193
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 /*
1195 * On success, find_next_reservable_window() returns the
1196 * reservation window where there is a reservable space after it.
1197 * Before we reserve this reservable space, we need
1198 * to make sure there is at least a free block inside this region.
1199 *
1200 * searching the first free bit on the block bitmap and copy of
1201 * last committed bitmap alternatively, until we found a allocatable
1202 * block. Search start from the start block of the reservable space
1203 * we just found.
1204 */
Mingming Cao21fe3472005-06-28 20:45:16 -07001205 spin_unlock(rsv_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 first_free_block = bitmap_search_next_usable_block(
Mingming Cao21fe3472005-06-28 20:45:16 -07001207 my_rsv->rsv_start - group_first_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 bitmap_bh, group_end_block - group_first_block + 1);
1209
1210 if (first_free_block < 0) {
1211 /*
1212 * no free block left on the bitmap, no point
1213 * to reserve the space. return failed.
1214 */
Mingming Cao21fe3472005-06-28 20:45:16 -07001215 spin_lock(rsv_lock);
1216 if (!rsv_is_empty(&my_rsv->rsv_window))
1217 rsv_window_remove(sb, my_rsv);
1218 spin_unlock(rsv_lock);
1219 return -1; /* failed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 }
Mingming Cao21fe3472005-06-28 20:45:16 -07001221
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 start_block = first_free_block + group_first_block;
1223 /*
1224 * check if the first free block is within the
Mingming Cao21fe3472005-06-28 20:45:16 -07001225 * free space we just reserved
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 */
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001227 if (start_block >= my_rsv->rsv_start &&
1228 start_block <= my_rsv->rsv_end) {
1229 trace_ext3_reserved(sb, start_block, my_rsv);
Mingming Cao21fe3472005-06-28 20:45:16 -07001230 return 0; /* success */
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001231 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 /*
1233 * if the first free bit we found is out of the reservable space
Mingming Cao21fe3472005-06-28 20:45:16 -07001234 * continue search for next reservable space,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 * start from where the free block is,
1236 * we also shift the list head to where we stopped last time
1237 */
Mingming Cao21fe3472005-06-28 20:45:16 -07001238 search_head = my_rsv;
1239 spin_lock(rsv_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241}
1242
Mingming Cao36faadc142006-09-27 01:49:32 -07001243/**
1244 * try_to_extend_reservation()
1245 * @my_rsv: given reservation window
1246 * @sb: super block
1247 * @size: the delta to extend
1248 *
1249 * Attempt to expand the reservation window large enough to have
1250 * required number of free blocks
1251 *
1252 * Since ext3_try_to_allocate() will always allocate blocks within
1253 * the reservation window range, if the window size is too small,
1254 * multiple blocks allocation has to stop at the end of the reservation
1255 * window. To make this more efficient, given the total number of
1256 * blocks needed and the current size of the window, we try to
1257 * expand the reservation window size if necessary on a best-effort
1258 * basis before ext3_new_blocks() tries to allocate blocks,
1259 */
Mingming Caod48589b2006-03-26 01:37:59 -08001260static void try_to_extend_reservation(struct ext3_reserve_window_node *my_rsv,
1261 struct super_block *sb, int size)
1262{
1263 struct ext3_reserve_window_node *next_rsv;
1264 struct rb_node *next;
1265 spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1266
1267 if (!spin_trylock(rsv_lock))
1268 return;
1269
1270 next = rb_next(&my_rsv->rsv_node);
1271
1272 if (!next)
1273 my_rsv->rsv_end += size;
1274 else {
Hugh Dickinsc56d2562006-12-06 20:41:27 -08001275 next_rsv = rb_entry(next, struct ext3_reserve_window_node, rsv_node);
Mingming Caod48589b2006-03-26 01:37:59 -08001276
1277 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1278 my_rsv->rsv_end += size;
1279 else
1280 my_rsv->rsv_end = next_rsv->rsv_start - 1;
1281 }
1282 spin_unlock(rsv_lock);
1283}
1284
Mingming Cao36faadc142006-09-27 01:49:32 -07001285/**
1286 * ext3_try_to_allocate_with_rsv()
1287 * @sb: superblock
1288 * @handle: handle to this transaction
1289 * @group: given allocation block group
1290 * @bitmap_bh: bufferhead holds the block bitmap
1291 * @grp_goal: given target block within the group
Mingming Cao36faadc142006-09-27 01:49:32 -07001292 * @my_rsv: reservation window
Namhyung Kima4c18ad2010-10-19 00:34:35 +09001293 * @count: target number of blocks to allocate
Mingming Cao36faadc142006-09-27 01:49:32 -07001294 * @errp: pointer to store the error code
1295 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 * This is the main function used to allocate a new block and its reservation
1297 * window.
1298 *
1299 * Each time when a new block allocation is need, first try to allocate from
1300 * its own reservation. If it does not have a reservation window, instead of
1301 * looking for a free bit on bitmap first, then look up the reservation list to
1302 * see if it is inside somebody else's reservation window, we try to allocate a
1303 * reservation window for it starting from the goal first. Then do the block
1304 * allocation within the reservation window.
1305 *
1306 * This will avoid keeping on searching the reservation list again and
Glauber de Oliveira Costa5b116872005-10-30 15:02:48 -08001307 * again when somebody is looking for a free block (without
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 * reservation), and there are lots of free blocks, but they are all
1309 * being reserved.
1310 *
Mingming Cao36faadc142006-09-27 01:49:32 -07001311 * We use a red-black tree for the per-filesystem reservation list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 *
1313 */
Mingming Cao1c2bf372006-06-25 05:48:06 -07001314static ext3_grpblk_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1316 unsigned int group, struct buffer_head *bitmap_bh,
Mingming Cao1c2bf372006-06-25 05:48:06 -07001317 ext3_grpblk_t grp_goal,
1318 struct ext3_reserve_window_node * my_rsv,
Mingming Caob54e41e2006-03-26 01:37:57 -08001319 unsigned long *count, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320{
Eric Sandeen32c2d2b2006-09-27 01:49:36 -07001321 ext3_fsblk_t group_first_block, group_last_block;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001322 ext3_grpblk_t ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 int fatal;
Mingming Caob54e41e2006-03-26 01:37:57 -08001324 unsigned long num = *count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
1326 *errp = 0;
1327
1328 /*
1329 * Make sure we use undo access for the bitmap, because it is critical
1330 * that we do the frozen_data COW on bitmap buffers in all cases even
1331 * if the buffer is in BJ_Forget state in the committing transaction.
1332 */
1333 BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1334 fatal = ext3_journal_get_undo_access(handle, bitmap_bh);
1335 if (fatal) {
1336 *errp = fatal;
1337 return -1;
1338 }
1339
1340 /*
1341 * we don't deal with reservation when
1342 * filesystem is mounted without reservation
1343 * or the file is not a regular file
1344 * or last attempt to allocate a block with reservation turned on failed
1345 */
1346 if (my_rsv == NULL ) {
Mingming Caob54e41e2006-03-26 01:37:57 -08001347 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
Mingming Cao1c2bf372006-06-25 05:48:06 -07001348 grp_goal, count, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 goto out;
1350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 /*
Mingming Cao1c2bf372006-06-25 05:48:06 -07001352 * grp_goal is a group relative block number (if there is a goal)
Hugh Dickins16502422006-12-06 20:41:25 -08001353 * 0 <= grp_goal < EXT3_BLOCKS_PER_GROUP(sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 * first block is a filesystem wide block number
1355 * first block is the block number of the first block in this group
1356 */
Mingming Cao43d23f92006-06-25 05:48:07 -07001357 group_first_block = ext3_group_first_block_no(sb, group);
Eric Sandeen32c2d2b2006-09-27 01:49:36 -07001358 group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
1360 /*
1361 * Basically we will allocate a new block from inode's reservation
1362 * window.
1363 *
1364 * We need to allocate a new reservation window, if:
1365 * a) inode does not have a reservation window; or
1366 * b) last attempt to allocate a block from existing reservation
1367 * failed; or
1368 * c) we come here with a goal and with a reservation window
1369 *
1370 * We do not need to allocate a new reservation window if we come here
1371 * at the beginning with a goal and the goal is inside the window, or
1372 * we don't have a goal but already have a reservation window.
1373 * then we could go to allocate from the reservation window directly.
1374 */
1375 while (1) {
Mingming Cao21fe3472005-06-28 20:45:16 -07001376 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
Mingming Cao36faadc142006-09-27 01:49:32 -07001377 !goal_in_my_reservation(&my_rsv->rsv_window,
1378 grp_goal, group, sb)) {
Mingming Caod48589b2006-03-26 01:37:59 -08001379 if (my_rsv->rsv_goal_size < *count)
1380 my_rsv->rsv_goal_size = *count;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001381 ret = alloc_new_reservation(my_rsv, grp_goal, sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 group, bitmap_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 if (ret < 0)
1384 break; /* failed */
1385
Mingming Cao36faadc142006-09-27 01:49:32 -07001386 if (!goal_in_my_reservation(&my_rsv->rsv_window,
1387 grp_goal, group, sb))
Mingming Cao1c2bf372006-06-25 05:48:06 -07001388 grp_goal = -1;
Hugh Dickins16502422006-12-06 20:41:25 -08001389 } else if (grp_goal >= 0) {
Mingming Cao2bd94bd2006-12-06 20:38:18 -08001390 int curr = my_rsv->rsv_end -
1391 (grp_goal + group_first_block) + 1;
1392
1393 if (curr < *count)
1394 try_to_extend_reservation(my_rsv, sb,
1395 *count - curr);
1396 }
Mingming Caod48589b2006-03-26 01:37:59 -08001397
Eric Sandeen32c2d2b2006-09-27 01:49:36 -07001398 if ((my_rsv->rsv_start > group_last_block) ||
1399 (my_rsv->rsv_end < group_first_block)) {
Mingming Cao321fb9e2006-09-27 01:49:32 -07001400 rsv_window_dump(&EXT3_SB(sb)->s_rsv_window_root, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 BUG();
Mingming Cao321fb9e2006-09-27 01:49:32 -07001402 }
Mingming Cao36faadc142006-09-27 01:49:32 -07001403 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1404 grp_goal, &num, &my_rsv->rsv_window);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 if (ret >= 0) {
Mingming Caob54e41e2006-03-26 01:37:57 -08001406 my_rsv->rsv_alloc_hit += num;
1407 *count = num;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 break; /* succeed */
1409 }
Mingming Caob54e41e2006-03-26 01:37:57 -08001410 num = *count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 }
1412out:
1413 if (ret >= 0) {
1414 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1415 "bitmap block");
1416 fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1417 if (fatal) {
1418 *errp = fatal;
1419 return -1;
1420 }
1421 return ret;
1422 }
1423
1424 BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1425 ext3_journal_release_buffer(handle, bitmap_bh);
1426 return ret;
1427}
1428
Mingming Cao36faadc142006-09-27 01:49:32 -07001429/**
1430 * ext3_has_free_blocks()
1431 * @sbi: in-core super block structure.
1432 *
1433 * Check if filesystem has at least 1 free block available for allocation.
1434 */
Dmitry Monakhov6360e212011-10-19 14:16:23 +04001435static int ext3_has_free_blocks(struct ext3_sb_info *sbi, int use_reservation)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436{
Mingming Cao1c2bf372006-06-25 05:48:06 -07001437 ext3_fsblk_t free_blocks, root_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
1439 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1440 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1441 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
Eric W. Biederman15232992012-02-07 15:41:24 -08001442 !use_reservation && !uid_eq(sbi->s_resuid, current_fsuid()) &&
1443 (gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1444 !in_group_p (sbi->s_resgid))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 return 0;
1446 }
1447 return 1;
1448}
1449
Mingming Cao36faadc142006-09-27 01:49:32 -07001450/**
1451 * ext3_should_retry_alloc()
1452 * @sb: super block
1453 * @retries number of attemps has been made
1454 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1456 * it is profitable to retry the operation, this function will wait
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001457 * for the current or committing transaction to complete, and then
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 * return TRUE.
Mingming Cao36faadc142006-09-27 01:49:32 -07001459 *
1460 * if the total number of retries exceed three times, return FALSE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 */
1462int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1463{
Dmitry Monakhov6360e212011-10-19 14:16:23 +04001464 if (!ext3_has_free_blocks(EXT3_SB(sb), 0) || (*retries)++ > 3)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 return 0;
1466
1467 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1468
1469 return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1470}
1471
Mingming Cao36faadc142006-09-27 01:49:32 -07001472/**
1473 * ext3_new_blocks() -- core block(s) allocation function
1474 * @handle: handle to this transaction
1475 * @inode: file inode
1476 * @goal: given target block(filesystem wide)
1477 * @count: target number of blocks to allocate
1478 * @errp: error code
1479 *
1480 * ext3_new_blocks uses a goal block to assist allocation. It tries to
1481 * allocate block(s) from the block group contains the goal block first. If that
1482 * fails, it will try to allocate block(s) from other block groups without
1483 * any specific goal block.
1484 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 */
Mingming Cao1c2bf372006-06-25 05:48:06 -07001486ext3_fsblk_t ext3_new_blocks(handle_t *handle, struct inode *inode,
1487 ext3_fsblk_t goal, unsigned long *count, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488{
1489 struct buffer_head *bitmap_bh = NULL;
1490 struct buffer_head *gdp_bh;
1491 int group_no;
1492 int goal_group;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001493 ext3_grpblk_t grp_target_blk; /* blockgroup relative goal block */
1494 ext3_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
1495 ext3_fsblk_t ret_block; /* filesyetem-wide allocated block */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 int bgi; /* blockgroup iteration index */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 int fatal = 0, err;
1498 int performed_allocation = 0;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001499 ext3_grpblk_t free_blocks; /* number of free blocks in a group */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 struct super_block *sb;
1501 struct ext3_group_desc *gdp;
1502 struct ext3_super_block *es;
1503 struct ext3_sb_info *sbi;
1504 struct ext3_reserve_window_node *my_rsv = NULL;
1505 struct ext3_block_alloc_info *block_i;
1506 unsigned short windowsz = 0;
1507#ifdef EXT3FS_DEBUG
1508 static int goal_hits, goal_attempts;
1509#endif
1510 unsigned long ngroups;
Mingming Caob54e41e2006-03-26 01:37:57 -08001511 unsigned long num = *count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
1513 *errp = -ENOSPC;
1514 sb = inode->i_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516 /*
1517 * Check quota for allocation of this block.
1518 */
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001519 err = dquot_alloc_block(inode, num);
1520 if (err) {
1521 *errp = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 return 0;
1523 }
1524
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001525 trace_ext3_request_blocks(inode, goal, num);
1526
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 sbi = EXT3_SB(sb);
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001528 es = sbi->s_es;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 ext3_debug("goal=%lu.\n", goal);
1530 /*
1531 * Allocate a block from reservation only when
1532 * filesystem is mounted with reservation(default,-o reservation), and
1533 * it's a regular file, and
1534 * the desired window size is greater than 0 (One could use ioctl
1535 * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1536 * reservation on that particular file)
1537 */
1538 block_i = EXT3_I(inode)->i_block_alloc_info;
1539 if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1540 my_rsv = &block_i->rsv_window_node;
1541
Dmitry Monakhov6360e212011-10-19 14:16:23 +04001542 if (!ext3_has_free_blocks(sbi, IS_NOQUOTA(inode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 *errp = -ENOSPC;
1544 goto out;
1545 }
1546
1547 /*
1548 * First, test whether the goal block is free.
1549 */
1550 if (goal < le32_to_cpu(es->s_first_data_block) ||
1551 goal >= le32_to_cpu(es->s_blocks_count))
1552 goal = le32_to_cpu(es->s_first_data_block);
1553 group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1554 EXT3_BLOCKS_PER_GROUP(sb);
Mingming Cao08fb306f2006-08-27 01:23:44 -07001555 goal_group = group_no;
1556retry_alloc:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1558 if (!gdp)
1559 goto io_error;
1560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1562 /*
1563 * if there is not enough free blocks to make a new resevation
1564 * turn off reservation for this allocation
1565 */
1566 if (my_rsv && (free_blocks < windowsz)
Mingming Cao46d01a22008-10-18 20:27:56 -07001567 && (free_blocks > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 && (rsv_is_empty(&my_rsv->rsv_window)))
1569 my_rsv = NULL;
1570
1571 if (free_blocks > 0) {
Mingming Cao1c2bf372006-06-25 05:48:06 -07001572 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 EXT3_BLOCKS_PER_GROUP(sb));
1574 bitmap_bh = read_block_bitmap(sb, group_no);
1575 if (!bitmap_bh)
1576 goto io_error;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001577 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1578 group_no, bitmap_bh, grp_target_blk,
1579 my_rsv, &num, &fatal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 if (fatal)
1581 goto out;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001582 if (grp_alloc_blk >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 goto allocated;
1584 }
1585
1586 ngroups = EXT3_SB(sb)->s_groups_count;
1587 smp_rmb();
1588
1589 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001590 * Now search the rest of the groups. We assume that
Akinobu Mita144704e2008-02-06 01:40:15 -08001591 * group_no and gdp correctly point to the last group visited.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 */
1593 for (bgi = 0; bgi < ngroups; bgi++) {
1594 group_no++;
1595 if (group_no >= ngroups)
1596 group_no = 0;
1597 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
Hugh Dickins2823b552006-12-06 20:41:28 -08001598 if (!gdp)
1599 goto io_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1601 /*
Frans van de Wiel8cef1072010-03-15 19:29:34 +01001602 * skip this group (and avoid loading bitmap) if there
1603 * are no free blocks
1604 */
1605 if (!free_blocks)
1606 continue;
1607 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 * skip this group if the number of
1609 * free blocks is less than half of the reservation
1610 * window size.
1611 */
Mingming Cao46d01a22008-10-18 20:27:56 -07001612 if (my_rsv && (free_blocks <= (windowsz/2)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 continue;
1614
1615 brelse(bitmap_bh);
1616 bitmap_bh = read_block_bitmap(sb, group_no);
1617 if (!bitmap_bh)
1618 goto io_error;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001619 /*
1620 * try to allocate block(s) from this group, without a goal(-1).
1621 */
1622 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1623 group_no, bitmap_bh, -1, my_rsv,
1624 &num, &fatal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 if (fatal)
1626 goto out;
Mingming Cao1c2bf372006-06-25 05:48:06 -07001627 if (grp_alloc_blk >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 goto allocated;
1629 }
1630 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001631 * We may end up a bogus earlier ENOSPC error due to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 * filesystem is "full" of reservations, but
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001633 * there maybe indeed free blocks available on disk
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 * In this case, we just forget about the reservations
1635 * just do block allocation as without reservations.
1636 */
1637 if (my_rsv) {
1638 my_rsv = NULL;
Hugh Dickinsef503672006-12-06 20:41:23 -08001639 windowsz = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 group_no = goal_group;
Mingming Cao08fb306f2006-08-27 01:23:44 -07001641 goto retry_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 }
1643 /* No space left on the device */
1644 *errp = -ENOSPC;
1645 goto out;
1646
1647allocated:
1648
1649 ext3_debug("using block group %d(%d)\n",
1650 group_no, gdp->bg_free_blocks_count);
1651
1652 BUFFER_TRACE(gdp_bh, "get_write_access");
1653 fatal = ext3_journal_get_write_access(handle, gdp_bh);
1654 if (fatal)
1655 goto out;
1656
Mingming Cao43d23f92006-06-25 05:48:07 -07001657 ret_block = grp_alloc_blk + ext3_group_first_block_no(sb, group_no);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
Mingming Cao1c2bf372006-06-25 05:48:06 -07001659 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1660 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1661 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
Mingming Caofaa56972006-03-26 01:37:58 -08001662 EXT3_SB(sb)->s_itb_per_group) ||
Mingming Cao1c2bf372006-06-25 05:48:06 -07001663 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
Aneesh Kumar K.Vfeda58d2008-02-06 01:40:13 -08001664 EXT3_SB(sb)->s_itb_per_group)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 ext3_error(sb, "ext3_new_block",
1666 "Allocating block in system zone - "
Mingming Cao1c2bf372006-06-25 05:48:06 -07001667 "blocks from "E3FSBLK", length %lu",
1668 ret_block, num);
Aneesh Kumar K.V2588ef82008-04-28 02:16:14 -07001669 /*
1670 * claim_block() marked the blocks we allocated as in use. So we
1671 * may want to selectively mark some of the blocks as free.
1672 */
1673 goto retry_alloc;
Aneesh Kumar K.Vfeda58d2008-02-06 01:40:13 -08001674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676 performed_allocation = 1;
1677
1678#ifdef CONFIG_JBD_DEBUG
1679 {
1680 struct buffer_head *debug_bh;
1681
1682 /* Record bitmap buffer state in the newly allocated block */
Mingming Cao1c2bf372006-06-25 05:48:06 -07001683 debug_bh = sb_find_get_block(sb, ret_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 if (debug_bh) {
1685 BUFFER_TRACE(debug_bh, "state when allocated");
1686 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1687 brelse(debug_bh);
1688 }
1689 }
1690 jbd_lock_bh_state(bitmap_bh);
1691 spin_lock(sb_bgl_lock(sbi, group_no));
1692 if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
Mingming Caofaa56972006-03-26 01:37:58 -08001693 int i;
1694
1695 for (i = 0; i < num; i++) {
Mingming Cao1c2bf372006-06-25 05:48:06 -07001696 if (ext3_test_bit(grp_alloc_blk+i,
Mingming Caofaa56972006-03-26 01:37:58 -08001697 bh2jh(bitmap_bh)->b_committed_data)) {
1698 printk("%s: block was unexpectedly set in "
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001699 "b_committed_data\n", __func__);
Mingming Caofaa56972006-03-26 01:37:58 -08001700 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 }
1702 }
Mingming Cao1c2bf372006-06-25 05:48:06 -07001703 ext3_debug("found bit %d\n", grp_alloc_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 spin_unlock(sb_bgl_lock(sbi, group_no));
1705 jbd_unlock_bh_state(bitmap_bh);
1706#endif
1707
Mingming Caofaa56972006-03-26 01:37:58 -08001708 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 ext3_error(sb, "ext3_new_block",
Mingming Cao1c2bf372006-06-25 05:48:06 -07001710 "block("E3FSBLK") >= blocks count(%d) - "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 "block_group = %d, es == %p ", ret_block,
1712 le32_to_cpu(es->s_blocks_count), group_no, es);
1713 goto out;
1714 }
1715
1716 /*
1717 * It is up to the caller to add the new buffer to a journal
1718 * list of some description. We don't know in advance whether
1719 * the caller wants to use it as metadata or data.
1720 */
Mingming Cao1c2bf372006-06-25 05:48:06 -07001721 ext3_debug("allocating block %lu. Goal hits %d of %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 ret_block, goal_hits, goal_attempts);
1723
1724 spin_lock(sb_bgl_lock(sbi, group_no));
Marcin Slusarz50e8a282008-02-08 04:20:13 -08001725 le16_add_cpu(&gdp->bg_free_blocks_count, -num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 spin_unlock(sb_bgl_lock(sbi, group_no));
Peter Zijlstra3cb4f9f2007-10-16 23:25:42 -07001727 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
1729 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1730 err = ext3_journal_dirty_metadata(handle, gdp_bh);
1731 if (!fatal)
1732 fatal = err;
1733
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 if (fatal)
1735 goto out;
1736
1737 *errp = 0;
1738 brelse(bitmap_bh);
Kazuya Mio053800a2012-01-30 17:41:25 +09001739
1740 if (num < *count) {
1741 dquot_free_block(inode, *count-num);
1742 *count = num;
1743 }
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001744
1745 trace_ext3_allocate_blocks(inode, goal, num,
1746 (unsigned long long)ret_block);
1747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 return ret_block;
1749
1750io_error:
1751 *errp = -EIO;
1752out:
1753 if (fatal) {
1754 *errp = fatal;
1755 ext3_std_error(sb, fatal);
1756 }
1757 /*
1758 * Undo the block allocation
1759 */
1760 if (!performed_allocation)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001761 dquot_free_block(inode, *count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 brelse(bitmap_bh);
1763 return 0;
1764}
1765
Mingming Cao1c2bf372006-06-25 05:48:06 -07001766ext3_fsblk_t ext3_new_block(handle_t *handle, struct inode *inode,
1767 ext3_fsblk_t goal, int *errp)
Mingming Caob54e41e2006-03-26 01:37:57 -08001768{
1769 unsigned long count = 1;
1770
1771 return ext3_new_blocks(handle, inode, goal, &count, errp);
1772}
1773
Mingming Cao36faadc142006-09-27 01:49:32 -07001774/**
1775 * ext3_count_free_blocks() -- count filesystem free blocks
1776 * @sb: superblock
1777 *
1778 * Adds up the number of free blocks from each block group.
1779 */
Mingming Cao43d23f92006-06-25 05:48:07 -07001780ext3_fsblk_t ext3_count_free_blocks(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781{
Mingming Cao43d23f92006-06-25 05:48:07 -07001782 ext3_fsblk_t desc_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 struct ext3_group_desc *gdp;
1784 int i;
Glauber de Oliveira Costa8bdac5d2005-09-22 21:44:26 -07001785 unsigned long ngroups = EXT3_SB(sb)->s_groups_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786#ifdef EXT3FS_DEBUG
1787 struct ext3_super_block *es;
Mingming Cao43d23f92006-06-25 05:48:07 -07001788 ext3_fsblk_t bitmap_count;
1789 unsigned long x;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 struct buffer_head *bitmap_bh = NULL;
1791
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 es = EXT3_SB(sb)->s_es;
1793 desc_count = 0;
1794 bitmap_count = 0;
1795 gdp = NULL;
Glauber de Oliveira Costa8bdac5d2005-09-22 21:44:26 -07001796
Glauber de Oliveira Costa5b116872005-10-30 15:02:48 -08001797 smp_rmb();
Glauber de Oliveira Costa8bdac5d2005-09-22 21:44:26 -07001798 for (i = 0; i < ngroups; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 gdp = ext3_get_group_desc(sb, i, NULL);
1800 if (!gdp)
1801 continue;
1802 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1803 brelse(bitmap_bh);
1804 bitmap_bh = read_block_bitmap(sb, i);
1805 if (bitmap_bh == NULL)
1806 continue;
1807
1808 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1809 printk("group %d: stored = %d, counted = %lu\n",
1810 i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1811 bitmap_count += x;
1812 }
1813 brelse(bitmap_bh);
Mingming Cao43d23f92006-06-25 05:48:07 -07001814 printk("ext3_count_free_blocks: stored = "E3FSBLK
1815 ", computed = "E3FSBLK", "E3FSBLK"\n",
Akinobu Mita10d47082012-07-30 14:41:06 -07001816 (ext3_fsblk_t)le32_to_cpu(es->s_free_blocks_count),
Mingming Cao43d23f92006-06-25 05:48:07 -07001817 desc_count, bitmap_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 return bitmap_count;
1819#else
1820 desc_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 smp_rmb();
1822 for (i = 0; i < ngroups; i++) {
1823 gdp = ext3_get_group_desc(sb, i, NULL);
1824 if (!gdp)
1825 continue;
1826 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1827 }
1828
1829 return desc_count;
1830#endif
1831}
1832
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833static inline int test_root(int a, int b)
1834{
1835 int num = b;
1836
1837 while (a > num)
1838 num *= b;
1839 return num == a;
1840}
1841
1842static int ext3_group_sparse(int group)
1843{
1844 if (group <= 1)
1845 return 1;
1846 if (!(group & 1))
1847 return 0;
1848 return (test_root(group, 7) || test_root(group, 5) ||
1849 test_root(group, 3));
1850}
1851
1852/**
1853 * ext3_bg_has_super - number of blocks used by the superblock in group
1854 * @sb: superblock for filesystem
1855 * @group: group number to check
1856 *
1857 * Return the number of blocks used by the superblock (primary or backup)
1858 * in this group. Currently this will be only 0 or 1.
1859 */
1860int ext3_bg_has_super(struct super_block *sb, int group)
1861{
Glauber de Oliveira Costab5a7c4f2006-03-24 03:18:37 -08001862 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1863 EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1864 !ext3_group_sparse(group))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 return 0;
1866 return 1;
1867}
1868
Glauber de Oliveira Costab5a7c4f2006-03-24 03:18:37 -08001869static unsigned long ext3_bg_num_gdb_meta(struct super_block *sb, int group)
1870{
1871 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1872 unsigned long first = metagroup * EXT3_DESC_PER_BLOCK(sb);
1873 unsigned long last = first + EXT3_DESC_PER_BLOCK(sb) - 1;
1874
1875 if (group == first || group == first + 1 || group == last)
1876 return 1;
1877 return 0;
1878}
1879
1880static unsigned long ext3_bg_num_gdb_nometa(struct super_block *sb, int group)
1881{
Akinobu Mita859cb932008-02-06 01:40:17 -08001882 return ext3_bg_has_super(sb, group) ? EXT3_SB(sb)->s_gdb_count : 0;
Glauber de Oliveira Costab5a7c4f2006-03-24 03:18:37 -08001883}
1884
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885/**
1886 * ext3_bg_num_gdb - number of blocks used by the group table in group
1887 * @sb: superblock for filesystem
1888 * @group: group number to check
1889 *
1890 * Return the number of blocks used by the group descriptor table
1891 * (primary or backup) in this group. In the future there may be a
1892 * different number of descriptor blocks in each group.
1893 */
1894unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1895{
Glauber de Oliveira Costab5a7c4f2006-03-24 03:18:37 -08001896 unsigned long first_meta_bg =
1897 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_meta_bg);
1898 unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899
Glauber de Oliveira Costab5a7c4f2006-03-24 03:18:37 -08001900 if (!EXT3_HAS_INCOMPAT_FEATURE(sb,EXT3_FEATURE_INCOMPAT_META_BG) ||
1901 metagroup < first_meta_bg)
1902 return ext3_bg_num_gdb_nometa(sb,group);
1903
1904 return ext3_bg_num_gdb_meta(sb,group);
1905
1906}
Lukas Czernerb853b962010-11-22 12:29:17 +01001907
1908/**
1909 * ext3_trim_all_free -- function to trim all free space in alloc. group
1910 * @sb: super block for file system
1911 * @group: allocation group to trim
1912 * @start: first group block to examine
1913 * @max: last group block to examine
1914 * @gdp: allocation group description structure
1915 * @minblocks: minimum extent block count
1916 *
1917 * ext3_trim_all_free walks through group's block bitmap searching for free
1918 * blocks. When the free block is found, it tries to allocate this block and
1919 * consequent free block to get the biggest free extent possible, until it
1920 * reaches any used block. Then issue a TRIM command on this extent and free
1921 * the extent in the block bitmap. This is done until whole group is scanned.
1922 */
H Hartley Sweeten3ee77f22011-09-23 13:55:52 -07001923static ext3_grpblk_t ext3_trim_all_free(struct super_block *sb,
1924 unsigned int group,
1925 ext3_grpblk_t start, ext3_grpblk_t max,
1926 ext3_grpblk_t minblocks)
Lukas Czernerb853b962010-11-22 12:29:17 +01001927{
1928 handle_t *handle;
1929 ext3_grpblk_t next, free_blocks, bit, freed, count = 0;
1930 ext3_fsblk_t discard_block;
1931 struct ext3_sb_info *sbi;
1932 struct buffer_head *gdp_bh, *bitmap_bh = NULL;
1933 struct ext3_group_desc *gdp;
1934 int err = 0, ret = 0;
1935
1936 /*
1937 * We will update one block bitmap, and one group descriptor
1938 */
1939 handle = ext3_journal_start_sb(sb, 2);
1940 if (IS_ERR(handle))
1941 return PTR_ERR(handle);
1942
1943 bitmap_bh = read_block_bitmap(sb, group);
1944 if (!bitmap_bh) {
1945 err = -EIO;
1946 goto err_out;
1947 }
1948
1949 BUFFER_TRACE(bitmap_bh, "getting undo access");
1950 err = ext3_journal_get_undo_access(handle, bitmap_bh);
1951 if (err)
1952 goto err_out;
1953
1954 gdp = ext3_get_group_desc(sb, group, &gdp_bh);
1955 if (!gdp) {
1956 err = -EIO;
1957 goto err_out;
1958 }
1959
1960 BUFFER_TRACE(gdp_bh, "get_write_access");
1961 err = ext3_journal_get_write_access(handle, gdp_bh);
1962 if (err)
1963 goto err_out;
1964
1965 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1966 sbi = EXT3_SB(sb);
1967
1968 /* Walk through the whole group */
Lukas Czernere703c202012-03-02 13:09:05 +01001969 while (start <= max) {
Lukas Czernerb853b962010-11-22 12:29:17 +01001970 start = bitmap_search_next_usable_block(start, bitmap_bh, max);
1971 if (start < 0)
1972 break;
1973 next = start;
1974
1975 /*
1976 * Allocate contiguous free extents by setting bits in the
1977 * block bitmap
1978 */
Lukas Czernere703c202012-03-02 13:09:05 +01001979 while (next <= max
Lukas Czernerb853b962010-11-22 12:29:17 +01001980 && claim_block(sb_bgl_lock(sbi, group),
1981 next, bitmap_bh)) {
1982 next++;
1983 }
1984
1985 /* We did not claim any blocks */
1986 if (next == start)
1987 continue;
1988
1989 discard_block = (ext3_fsblk_t)start +
1990 ext3_group_first_block_no(sb, group);
1991
1992 /* Update counters */
1993 spin_lock(sb_bgl_lock(sbi, group));
1994 le16_add_cpu(&gdp->bg_free_blocks_count, start - next);
1995 spin_unlock(sb_bgl_lock(sbi, group));
1996 percpu_counter_sub(&sbi->s_freeblocks_counter, next - start);
1997
Tao Mabbac7512011-01-21 22:09:20 +08001998 free_blocks -= next - start;
Lukas Czernerb853b962010-11-22 12:29:17 +01001999 /* Do not issue a TRIM on extents smaller than minblocks */
2000 if ((next - start) < minblocks)
2001 goto free_extent;
2002
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002003 trace_ext3_discard_blocks(sb, discard_block, next - start);
Lukas Czernerb853b962010-11-22 12:29:17 +01002004 /* Send the TRIM command down to the device */
2005 err = sb_issue_discard(sb, discard_block, next - start,
2006 GFP_NOFS, 0);
2007 count += (next - start);
2008free_extent:
2009 freed = 0;
2010
2011 /*
2012 * Clear bits in the bitmap
2013 */
2014 for (bit = start; bit < next; bit++) {
2015 BUFFER_TRACE(bitmap_bh, "clear bit");
2016 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, group),
2017 bit, bitmap_bh->b_data)) {
2018 ext3_error(sb, __func__,
2019 "bit already cleared for block "E3FSBLK,
2020 (unsigned long)bit);
2021 BUFFER_TRACE(bitmap_bh, "bit already cleared");
2022 } else {
2023 freed++;
2024 }
2025 }
2026
2027 /* Update couters */
2028 spin_lock(sb_bgl_lock(sbi, group));
2029 le16_add_cpu(&gdp->bg_free_blocks_count, freed);
2030 spin_unlock(sb_bgl_lock(sbi, group));
2031 percpu_counter_add(&sbi->s_freeblocks_counter, freed);
2032
2033 start = next;
2034 if (err < 0) {
2035 if (err != -EOPNOTSUPP)
2036 ext3_warning(sb, __func__, "Discard command "
2037 "returned error %d\n", err);
2038 break;
2039 }
2040
2041 if (fatal_signal_pending(current)) {
2042 err = -ERESTARTSYS;
2043 break;
2044 }
2045
2046 cond_resched();
2047
2048 /* No more suitable extents */
Tao Mabbac7512011-01-21 22:09:20 +08002049 if (free_blocks < minblocks)
Lukas Czernerb853b962010-11-22 12:29:17 +01002050 break;
2051 }
2052
2053 /* We dirtied the bitmap block */
2054 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
2055 ret = ext3_journal_dirty_metadata(handle, bitmap_bh);
2056 if (!err)
2057 err = ret;
2058
2059 /* And the group descriptor block */
2060 BUFFER_TRACE(gdp_bh, "dirtied group descriptor block");
2061 ret = ext3_journal_dirty_metadata(handle, gdp_bh);
2062 if (!err)
2063 err = ret;
2064
2065 ext3_debug("trimmed %d blocks in the group %d\n",
2066 count, group);
2067
2068err_out:
2069 if (err)
2070 count = err;
2071 ext3_journal_stop(handle);
2072 brelse(bitmap_bh);
2073
2074 return count;
2075}
2076
2077/**
2078 * ext3_trim_fs() -- trim ioctl handle function
2079 * @sb: superblock for filesystem
2080 * @start: First Byte to trim
2081 * @len: number of Bytes to trim from start
2082 * @minlen: minimum extent length in Bytes
2083 *
2084 * ext3_trim_fs goes through all allocation groups containing Bytes from
2085 * start to start+len. For each such a group ext3_trim_all_free function
2086 * is invoked to trim all free space.
2087 */
2088int ext3_trim_fs(struct super_block *sb, struct fstrim_range *range)
2089{
Lukas Czernere703c202012-03-02 13:09:05 +01002090 ext3_grpblk_t last_block, first_block;
2091 unsigned long group, first_group, last_group;
Lukas Czernerb853b962010-11-22 12:29:17 +01002092 struct ext3_group_desc *gdp;
2093 struct ext3_super_block *es = EXT3_SB(sb)->s_es;
Lukas Czernere703c202012-03-02 13:09:05 +01002094 uint64_t start, minlen, end, trimmed = 0;
2095 ext3_fsblk_t first_data_blk =
2096 le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block);
Lukas Czernerb853b962010-11-22 12:29:17 +01002097 ext3_fsblk_t max_blks = le32_to_cpu(es->s_blocks_count);
2098 int ret = 0;
2099
Lukas Czernere703c202012-03-02 13:09:05 +01002100 start = range->start >> sb->s_blocksize_bits;
2101 end = start + (range->len >> sb->s_blocksize_bits) - 1;
Lukas Czernerb853b962010-11-22 12:29:17 +01002102 minlen = range->minlen >> sb->s_blocksize_bits;
Lukas Czernerb853b962010-11-22 12:29:17 +01002103
Lukas Czernerae49eee2012-10-11 12:28:38 +02002104 if (minlen > EXT3_BLOCKS_PER_GROUP(sb) ||
2105 start >= max_blks ||
2106 range->len < sb->s_blocksize)
Lukas Czernerb853b962010-11-22 12:29:17 +01002107 return -EINVAL;
Lukas Czernere703c202012-03-02 13:09:05 +01002108 if (end >= max_blks)
2109 end = max_blks - 1;
2110 if (end <= first_data_blk)
2111 goto out;
2112 if (start < first_data_blk)
2113 start = first_data_blk;
Lukas Czernerb853b962010-11-22 12:29:17 +01002114
Lukas Czernerb853b962010-11-22 12:29:17 +01002115 smp_rmb();
2116
2117 /* Determine first and last group to examine based on start and len */
2118 ext3_get_group_no_and_offset(sb, (ext3_fsblk_t) start,
2119 &first_group, &first_block);
Lukas Czernere703c202012-03-02 13:09:05 +01002120 ext3_get_group_no_and_offset(sb, (ext3_fsblk_t) end,
Lukas Czernerb853b962010-11-22 12:29:17 +01002121 &last_group, &last_block);
Lukas Czernerb853b962010-11-22 12:29:17 +01002122
Lukas Czernere703c202012-03-02 13:09:05 +01002123 /* end now represents the last block to discard in this group */
2124 end = EXT3_BLOCKS_PER_GROUP(sb) - 1;
Lukas Czernerb853b962010-11-22 12:29:17 +01002125
2126 for (group = first_group; group <= last_group; group++) {
2127 gdp = ext3_get_group_desc(sb, group, NULL);
2128 if (!gdp)
2129 break;
2130
Tao Ma425fa412011-03-03 22:58:37 +08002131 /*
2132 * For all the groups except the last one, last block will
Lukas Czernere703c202012-03-02 13:09:05 +01002133 * always be EXT3_BLOCKS_PER_GROUP(sb)-1, so we only need to
2134 * change it for the last group, note that last_block is
2135 * already computed earlier by ext3_get_group_no_and_offset()
Tao Ma425fa412011-03-03 22:58:37 +08002136 */
Lukas Czernere703c202012-03-02 13:09:05 +01002137 if (group == last_group)
2138 end = last_block;
Lukas Czernerb853b962010-11-22 12:29:17 +01002139
Lukas Czernere703c202012-03-02 13:09:05 +01002140 if (le16_to_cpu(gdp->bg_free_blocks_count) >= minlen) {
2141 ret = ext3_trim_all_free(sb, group, first_block,
2142 end, minlen);
2143 if (ret < 0)
2144 break;
2145 trimmed += ret;
2146 }
Lukas Czernerb853b962010-11-22 12:29:17 +01002147
Lukas Czernere703c202012-03-02 13:09:05 +01002148 /*
2149 * For every group except the first one, we are sure
2150 * that the first block to discard will be block #0.
2151 */
Lukas Czernerb853b962010-11-22 12:29:17 +01002152 first_block = 0;
2153 }
2154
Lukas Czernere703c202012-03-02 13:09:05 +01002155 if (ret > 0)
Lukas Czernerb853b962010-11-22 12:29:17 +01002156 ret = 0;
Lukas Czernerb853b962010-11-22 12:29:17 +01002157
Lukas Czernere703c202012-03-02 13:09:05 +01002158out:
2159 range->len = trimmed * sb->s_blocksize;
Lukas Czernerb853b962010-11-22 12:29:17 +01002160 return ret;
2161}