Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/balloc.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993 | |
10 | * Big-endian to little-endian byte-swapping/bitmaps by | |
11 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
12 | */ | |
13 | ||
14 | #include <linux/time.h> | |
15 | #include <linux/capability.h> | |
16 | #include <linux/fs.h> | |
dab291af | 17 | #include <linux/jbd2.h> |
ac27a0ec DK |
18 | #include <linux/quotaops.h> |
19 | #include <linux/buffer_head.h> | |
3dcf5451 CH |
20 | #include "ext4.h" |
21 | #include "ext4_jbd2.h" | |
717d50e4 | 22 | #include "group.h" |
3dcf5451 | 23 | |
ac27a0ec DK |
24 | /* |
25 | * balloc.c contains the blocks allocation and deallocation routines | |
26 | */ | |
27 | ||
72b64b59 AM |
28 | /* |
29 | * Calculate the block group number and offset, given a block number | |
30 | */ | |
31 | void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr, | |
fd2d4291 | 32 | ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp) |
72b64b59 | 33 | { |
8c55e204 | 34 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
72b64b59 AM |
35 | ext4_grpblk_t offset; |
36 | ||
8c55e204 | 37 | blocknr = blocknr - le32_to_cpu(es->s_first_data_block); |
f4e5bc24 | 38 | offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)); |
72b64b59 AM |
39 | if (offsetp) |
40 | *offsetp = offset; | |
41 | if (blockgrpp) | |
8c55e204 | 42 | *blockgrpp = blocknr; |
72b64b59 AM |
43 | |
44 | } | |
45 | ||
0bf7e837 JS |
46 | static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block, |
47 | ext4_group_t block_group) | |
48 | { | |
49 | ext4_group_t actual_group; | |
7477827f | 50 | ext4_get_group_no_and_offset(sb, block, &actual_group, NULL); |
0bf7e837 JS |
51 | if (actual_group == block_group) |
52 | return 1; | |
53 | return 0; | |
54 | } | |
55 | ||
56 | static int ext4_group_used_meta_blocks(struct super_block *sb, | |
57 | ext4_group_t block_group) | |
58 | { | |
59 | ext4_fsblk_t tmp; | |
60 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
61 | /* block bitmap, inode bitmap, and inode table blocks */ | |
62 | int used_blocks = sbi->s_itb_per_group + 2; | |
63 | ||
64 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) { | |
65 | struct ext4_group_desc *gdp; | |
66 | struct buffer_head *bh; | |
67 | ||
68 | gdp = ext4_get_group_desc(sb, block_group, &bh); | |
69 | if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), | |
70 | block_group)) | |
71 | used_blocks--; | |
72 | ||
73 | if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), | |
74 | block_group)) | |
75 | used_blocks--; | |
76 | ||
77 | tmp = ext4_inode_table(sb, gdp); | |
78 | for (; tmp < ext4_inode_table(sb, gdp) + | |
79 | sbi->s_itb_per_group; tmp++) { | |
80 | if (!ext4_block_in_group(sb, tmp, block_group)) | |
81 | used_blocks -= 1; | |
82 | } | |
83 | } | |
84 | return used_blocks; | |
85 | } | |
717d50e4 AD |
86 | /* Initializes an uninitialized block bitmap if given, and returns the |
87 | * number of blocks free in the group. */ | |
88 | unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh, | |
fd2d4291 | 89 | ext4_group_t block_group, struct ext4_group_desc *gdp) |
717d50e4 | 90 | { |
717d50e4 AD |
91 | int bit, bit_max; |
92 | unsigned free_blocks, group_blocks; | |
93 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
94 | ||
95 | if (bh) { | |
96 | J_ASSERT_BH(bh, buffer_locked(bh)); | |
97 | ||
98 | /* If checksum is bad mark all blocks used to prevent allocation | |
99 | * essentially implementing a per-group read-only flag. */ | |
100 | if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) { | |
46e665e9 | 101 | ext4_error(sb, __func__, |
fd2d4291 | 102 | "Checksum bad for group %lu\n", block_group); |
717d50e4 AD |
103 | gdp->bg_free_blocks_count = 0; |
104 | gdp->bg_free_inodes_count = 0; | |
105 | gdp->bg_itable_unused = 0; | |
106 | memset(bh->b_data, 0xff, sb->s_blocksize); | |
107 | return 0; | |
108 | } | |
109 | memset(bh->b_data, 0, sb->s_blocksize); | |
110 | } | |
111 | ||
112 | /* Check for superblock and gdt backups in this group */ | |
113 | bit_max = ext4_bg_has_super(sb, block_group); | |
114 | ||
115 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) || | |
116 | block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) * | |
117 | sbi->s_desc_per_block) { | |
118 | if (bit_max) { | |
119 | bit_max += ext4_bg_num_gdb(sb, block_group); | |
120 | bit_max += | |
121 | le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); | |
122 | } | |
123 | } else { /* For META_BG_BLOCK_GROUPS */ | |
6afd6707 | 124 | bit_max += ext4_bg_num_gdb(sb, block_group); |
717d50e4 AD |
125 | } |
126 | ||
127 | if (block_group == sbi->s_groups_count - 1) { | |
128 | /* | |
129 | * Even though mke2fs always initialize first and last group | |
130 | * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need | |
131 | * to make sure we calculate the right free blocks | |
132 | */ | |
133 | group_blocks = ext4_blocks_count(sbi->s_es) - | |
134 | le32_to_cpu(sbi->s_es->s_first_data_block) - | |
af5bc92d | 135 | (EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count - 1)); |
717d50e4 AD |
136 | } else { |
137 | group_blocks = EXT4_BLOCKS_PER_GROUP(sb); | |
138 | } | |
139 | ||
140 | free_blocks = group_blocks - bit_max; | |
141 | ||
142 | if (bh) { | |
0bf7e837 JS |
143 | ext4_fsblk_t start, tmp; |
144 | int flex_bg = 0; | |
d00a6d7b | 145 | |
717d50e4 AD |
146 | for (bit = 0; bit < bit_max; bit++) |
147 | ext4_set_bit(bit, bh->b_data); | |
148 | ||
d00a6d7b | 149 | start = ext4_group_first_block_no(sb, block_group); |
717d50e4 | 150 | |
0bf7e837 JS |
151 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, |
152 | EXT4_FEATURE_INCOMPAT_FLEX_BG)) | |
153 | flex_bg = 1; | |
717d50e4 | 154 | |
0bf7e837 JS |
155 | /* Set bits for block and inode bitmaps, and inode table */ |
156 | tmp = ext4_block_bitmap(sb, gdp); | |
157 | if (!flex_bg || ext4_block_in_group(sb, tmp, block_group)) | |
158 | ext4_set_bit(tmp - start, bh->b_data); | |
159 | ||
160 | tmp = ext4_inode_bitmap(sb, gdp); | |
161 | if (!flex_bg || ext4_block_in_group(sb, tmp, block_group)) | |
162 | ext4_set_bit(tmp - start, bh->b_data); | |
163 | ||
164 | tmp = ext4_inode_table(sb, gdp); | |
165 | for (; tmp < ext4_inode_table(sb, gdp) + | |
166 | sbi->s_itb_per_group; tmp++) { | |
167 | if (!flex_bg || | |
168 | ext4_block_in_group(sb, tmp, block_group)) | |
169 | ext4_set_bit(tmp - start, bh->b_data); | |
170 | } | |
717d50e4 AD |
171 | /* |
172 | * Also if the number of blocks within the group is | |
173 | * less than the blocksize * 8 ( which is the size | |
174 | * of bitmap ), set rest of the block bitmap to 1 | |
175 | */ | |
176 | mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data); | |
177 | } | |
0bf7e837 | 178 | return free_blocks - ext4_group_used_meta_blocks(sb, block_group); |
717d50e4 AD |
179 | } |
180 | ||
181 | ||
ac27a0ec DK |
182 | /* |
183 | * The free blocks are managed by bitmaps. A file system contains several | |
184 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
185 | * block for inodes, N blocks for the inode table and data blocks. | |
186 | * | |
187 | * The file system contains group descriptors which are located after the | |
188 | * super block. Each descriptor contains the number of the bitmap block and | |
189 | * the free blocks count in the block. The descriptors are loaded in memory | |
e627432c | 190 | * when a file system is mounted (see ext4_fill_super). |
ac27a0ec DK |
191 | */ |
192 | ||
193 | ||
194 | #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1) | |
195 | ||
196 | /** | |
617ba13b | 197 | * ext4_get_group_desc() -- load group descriptor from disk |
ac27a0ec DK |
198 | * @sb: super block |
199 | * @block_group: given block group | |
200 | * @bh: pointer to the buffer head to store the block | |
201 | * group descriptor | |
202 | */ | |
af5bc92d | 203 | struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb, |
fd2d4291 | 204 | ext4_group_t block_group, |
af5bc92d | 205 | struct buffer_head **bh) |
ac27a0ec DK |
206 | { |
207 | unsigned long group_desc; | |
208 | unsigned long offset; | |
af5bc92d | 209 | struct ext4_group_desc *desc; |
617ba13b | 210 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
ac27a0ec DK |
211 | |
212 | if (block_group >= sbi->s_groups_count) { | |
af5bc92d TT |
213 | ext4_error(sb, "ext4_get_group_desc", |
214 | "block_group >= groups_count - " | |
215 | "block_group = %lu, groups_count = %lu", | |
216 | block_group, sbi->s_groups_count); | |
ac27a0ec DK |
217 | |
218 | return NULL; | |
219 | } | |
220 | smp_rmb(); | |
221 | ||
617ba13b MC |
222 | group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); |
223 | offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); | |
ac27a0ec | 224 | if (!sbi->s_group_desc[group_desc]) { |
af5bc92d TT |
225 | ext4_error(sb, "ext4_get_group_desc", |
226 | "Group descriptor not loaded - " | |
227 | "block_group = %lu, group_desc = %lu, desc = %lu", | |
228 | block_group, group_desc, offset); | |
ac27a0ec DK |
229 | return NULL; |
230 | } | |
231 | ||
0d1ee42f AR |
232 | desc = (struct ext4_group_desc *)( |
233 | (__u8 *)sbi->s_group_desc[group_desc]->b_data + | |
234 | offset * EXT4_DESC_SIZE(sb)); | |
ac27a0ec DK |
235 | if (bh) |
236 | *bh = sbi->s_group_desc[group_desc]; | |
0d1ee42f | 237 | return desc; |
ac27a0ec DK |
238 | } |
239 | ||
abcb2947 AK |
240 | static int ext4_valid_block_bitmap(struct super_block *sb, |
241 | struct ext4_group_desc *desc, | |
242 | unsigned int block_group, | |
243 | struct buffer_head *bh) | |
244 | { | |
245 | ext4_grpblk_t offset; | |
246 | ext4_grpblk_t next_zero_bit; | |
247 | ext4_fsblk_t bitmap_blk; | |
248 | ext4_fsblk_t group_first_block; | |
249 | ||
250 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) { | |
251 | /* with FLEX_BG, the inode/block bitmaps and itable | |
252 | * blocks may not be in the group at all | |
253 | * so the bitmap validation will be skipped for those groups | |
254 | * or it has to also read the block group where the bitmaps | |
255 | * are located to verify they are set. | |
256 | */ | |
257 | return 1; | |
258 | } | |
259 | group_first_block = ext4_group_first_block_no(sb, block_group); | |
260 | ||
261 | /* check whether block bitmap block number is set */ | |
262 | bitmap_blk = ext4_block_bitmap(sb, desc); | |
263 | offset = bitmap_blk - group_first_block; | |
264 | if (!ext4_test_bit(offset, bh->b_data)) | |
265 | /* bad block bitmap */ | |
266 | goto err_out; | |
267 | ||
268 | /* check whether the inode bitmap block number is set */ | |
269 | bitmap_blk = ext4_inode_bitmap(sb, desc); | |
270 | offset = bitmap_blk - group_first_block; | |
271 | if (!ext4_test_bit(offset, bh->b_data)) | |
272 | /* bad block bitmap */ | |
273 | goto err_out; | |
274 | ||
275 | /* check whether the inode table block number is set */ | |
276 | bitmap_blk = ext4_inode_table(sb, desc); | |
277 | offset = bitmap_blk - group_first_block; | |
278 | next_zero_bit = ext4_find_next_zero_bit(bh->b_data, | |
279 | offset + EXT4_SB(sb)->s_itb_per_group, | |
280 | offset); | |
281 | if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group) | |
282 | /* good bitmap for inode tables */ | |
283 | return 1; | |
284 | ||
285 | err_out: | |
46e665e9 | 286 | ext4_error(sb, __func__, |
abcb2947 AK |
287 | "Invalid block bitmap - " |
288 | "block_group = %d, block = %llu", | |
289 | block_group, bitmap_blk); | |
290 | return 0; | |
291 | } | |
ac27a0ec | 292 | /** |
574ca174 | 293 | * ext4_read_block_bitmap() |
ac27a0ec DK |
294 | * @sb: super block |
295 | * @block_group: given block group | |
296 | * | |
abcb2947 AK |
297 | * Read the bitmap for a given block_group,and validate the |
298 | * bits for block/inode/inode tables are set in the bitmaps | |
ac27a0ec DK |
299 | * |
300 | * Return buffer_head on success or NULL in case of failure. | |
301 | */ | |
717d50e4 | 302 | struct buffer_head * |
574ca174 | 303 | ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 304 | { |
af5bc92d TT |
305 | struct ext4_group_desc *desc; |
306 | struct buffer_head *bh = NULL; | |
7c9e69fa | 307 | ext4_fsblk_t bitmap_blk; |
ac27a0ec | 308 | |
717d50e4 | 309 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 310 | if (!desc) |
7c9e69fa AK |
311 | return NULL; |
312 | bitmap_blk = ext4_block_bitmap(sb, desc); | |
abcb2947 AK |
313 | bh = sb_getblk(sb, bitmap_blk); |
314 | if (unlikely(!bh)) { | |
46e665e9 | 315 | ext4_error(sb, __func__, |
abcb2947 | 316 | "Cannot read block bitmap - " |
e29d1cde ES |
317 | "block_group = %lu, block_bitmap = %llu", |
318 | block_group, bitmap_blk); | |
abcb2947 AK |
319 | return NULL; |
320 | } | |
321 | if (bh_uptodate_or_lock(bh)) | |
322 | return bh; | |
323 | ||
b5f10eed | 324 | spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group)); |
717d50e4 | 325 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
abcb2947 AK |
326 | ext4_init_block_bitmap(sb, bh, block_group, desc); |
327 | set_buffer_uptodate(bh); | |
328 | unlock_buffer(bh); | |
b5f10eed | 329 | spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group)); |
abcb2947 | 330 | return bh; |
717d50e4 | 331 | } |
b5f10eed | 332 | spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group)); |
abcb2947 AK |
333 | if (bh_submit_read(bh) < 0) { |
334 | put_bh(bh); | |
46e665e9 | 335 | ext4_error(sb, __func__, |
ac27a0ec | 336 | "Cannot read block bitmap - " |
e29d1cde ES |
337 | "block_group = %lu, block_bitmap = %llu", |
338 | block_group, bitmap_blk); | |
abcb2947 AK |
339 | return NULL; |
340 | } | |
519deca0 AK |
341 | ext4_valid_block_bitmap(sb, desc, block_group, bh); |
342 | /* | |
343 | * file system mounted not to panic on error, | |
344 | * continue with corrupt bitmap | |
345 | */ | |
ac27a0ec DK |
346 | return bh; |
347 | } | |
348 | /* | |
349 | * The reservation window structure operations | |
350 | * -------------------------------------------- | |
351 | * Operations include: | |
352 | * dump, find, add, remove, is_empty, find_next_reservable_window, etc. | |
353 | * | |
354 | * We use a red-black tree to represent per-filesystem reservation | |
355 | * windows. | |
356 | * | |
357 | */ | |
358 | ||
359 | /** | |
360 | * __rsv_window_dump() -- Dump the filesystem block allocation reservation map | |
361 | * @rb_root: root of per-filesystem reservation rb tree | |
362 | * @verbose: verbose mode | |
363 | * @fn: function which wishes to dump the reservation map | |
364 | * | |
365 | * If verbose is turned on, it will print the whole block reservation | |
366 | * windows(start, end). Otherwise, it will only print out the "bad" windows, | |
367 | * those windows that overlap with their immediate neighbors. | |
368 | */ | |
369 | #if 1 | |
370 | static void __rsv_window_dump(struct rb_root *root, int verbose, | |
371 | const char *fn) | |
372 | { | |
373 | struct rb_node *n; | |
617ba13b | 374 | struct ext4_reserve_window_node *rsv, *prev; |
ac27a0ec DK |
375 | int bad; |
376 | ||
377 | restart: | |
378 | n = rb_first(root); | |
379 | bad = 0; | |
380 | prev = NULL; | |
381 | ||
4776004f TT |
382 | printk(KERN_DEBUG "Block Allocation Reservation " |
383 | "Windows Map (%s):\n", fn); | |
ac27a0ec | 384 | while (n) { |
b78a657f | 385 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec | 386 | if (verbose) |
4776004f | 387 | printk(KERN_DEBUG "reservation window 0x%p " |
2ae02107 | 388 | "start: %llu, end: %llu\n", |
ac27a0ec DK |
389 | rsv, rsv->rsv_start, rsv->rsv_end); |
390 | if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) { | |
4776004f | 391 | printk(KERN_DEBUG "Bad reservation %p (start >= end)\n", |
ac27a0ec DK |
392 | rsv); |
393 | bad = 1; | |
394 | } | |
395 | if (prev && prev->rsv_end >= rsv->rsv_start) { | |
4776004f TT |
396 | printk(KERN_DEBUG "Bad reservation %p " |
397 | "(prev->end >= start)\n", rsv); | |
ac27a0ec DK |
398 | bad = 1; |
399 | } | |
400 | if (bad) { | |
401 | if (!verbose) { | |
4776004f TT |
402 | printk(KERN_DEBUG "Restarting reservation " |
403 | "walk in verbose mode\n"); | |
ac27a0ec DK |
404 | verbose = 1; |
405 | goto restart; | |
406 | } | |
407 | } | |
408 | n = rb_next(n); | |
409 | prev = rsv; | |
410 | } | |
4776004f | 411 | printk(KERN_DEBUG "Window map complete.\n"); |
07d45f12 | 412 | BUG_ON(bad); |
ac27a0ec DK |
413 | } |
414 | #define rsv_window_dump(root, verbose) \ | |
46e665e9 | 415 | __rsv_window_dump((root), (verbose), __func__) |
ac27a0ec DK |
416 | #else |
417 | #define rsv_window_dump(root, verbose) do {} while (0) | |
418 | #endif | |
419 | ||
420 | /** | |
421 | * goal_in_my_reservation() | |
422 | * @rsv: inode's reservation window | |
423 | * @grp_goal: given goal block relative to the allocation block group | |
424 | * @group: the current allocation block group | |
425 | * @sb: filesystem super block | |
426 | * | |
427 | * Test if the given goal block (group relative) is within the file's | |
428 | * own block reservation window range. | |
429 | * | |
430 | * If the reservation window is outside the goal allocation group, return 0; | |
431 | * grp_goal (given goal block) could be -1, which means no specific | |
432 | * goal block. In this case, always return 1. | |
433 | * If the goal block is within the reservation window, return 1; | |
434 | * otherwise, return 0; | |
435 | */ | |
436 | static int | |
617ba13b | 437 | goal_in_my_reservation(struct ext4_reserve_window *rsv, ext4_grpblk_t grp_goal, |
fd2d4291 | 438 | ext4_group_t group, struct super_block *sb) |
ac27a0ec | 439 | { |
617ba13b | 440 | ext4_fsblk_t group_first_block, group_last_block; |
ac27a0ec | 441 | |
617ba13b MC |
442 | group_first_block = ext4_group_first_block_no(sb, group); |
443 | group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
444 | |
445 | if ((rsv->_rsv_start > group_last_block) || | |
446 | (rsv->_rsv_end < group_first_block)) | |
447 | return 0; | |
448 | if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start) | |
449 | || (grp_goal + group_first_block > rsv->_rsv_end))) | |
450 | return 0; | |
451 | return 1; | |
452 | } | |
453 | ||
454 | /** | |
455 | * search_reserve_window() | |
456 | * @rb_root: root of reservation tree | |
457 | * @goal: target allocation block | |
458 | * | |
459 | * Find the reserved window which includes the goal, or the previous one | |
460 | * if the goal is not in any window. | |
461 | * Returns NULL if there are no windows or if all windows start after the goal. | |
462 | */ | |
617ba13b MC |
463 | static struct ext4_reserve_window_node * |
464 | search_reserve_window(struct rb_root *root, ext4_fsblk_t goal) | |
ac27a0ec DK |
465 | { |
466 | struct rb_node *n = root->rb_node; | |
617ba13b | 467 | struct ext4_reserve_window_node *rsv; |
ac27a0ec DK |
468 | |
469 | if (!n) | |
470 | return NULL; | |
471 | ||
472 | do { | |
617ba13b | 473 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
474 | |
475 | if (goal < rsv->rsv_start) | |
476 | n = n->rb_left; | |
477 | else if (goal > rsv->rsv_end) | |
478 | n = n->rb_right; | |
479 | else | |
480 | return rsv; | |
481 | } while (n); | |
482 | /* | |
483 | * We've fallen off the end of the tree: the goal wasn't inside | |
484 | * any particular node. OK, the previous node must be to one | |
485 | * side of the interval containing the goal. If it's the RHS, | |
486 | * we need to back up one. | |
487 | */ | |
488 | if (rsv->rsv_start > goal) { | |
489 | n = rb_prev(&rsv->rsv_node); | |
617ba13b | 490 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
491 | } |
492 | return rsv; | |
493 | } | |
494 | ||
495 | /** | |
617ba13b | 496 | * ext4_rsv_window_add() -- Insert a window to the block reservation rb tree. |
ac27a0ec DK |
497 | * @sb: super block |
498 | * @rsv: reservation window to add | |
499 | * | |
500 | * Must be called with rsv_lock hold. | |
501 | */ | |
617ba13b MC |
502 | void ext4_rsv_window_add(struct super_block *sb, |
503 | struct ext4_reserve_window_node *rsv) | |
ac27a0ec | 504 | { |
617ba13b | 505 | struct rb_root *root = &EXT4_SB(sb)->s_rsv_window_root; |
ac27a0ec | 506 | struct rb_node *node = &rsv->rsv_node; |
617ba13b | 507 | ext4_fsblk_t start = rsv->rsv_start; |
ac27a0ec | 508 | |
af5bc92d TT |
509 | struct rb_node **p = &root->rb_node; |
510 | struct rb_node *parent = NULL; | |
617ba13b | 511 | struct ext4_reserve_window_node *this; |
ac27a0ec DK |
512 | |
513 | while (*p) | |
514 | { | |
515 | parent = *p; | |
617ba13b | 516 | this = rb_entry(parent, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
517 | |
518 | if (start < this->rsv_start) | |
519 | p = &(*p)->rb_left; | |
520 | else if (start > this->rsv_end) | |
521 | p = &(*p)->rb_right; | |
522 | else { | |
523 | rsv_window_dump(root, 1); | |
524 | BUG(); | |
525 | } | |
526 | } | |
527 | ||
528 | rb_link_node(node, parent, p); | |
529 | rb_insert_color(node, root); | |
530 | } | |
531 | ||
532 | /** | |
617ba13b | 533 | * ext4_rsv_window_remove() -- unlink a window from the reservation rb tree |
ac27a0ec DK |
534 | * @sb: super block |
535 | * @rsv: reservation window to remove | |
536 | * | |
537 | * Mark the block reservation window as not allocated, and unlink it | |
538 | * from the filesystem reservation window rb tree. Must be called with | |
539 | * rsv_lock hold. | |
540 | */ | |
541 | static void rsv_window_remove(struct super_block *sb, | |
617ba13b | 542 | struct ext4_reserve_window_node *rsv) |
ac27a0ec | 543 | { |
617ba13b MC |
544 | rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
545 | rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; | |
ac27a0ec | 546 | rsv->rsv_alloc_hit = 0; |
617ba13b | 547 | rb_erase(&rsv->rsv_node, &EXT4_SB(sb)->s_rsv_window_root); |
ac27a0ec DK |
548 | } |
549 | ||
550 | /* | |
551 | * rsv_is_empty() -- Check if the reservation window is allocated. | |
552 | * @rsv: given reservation window to check | |
553 | * | |
617ba13b | 554 | * returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED. |
ac27a0ec | 555 | */ |
617ba13b | 556 | static inline int rsv_is_empty(struct ext4_reserve_window *rsv) |
ac27a0ec DK |
557 | { |
558 | /* a valid reservation end block could not be 0 */ | |
617ba13b | 559 | return rsv->_rsv_end == EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
ac27a0ec DK |
560 | } |
561 | ||
562 | /** | |
617ba13b | 563 | * ext4_init_block_alloc_info() |
ac27a0ec DK |
564 | * @inode: file inode structure |
565 | * | |
566 | * Allocate and initialize the reservation window structure, and | |
617ba13b | 567 | * link the window to the ext4 inode structure at last |
ac27a0ec DK |
568 | * |
569 | * The reservation window structure is only dynamically allocated | |
617ba13b MC |
570 | * and linked to ext4 inode the first time the open file |
571 | * needs a new block. So, before every ext4_new_block(s) call, for | |
ac27a0ec DK |
572 | * regular files, we should check whether the reservation window |
573 | * structure exists or not. In the latter case, this function is called. | |
574 | * Fail to do so will result in block reservation being turned off for that | |
575 | * open file. | |
576 | * | |
617ba13b | 577 | * This function is called from ext4_get_blocks_handle(), also called |
ac27a0ec DK |
578 | * when setting the reservation window size through ioctl before the file |
579 | * is open for write (needs block allocation). | |
580 | * | |
0e855ac8 | 581 | * Needs down_write(i_data_sem) protection prior to call this function. |
ac27a0ec | 582 | */ |
617ba13b | 583 | void ext4_init_block_alloc_info(struct inode *inode) |
ac27a0ec | 584 | { |
617ba13b MC |
585 | struct ext4_inode_info *ei = EXT4_I(inode); |
586 | struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info; | |
ac27a0ec DK |
587 | struct super_block *sb = inode->i_sb; |
588 | ||
589 | block_i = kmalloc(sizeof(*block_i), GFP_NOFS); | |
590 | if (block_i) { | |
617ba13b | 591 | struct ext4_reserve_window_node *rsv = &block_i->rsv_window_node; |
ac27a0ec | 592 | |
617ba13b MC |
593 | rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
594 | rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; | |
ac27a0ec DK |
595 | |
596 | /* | |
597 | * if filesystem is mounted with NORESERVATION, the goal | |
598 | * reservation window size is set to zero to indicate | |
599 | * block reservation is off | |
600 | */ | |
601 | if (!test_opt(sb, RESERVATION)) | |
602 | rsv->rsv_goal_size = 0; | |
603 | else | |
617ba13b | 604 | rsv->rsv_goal_size = EXT4_DEFAULT_RESERVE_BLOCKS; |
ac27a0ec DK |
605 | rsv->rsv_alloc_hit = 0; |
606 | block_i->last_alloc_logical_block = 0; | |
607 | block_i->last_alloc_physical_block = 0; | |
608 | } | |
609 | ei->i_block_alloc_info = block_i; | |
610 | } | |
611 | ||
612 | /** | |
617ba13b | 613 | * ext4_discard_reservation() |
ac27a0ec DK |
614 | * @inode: inode |
615 | * | |
616 | * Discard(free) block reservation window on last file close, or truncate | |
617 | * or at last iput(). | |
618 | * | |
619 | * It is being called in three cases: | |
617ba13b MC |
620 | * ext4_release_file(): last writer close the file |
621 | * ext4_clear_inode(): last iput(), when nobody link to this file. | |
622 | * ext4_truncate(): when the block indirect map is about to change. | |
ac27a0ec DK |
623 | * |
624 | */ | |
617ba13b | 625 | void ext4_discard_reservation(struct inode *inode) |
ac27a0ec | 626 | { |
617ba13b MC |
627 | struct ext4_inode_info *ei = EXT4_I(inode); |
628 | struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info; | |
629 | struct ext4_reserve_window_node *rsv; | |
630 | spinlock_t *rsv_lock = &EXT4_SB(inode->i_sb)->s_rsv_window_lock; | |
ac27a0ec | 631 | |
c9de560d AT |
632 | ext4_mb_discard_inode_preallocations(inode); |
633 | ||
ac27a0ec DK |
634 | if (!block_i) |
635 | return; | |
636 | ||
637 | rsv = &block_i->rsv_window_node; | |
638 | if (!rsv_is_empty(&rsv->rsv_window)) { | |
639 | spin_lock(rsv_lock); | |
640 | if (!rsv_is_empty(&rsv->rsv_window)) | |
641 | rsv_window_remove(inode->i_sb, rsv); | |
642 | spin_unlock(rsv_lock); | |
643 | } | |
644 | } | |
645 | ||
646 | /** | |
617ba13b | 647 | * ext4_free_blocks_sb() -- Free given blocks and update quota |
ac27a0ec DK |
648 | * @handle: handle to this transaction |
649 | * @sb: super block | |
650 | * @block: start physcial block to free | |
651 | * @count: number of blocks to free | |
652 | * @pdquot_freed_blocks: pointer to quota | |
653 | */ | |
617ba13b MC |
654 | void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb, |
655 | ext4_fsblk_t block, unsigned long count, | |
ac27a0ec DK |
656 | unsigned long *pdquot_freed_blocks) |
657 | { | |
658 | struct buffer_head *bitmap_bh = NULL; | |
659 | struct buffer_head *gd_bh; | |
fd2d4291 | 660 | ext4_group_t block_group; |
617ba13b | 661 | ext4_grpblk_t bit; |
ac27a0ec DK |
662 | unsigned long i; |
663 | unsigned long overflow; | |
af5bc92d TT |
664 | struct ext4_group_desc *desc; |
665 | struct ext4_super_block *es; | |
617ba13b | 666 | struct ext4_sb_info *sbi; |
ac27a0ec | 667 | int err = 0, ret; |
617ba13b | 668 | ext4_grpblk_t group_freed; |
ac27a0ec DK |
669 | |
670 | *pdquot_freed_blocks = 0; | |
617ba13b | 671 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
672 | es = sbi->s_es; |
673 | if (block < le32_to_cpu(es->s_first_data_block) || | |
674 | block + count < block || | |
bd81d8ee | 675 | block + count > ext4_blocks_count(es)) { |
af5bc92d TT |
676 | ext4_error(sb, "ext4_free_blocks", |
677 | "Freeing blocks not in datazone - " | |
678 | "block = %llu, count = %lu", block, count); | |
ac27a0ec DK |
679 | goto error_return; |
680 | } | |
681 | ||
af5bc92d | 682 | ext4_debug("freeing block(s) %llu-%llu\n", block, block + count - 1); |
ac27a0ec DK |
683 | |
684 | do_more: | |
685 | overflow = 0; | |
3a5b2ecd | 686 | ext4_get_group_no_and_offset(sb, block, &block_group, &bit); |
ac27a0ec DK |
687 | /* |
688 | * Check to see if we are freeing blocks across a group | |
689 | * boundary. | |
690 | */ | |
617ba13b MC |
691 | if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) { |
692 | overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb); | |
ac27a0ec DK |
693 | count -= overflow; |
694 | } | |
695 | brelse(bitmap_bh); | |
574ca174 | 696 | bitmap_bh = ext4_read_block_bitmap(sb, block_group); |
ac27a0ec DK |
697 | if (!bitmap_bh) |
698 | goto error_return; | |
af5bc92d | 699 | desc = ext4_get_group_desc(sb, block_group, &gd_bh); |
ac27a0ec DK |
700 | if (!desc) |
701 | goto error_return; | |
702 | ||
8fadc143 AR |
703 | if (in_range(ext4_block_bitmap(sb, desc), block, count) || |
704 | in_range(ext4_inode_bitmap(sb, desc), block, count) || | |
705 | in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) || | |
706 | in_range(block + count - 1, ext4_inode_table(sb, desc), | |
cb47dce7 | 707 | sbi->s_itb_per_group)) { |
af5bc92d TT |
708 | ext4_error(sb, "ext4_free_blocks", |
709 | "Freeing blocks in system zones - " | |
710 | "Block = %llu, count = %lu", | |
711 | block, count); | |
cb47dce7 AK |
712 | goto error_return; |
713 | } | |
ac27a0ec DK |
714 | |
715 | /* | |
716 | * We are about to start releasing blocks in the bitmap, | |
717 | * so we need undo access. | |
718 | */ | |
719 | /* @@@ check errors */ | |
720 | BUFFER_TRACE(bitmap_bh, "getting undo access"); | |
617ba13b | 721 | err = ext4_journal_get_undo_access(handle, bitmap_bh); |
ac27a0ec DK |
722 | if (err) |
723 | goto error_return; | |
724 | ||
725 | /* | |
726 | * We are about to modify some metadata. Call the journal APIs | |
727 | * to unshare ->b_data if a currently-committing transaction is | |
728 | * using it | |
729 | */ | |
730 | BUFFER_TRACE(gd_bh, "get_write_access"); | |
617ba13b | 731 | err = ext4_journal_get_write_access(handle, gd_bh); |
ac27a0ec DK |
732 | if (err) |
733 | goto error_return; | |
734 | ||
735 | jbd_lock_bh_state(bitmap_bh); | |
736 | ||
737 | for (i = 0, group_freed = 0; i < count; i++) { | |
738 | /* | |
739 | * An HJ special. This is expensive... | |
740 | */ | |
e23291b9 | 741 | #ifdef CONFIG_JBD2_DEBUG |
ac27a0ec DK |
742 | jbd_unlock_bh_state(bitmap_bh); |
743 | { | |
744 | struct buffer_head *debug_bh; | |
745 | debug_bh = sb_find_get_block(sb, block + i); | |
746 | if (debug_bh) { | |
747 | BUFFER_TRACE(debug_bh, "Deleted!"); | |
748 | if (!bh2jh(bitmap_bh)->b_committed_data) | |
749 | BUFFER_TRACE(debug_bh, | |
750 | "No commited data in bitmap"); | |
751 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap"); | |
752 | __brelse(debug_bh); | |
753 | } | |
754 | } | |
755 | jbd_lock_bh_state(bitmap_bh); | |
756 | #endif | |
757 | if (need_resched()) { | |
758 | jbd_unlock_bh_state(bitmap_bh); | |
759 | cond_resched(); | |
760 | jbd_lock_bh_state(bitmap_bh); | |
761 | } | |
762 | /* @@@ This prevents newly-allocated data from being | |
763 | * freed and then reallocated within the same | |
764 | * transaction. | |
765 | * | |
766 | * Ideally we would want to allow that to happen, but to | |
dab291af | 767 | * do so requires making jbd2_journal_forget() capable of |
ac27a0ec DK |
768 | * revoking the queued write of a data block, which |
769 | * implies blocking on the journal lock. *forget() | |
770 | * cannot block due to truncate races. | |
771 | * | |
dab291af | 772 | * Eventually we can fix this by making jbd2_journal_forget() |
ac27a0ec DK |
773 | * return a status indicating whether or not it was able |
774 | * to revoke the buffer. On successful revoke, it is | |
775 | * safe not to set the allocation bit in the committed | |
776 | * bitmap, because we know that there is no outstanding | |
777 | * activity on the buffer any more and so it is safe to | |
778 | * reallocate it. | |
779 | */ | |
780 | BUFFER_TRACE(bitmap_bh, "set in b_committed_data"); | |
781 | J_ASSERT_BH(bitmap_bh, | |
782 | bh2jh(bitmap_bh)->b_committed_data != NULL); | |
617ba13b | 783 | ext4_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i, |
ac27a0ec DK |
784 | bh2jh(bitmap_bh)->b_committed_data); |
785 | ||
786 | /* | |
787 | * We clear the bit in the bitmap after setting the committed | |
788 | * data bit, because this is the reverse order to that which | |
789 | * the allocator uses. | |
790 | */ | |
791 | BUFFER_TRACE(bitmap_bh, "clear bit"); | |
617ba13b | 792 | if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group), |
ac27a0ec DK |
793 | bit + i, bitmap_bh->b_data)) { |
794 | jbd_unlock_bh_state(bitmap_bh); | |
46e665e9 | 795 | ext4_error(sb, __func__, |
2ae02107 | 796 | "bit already cleared for block %llu", |
bd81d8ee | 797 | (ext4_fsblk_t)(block + i)); |
ac27a0ec DK |
798 | jbd_lock_bh_state(bitmap_bh); |
799 | BUFFER_TRACE(bitmap_bh, "bit already cleared"); | |
800 | } else { | |
801 | group_freed++; | |
802 | } | |
803 | } | |
804 | jbd_unlock_bh_state(bitmap_bh); | |
805 | ||
806 | spin_lock(sb_bgl_lock(sbi, block_group)); | |
e8546d06 | 807 | le16_add_cpu(&desc->bg_free_blocks_count, group_freed); |
717d50e4 | 808 | desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc); |
ac27a0ec | 809 | spin_unlock(sb_bgl_lock(sbi, block_group)); |
aa0dff2d | 810 | percpu_counter_add(&sbi->s_freeblocks_counter, count); |
ac27a0ec | 811 | |
772cb7c8 JS |
812 | if (sbi->s_log_groups_per_flex) { |
813 | ext4_group_t flex_group = ext4_flex_group(sbi, block_group); | |
814 | spin_lock(sb_bgl_lock(sbi, flex_group)); | |
815 | sbi->s_flex_groups[flex_group].free_blocks += count; | |
816 | spin_unlock(sb_bgl_lock(sbi, flex_group)); | |
817 | } | |
818 | ||
ac27a0ec DK |
819 | /* We dirtied the bitmap block */ |
820 | BUFFER_TRACE(bitmap_bh, "dirtied bitmap block"); | |
617ba13b | 821 | err = ext4_journal_dirty_metadata(handle, bitmap_bh); |
ac27a0ec DK |
822 | |
823 | /* And the group descriptor block */ | |
824 | BUFFER_TRACE(gd_bh, "dirtied group descriptor block"); | |
617ba13b | 825 | ret = ext4_journal_dirty_metadata(handle, gd_bh); |
ac27a0ec DK |
826 | if (!err) err = ret; |
827 | *pdquot_freed_blocks += group_freed; | |
828 | ||
829 | if (overflow && !err) { | |
830 | block += count; | |
831 | count = overflow; | |
832 | goto do_more; | |
833 | } | |
834 | sb->s_dirt = 1; | |
835 | error_return: | |
836 | brelse(bitmap_bh); | |
617ba13b | 837 | ext4_std_error(sb, err); |
ac27a0ec DK |
838 | return; |
839 | } | |
840 | ||
841 | /** | |
617ba13b | 842 | * ext4_free_blocks() -- Free given blocks and update quota |
ac27a0ec DK |
843 | * @handle: handle for this transaction |
844 | * @inode: inode | |
845 | * @block: start physical block to free | |
846 | * @count: number of blocks to count | |
c9de560d | 847 | * @metadata: Are these metadata blocks |
ac27a0ec | 848 | */ |
617ba13b | 849 | void ext4_free_blocks(handle_t *handle, struct inode *inode, |
c9de560d AT |
850 | ext4_fsblk_t block, unsigned long count, |
851 | int metadata) | |
ac27a0ec | 852 | { |
af5bc92d | 853 | struct super_block *sb; |
ac27a0ec DK |
854 | unsigned long dquot_freed_blocks; |
855 | ||
c9de560d AT |
856 | /* this isn't the right place to decide whether block is metadata |
857 | * inode.c/extents.c knows better, but for safety ... */ | |
858 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) || | |
859 | ext4_should_journal_data(inode)) | |
860 | metadata = 1; | |
861 | ||
ac27a0ec | 862 | sb = inode->i_sb; |
c9de560d AT |
863 | |
864 | if (!test_opt(sb, MBALLOC) || !EXT4_SB(sb)->s_group_info) | |
865 | ext4_free_blocks_sb(handle, sb, block, count, | |
866 | &dquot_freed_blocks); | |
867 | else | |
868 | ext4_mb_free_blocks(handle, inode, block, count, | |
869 | metadata, &dquot_freed_blocks); | |
ac27a0ec DK |
870 | if (dquot_freed_blocks) |
871 | DQUOT_FREE_BLOCK(inode, dquot_freed_blocks); | |
872 | return; | |
873 | } | |
874 | ||
875 | /** | |
617ba13b | 876 | * ext4_test_allocatable() |
ac27a0ec DK |
877 | * @nr: given allocation block group |
878 | * @bh: bufferhead contains the bitmap of the given block group | |
879 | * | |
617ba13b | 880 | * For ext4 allocations, we must not reuse any blocks which are |
ac27a0ec DK |
881 | * allocated in the bitmap buffer's "last committed data" copy. This |
882 | * prevents deletes from freeing up the page for reuse until we have | |
883 | * committed the delete transaction. | |
884 | * | |
885 | * If we didn't do this, then deleting something and reallocating it as | |
886 | * data would allow the old block to be overwritten before the | |
887 | * transaction committed (because we force data to disk before commit). | |
888 | * This would lead to corruption if we crashed between overwriting the | |
889 | * data and committing the delete. | |
890 | * | |
891 | * @@@ We may want to make this allocation behaviour conditional on | |
892 | * data-writes at some point, and disable it for metadata allocations or | |
893 | * sync-data inodes. | |
894 | */ | |
617ba13b | 895 | static int ext4_test_allocatable(ext4_grpblk_t nr, struct buffer_head *bh) |
ac27a0ec DK |
896 | { |
897 | int ret; | |
898 | struct journal_head *jh = bh2jh(bh); | |
899 | ||
617ba13b | 900 | if (ext4_test_bit(nr, bh->b_data)) |
ac27a0ec DK |
901 | return 0; |
902 | ||
903 | jbd_lock_bh_state(bh); | |
904 | if (!jh->b_committed_data) | |
905 | ret = 1; | |
906 | else | |
617ba13b | 907 | ret = !ext4_test_bit(nr, jh->b_committed_data); |
ac27a0ec DK |
908 | jbd_unlock_bh_state(bh); |
909 | return ret; | |
910 | } | |
911 | ||
912 | /** | |
913 | * bitmap_search_next_usable_block() | |
914 | * @start: the starting block (group relative) of the search | |
915 | * @bh: bufferhead contains the block group bitmap | |
916 | * @maxblocks: the ending block (group relative) of the reservation | |
917 | * | |
918 | * The bitmap search --- search forward alternately through the actual | |
919 | * bitmap on disk and the last-committed copy in journal, until we find a | |
920 | * bit free in both bitmaps. | |
921 | */ | |
617ba13b MC |
922 | static ext4_grpblk_t |
923 | bitmap_search_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh, | |
924 | ext4_grpblk_t maxblocks) | |
ac27a0ec | 925 | { |
617ba13b | 926 | ext4_grpblk_t next; |
ac27a0ec DK |
927 | struct journal_head *jh = bh2jh(bh); |
928 | ||
929 | while (start < maxblocks) { | |
617ba13b | 930 | next = ext4_find_next_zero_bit(bh->b_data, maxblocks, start); |
ac27a0ec DK |
931 | if (next >= maxblocks) |
932 | return -1; | |
617ba13b | 933 | if (ext4_test_allocatable(next, bh)) |
ac27a0ec DK |
934 | return next; |
935 | jbd_lock_bh_state(bh); | |
936 | if (jh->b_committed_data) | |
617ba13b | 937 | start = ext4_find_next_zero_bit(jh->b_committed_data, |
ac27a0ec DK |
938 | maxblocks, next); |
939 | jbd_unlock_bh_state(bh); | |
940 | } | |
941 | return -1; | |
942 | } | |
943 | ||
944 | /** | |
945 | * find_next_usable_block() | |
946 | * @start: the starting block (group relative) to find next | |
947 | * allocatable block in bitmap. | |
948 | * @bh: bufferhead contains the block group bitmap | |
949 | * @maxblocks: the ending block (group relative) for the search | |
950 | * | |
951 | * Find an allocatable block in a bitmap. We honor both the bitmap and | |
952 | * its last-committed copy (if that exists), and perform the "most | |
953 | * appropriate allocation" algorithm of looking for a free block near | |
954 | * the initial goal; then for a free byte somewhere in the bitmap; then | |
955 | * for any free bit in the bitmap. | |
956 | */ | |
617ba13b MC |
957 | static ext4_grpblk_t |
958 | find_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh, | |
959 | ext4_grpblk_t maxblocks) | |
ac27a0ec | 960 | { |
617ba13b | 961 | ext4_grpblk_t here, next; |
ac27a0ec DK |
962 | char *p, *r; |
963 | ||
964 | if (start > 0) { | |
965 | /* | |
966 | * The goal was occupied; search forward for a free | |
967 | * block within the next XX blocks. | |
968 | * | |
969 | * end_goal is more or less random, but it has to be | |
617ba13b | 970 | * less than EXT4_BLOCKS_PER_GROUP. Aligning up to the |
ac27a0ec DK |
971 | * next 64-bit boundary is simple.. |
972 | */ | |
617ba13b | 973 | ext4_grpblk_t end_goal = (start + 63) & ~63; |
ac27a0ec DK |
974 | if (end_goal > maxblocks) |
975 | end_goal = maxblocks; | |
617ba13b MC |
976 | here = ext4_find_next_zero_bit(bh->b_data, end_goal, start); |
977 | if (here < end_goal && ext4_test_allocatable(here, bh)) | |
ac27a0ec | 978 | return here; |
617ba13b | 979 | ext4_debug("Bit not found near goal\n"); |
ac27a0ec DK |
980 | } |
981 | ||
982 | here = start; | |
983 | if (here < 0) | |
984 | here = 0; | |
985 | ||
986 | p = ((char *)bh->b_data) + (here >> 3); | |
ec0837f2 | 987 | r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3)); |
ac27a0ec DK |
988 | next = (r - ((char *)bh->b_data)) << 3; |
989 | ||
617ba13b | 990 | if (next < maxblocks && next >= start && ext4_test_allocatable(next, bh)) |
ac27a0ec DK |
991 | return next; |
992 | ||
993 | /* | |
994 | * The bitmap search --- search forward alternately through the actual | |
995 | * bitmap and the last-committed copy until we find a bit free in | |
996 | * both | |
997 | */ | |
998 | here = bitmap_search_next_usable_block(here, bh, maxblocks); | |
999 | return here; | |
1000 | } | |
1001 | ||
1002 | /** | |
1003 | * claim_block() | |
1004 | * @block: the free block (group relative) to allocate | |
1005 | * @bh: the bufferhead containts the block group bitmap | |
1006 | * | |
1007 | * We think we can allocate this block in this bitmap. Try to set the bit. | |
1008 | * If that succeeds then check that nobody has allocated and then freed the | |
1009 | * block since we saw that is was not marked in b_committed_data. If it _was_ | |
1010 | * allocated and freed then clear the bit in the bitmap again and return | |
1011 | * zero (failure). | |
1012 | */ | |
1013 | static inline int | |
617ba13b | 1014 | claim_block(spinlock_t *lock, ext4_grpblk_t block, struct buffer_head *bh) |
ac27a0ec DK |
1015 | { |
1016 | struct journal_head *jh = bh2jh(bh); | |
1017 | int ret; | |
1018 | ||
617ba13b | 1019 | if (ext4_set_bit_atomic(lock, block, bh->b_data)) |
ac27a0ec DK |
1020 | return 0; |
1021 | jbd_lock_bh_state(bh); | |
af5bc92d | 1022 | if (jh->b_committed_data && ext4_test_bit(block, jh->b_committed_data)) { |
617ba13b | 1023 | ext4_clear_bit_atomic(lock, block, bh->b_data); |
ac27a0ec DK |
1024 | ret = 0; |
1025 | } else { | |
1026 | ret = 1; | |
1027 | } | |
1028 | jbd_unlock_bh_state(bh); | |
1029 | return ret; | |
1030 | } | |
1031 | ||
1032 | /** | |
617ba13b | 1033 | * ext4_try_to_allocate() |
ac27a0ec DK |
1034 | * @sb: superblock |
1035 | * @handle: handle to this transaction | |
1036 | * @group: given allocation block group | |
1037 | * @bitmap_bh: bufferhead holds the block bitmap | |
1038 | * @grp_goal: given target block within the group | |
1039 | * @count: target number of blocks to allocate | |
1040 | * @my_rsv: reservation window | |
1041 | * | |
1042 | * Attempt to allocate blocks within a give range. Set the range of allocation | |
1043 | * first, then find the first free bit(s) from the bitmap (within the range), | |
1044 | * and at last, allocate the blocks by claiming the found free bit as allocated. | |
1045 | * | |
1046 | * To set the range of this allocation: | |
1047 | * if there is a reservation window, only try to allocate block(s) from the | |
1048 | * file's own reservation window; | |
1049 | * Otherwise, the allocation range starts from the give goal block, ends at | |
1050 | * the block group's last block. | |
1051 | * | |
1052 | * If we failed to allocate the desired block then we may end up crossing to a | |
1053 | * new bitmap. In that case we must release write access to the old one via | |
617ba13b | 1054 | * ext4_journal_release_buffer(), else we'll run out of credits. |
ac27a0ec | 1055 | */ |
617ba13b | 1056 | static ext4_grpblk_t |
fd2d4291 AM |
1057 | ext4_try_to_allocate(struct super_block *sb, handle_t *handle, |
1058 | ext4_group_t group, struct buffer_head *bitmap_bh, | |
1059 | ext4_grpblk_t grp_goal, unsigned long *count, | |
1060 | struct ext4_reserve_window *my_rsv) | |
ac27a0ec | 1061 | { |
617ba13b MC |
1062 | ext4_fsblk_t group_first_block; |
1063 | ext4_grpblk_t start, end; | |
ac27a0ec DK |
1064 | unsigned long num = 0; |
1065 | ||
1066 | /* we do allocation within the reservation window if we have a window */ | |
1067 | if (my_rsv) { | |
617ba13b | 1068 | group_first_block = ext4_group_first_block_no(sb, group); |
ac27a0ec DK |
1069 | if (my_rsv->_rsv_start >= group_first_block) |
1070 | start = my_rsv->_rsv_start - group_first_block; | |
1071 | else | |
1072 | /* reservation window cross group boundary */ | |
1073 | start = 0; | |
1074 | end = my_rsv->_rsv_end - group_first_block + 1; | |
617ba13b | 1075 | if (end > EXT4_BLOCKS_PER_GROUP(sb)) |
ac27a0ec | 1076 | /* reservation window crosses group boundary */ |
617ba13b | 1077 | end = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
1078 | if ((start <= grp_goal) && (grp_goal < end)) |
1079 | start = grp_goal; | |
1080 | else | |
1081 | grp_goal = -1; | |
1082 | } else { | |
1083 | if (grp_goal > 0) | |
1084 | start = grp_goal; | |
1085 | else | |
1086 | start = 0; | |
617ba13b | 1087 | end = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
1088 | } |
1089 | ||
617ba13b | 1090 | BUG_ON(start > EXT4_BLOCKS_PER_GROUP(sb)); |
ac27a0ec DK |
1091 | |
1092 | repeat: | |
617ba13b | 1093 | if (grp_goal < 0 || !ext4_test_allocatable(grp_goal, bitmap_bh)) { |
ac27a0ec DK |
1094 | grp_goal = find_next_usable_block(start, bitmap_bh, end); |
1095 | if (grp_goal < 0) | |
1096 | goto fail_access; | |
1097 | if (!my_rsv) { | |
1098 | int i; | |
1099 | ||
1100 | for (i = 0; i < 7 && grp_goal > start && | |
617ba13b | 1101 | ext4_test_allocatable(grp_goal - 1, |
ac27a0ec DK |
1102 | bitmap_bh); |
1103 | i++, grp_goal--) | |
1104 | ; | |
1105 | } | |
1106 | } | |
1107 | start = grp_goal; | |
1108 | ||
617ba13b | 1109 | if (!claim_block(sb_bgl_lock(EXT4_SB(sb), group), |
ac27a0ec DK |
1110 | grp_goal, bitmap_bh)) { |
1111 | /* | |
1112 | * The block was allocated by another thread, or it was | |
1113 | * allocated and then freed by another thread | |
1114 | */ | |
1115 | start++; | |
1116 | grp_goal++; | |
1117 | if (start >= end) | |
1118 | goto fail_access; | |
1119 | goto repeat; | |
1120 | } | |
1121 | num++; | |
1122 | grp_goal++; | |
1123 | while (num < *count && grp_goal < end | |
617ba13b MC |
1124 | && ext4_test_allocatable(grp_goal, bitmap_bh) |
1125 | && claim_block(sb_bgl_lock(EXT4_SB(sb), group), | |
ac27a0ec DK |
1126 | grp_goal, bitmap_bh)) { |
1127 | num++; | |
1128 | grp_goal++; | |
1129 | } | |
1130 | *count = num; | |
1131 | return grp_goal - num; | |
1132 | fail_access: | |
1133 | *count = num; | |
1134 | return -1; | |
1135 | } | |
1136 | ||
1137 | /** | |
1138 | * find_next_reservable_window(): | |
1139 | * find a reservable space within the given range. | |
1140 | * It does not allocate the reservation window for now: | |
1141 | * alloc_new_reservation() will do the work later. | |
1142 | * | |
1143 | * @search_head: the head of the searching list; | |
1144 | * This is not necessarily the list head of the whole filesystem | |
1145 | * | |
1146 | * We have both head and start_block to assist the search | |
1147 | * for the reservable space. The list starts from head, | |
1148 | * but we will shift to the place where start_block is, | |
1149 | * then start from there, when looking for a reservable space. | |
1150 | * | |
1151 | * @size: the target new reservation window size | |
1152 | * | |
1153 | * @group_first_block: the first block we consider to start | |
1154 | * the real search from | |
1155 | * | |
1156 | * @last_block: | |
1157 | * the maximum block number that our goal reservable space | |
1158 | * could start from. This is normally the last block in this | |
1159 | * group. The search will end when we found the start of next | |
1160 | * possible reservable space is out of this boundary. | |
1161 | * This could handle the cross boundary reservation window | |
1162 | * request. | |
1163 | * | |
1164 | * basically we search from the given range, rather than the whole | |
1165 | * reservation double linked list, (start_block, last_block) | |
1166 | * to find a free region that is of my size and has not | |
1167 | * been reserved. | |
1168 | * | |
1169 | */ | |
1170 | static int find_next_reservable_window( | |
617ba13b MC |
1171 | struct ext4_reserve_window_node *search_head, |
1172 | struct ext4_reserve_window_node *my_rsv, | |
af5bc92d | 1173 | struct super_block *sb, |
617ba13b MC |
1174 | ext4_fsblk_t start_block, |
1175 | ext4_fsblk_t last_block) | |
ac27a0ec DK |
1176 | { |
1177 | struct rb_node *next; | |
617ba13b MC |
1178 | struct ext4_reserve_window_node *rsv, *prev; |
1179 | ext4_fsblk_t cur; | |
ac27a0ec DK |
1180 | int size = my_rsv->rsv_goal_size; |
1181 | ||
1182 | /* TODO: make the start of the reservation window byte-aligned */ | |
1183 | /* cur = *start_block & ~7;*/ | |
1184 | cur = start_block; | |
1185 | rsv = search_head; | |
1186 | if (!rsv) | |
1187 | return -1; | |
1188 | ||
1189 | while (1) { | |
1190 | if (cur <= rsv->rsv_end) | |
1191 | cur = rsv->rsv_end + 1; | |
1192 | ||
1193 | /* TODO? | |
1194 | * in the case we could not find a reservable space | |
1195 | * that is what is expected, during the re-search, we could | |
1196 | * remember what's the largest reservable space we could have | |
1197 | * and return that one. | |
1198 | * | |
1199 | * For now it will fail if we could not find the reservable | |
1200 | * space with expected-size (or more)... | |
1201 | */ | |
1202 | if (cur > last_block) | |
1203 | return -1; /* fail */ | |
1204 | ||
1205 | prev = rsv; | |
1206 | next = rb_next(&rsv->rsv_node); | |
af5bc92d | 1207 | rsv = rb_entry(next, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
1208 | |
1209 | /* | |
1210 | * Reached the last reservation, we can just append to the | |
1211 | * previous one. | |
1212 | */ | |
1213 | if (!next) | |
1214 | break; | |
1215 | ||
1216 | if (cur + size <= rsv->rsv_start) { | |
1217 | /* | |
1218 | * Found a reserveable space big enough. We could | |
1219 | * have a reservation across the group boundary here | |
1220 | */ | |
1221 | break; | |
1222 | } | |
1223 | } | |
1224 | /* | |
1225 | * we come here either : | |
1226 | * when we reach the end of the whole list, | |
1227 | * and there is empty reservable space after last entry in the list. | |
1228 | * append it to the end of the list. | |
1229 | * | |
1230 | * or we found one reservable space in the middle of the list, | |
1231 | * return the reservation window that we could append to. | |
1232 | * succeed. | |
1233 | */ | |
1234 | ||
1235 | if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window))) | |
1236 | rsv_window_remove(sb, my_rsv); | |
1237 | ||
1238 | /* | |
1239 | * Let's book the whole avaliable window for now. We will check the | |
1240 | * disk bitmap later and then, if there are free blocks then we adjust | |
1241 | * the window size if it's larger than requested. | |
1242 | * Otherwise, we will remove this node from the tree next time | |
1243 | * call find_next_reservable_window. | |
1244 | */ | |
1245 | my_rsv->rsv_start = cur; | |
1246 | my_rsv->rsv_end = cur + size - 1; | |
1247 | my_rsv->rsv_alloc_hit = 0; | |
1248 | ||
1249 | if (prev != my_rsv) | |
617ba13b | 1250 | ext4_rsv_window_add(sb, my_rsv); |
ac27a0ec DK |
1251 | |
1252 | return 0; | |
1253 | } | |
1254 | ||
1255 | /** | |
1256 | * alloc_new_reservation()--allocate a new reservation window | |
1257 | * | |
1258 | * To make a new reservation, we search part of the filesystem | |
1259 | * reservation list (the list that inside the group). We try to | |
1260 | * allocate a new reservation window near the allocation goal, | |
1261 | * or the beginning of the group, if there is no goal. | |
1262 | * | |
1263 | * We first find a reservable space after the goal, then from | |
1264 | * there, we check the bitmap for the first free block after | |
1265 | * it. If there is no free block until the end of group, then the | |
1266 | * whole group is full, we failed. Otherwise, check if the free | |
1267 | * block is inside the expected reservable space, if so, we | |
1268 | * succeed. | |
1269 | * If the first free block is outside the reservable space, then | |
1270 | * start from the first free block, we search for next available | |
1271 | * space, and go on. | |
1272 | * | |
1273 | * on succeed, a new reservation will be found and inserted into the list | |
1274 | * It contains at least one free block, and it does not overlap with other | |
1275 | * reservation windows. | |
1276 | * | |
1277 | * failed: we failed to find a reservation window in this group | |
1278 | * | |
1279 | * @rsv: the reservation | |
1280 | * | |
1281 | * @grp_goal: The goal (group-relative). It is where the search for a | |
1282 | * free reservable space should start from. | |
1283 | * if we have a grp_goal(grp_goal >0 ), then start from there, | |
1284 | * no grp_goal(grp_goal = -1), we start from the first block | |
1285 | * of the group. | |
1286 | * | |
1287 | * @sb: the super block | |
1288 | * @group: the group we are trying to allocate in | |
1289 | * @bitmap_bh: the block group block bitmap | |
1290 | * | |
1291 | */ | |
617ba13b MC |
1292 | static int alloc_new_reservation(struct ext4_reserve_window_node *my_rsv, |
1293 | ext4_grpblk_t grp_goal, struct super_block *sb, | |
fd2d4291 | 1294 | ext4_group_t group, struct buffer_head *bitmap_bh) |
ac27a0ec | 1295 | { |
617ba13b MC |
1296 | struct ext4_reserve_window_node *search_head; |
1297 | ext4_fsblk_t group_first_block, group_end_block, start_block; | |
1298 | ext4_grpblk_t first_free_block; | |
1299 | struct rb_root *fs_rsv_root = &EXT4_SB(sb)->s_rsv_window_root; | |
ac27a0ec DK |
1300 | unsigned long size; |
1301 | int ret; | |
617ba13b | 1302 | spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock; |
ac27a0ec | 1303 | |
617ba13b MC |
1304 | group_first_block = ext4_group_first_block_no(sb, group); |
1305 | group_end_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
1306 | |
1307 | if (grp_goal < 0) | |
1308 | start_block = group_first_block; | |
1309 | else | |
1310 | start_block = grp_goal + group_first_block; | |
1311 | ||
1312 | size = my_rsv->rsv_goal_size; | |
1313 | ||
1314 | if (!rsv_is_empty(&my_rsv->rsv_window)) { | |
1315 | /* | |
1316 | * if the old reservation is cross group boundary | |
1317 | * and if the goal is inside the old reservation window, | |
1318 | * we will come here when we just failed to allocate from | |
1319 | * the first part of the window. We still have another part | |
1320 | * that belongs to the next group. In this case, there is no | |
1321 | * point to discard our window and try to allocate a new one | |
1322 | * in this group(which will fail). we should | |
1323 | * keep the reservation window, just simply move on. | |
1324 | * | |
1325 | * Maybe we could shift the start block of the reservation | |
1326 | * window to the first block of next group. | |
1327 | */ | |
1328 | ||
1329 | if ((my_rsv->rsv_start <= group_end_block) && | |
1330 | (my_rsv->rsv_end > group_end_block) && | |
1331 | (start_block >= my_rsv->rsv_start)) | |
1332 | return -1; | |
1333 | ||
1334 | if ((my_rsv->rsv_alloc_hit > | |
1335 | (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) { | |
1336 | /* | |
1337 | * if the previously allocation hit ratio is | |
1338 | * greater than 1/2, then we double the size of | |
1339 | * the reservation window the next time, | |
1340 | * otherwise we keep the same size window | |
1341 | */ | |
1342 | size = size * 2; | |
617ba13b MC |
1343 | if (size > EXT4_MAX_RESERVE_BLOCKS) |
1344 | size = EXT4_MAX_RESERVE_BLOCKS; | |
af5bc92d | 1345 | my_rsv->rsv_goal_size = size; |
ac27a0ec DK |
1346 | } |
1347 | } | |
1348 | ||
1349 | spin_lock(rsv_lock); | |
1350 | /* | |
1351 | * shift the search start to the window near the goal block | |
1352 | */ | |
1353 | search_head = search_reserve_window(fs_rsv_root, start_block); | |
1354 | ||
1355 | /* | |
1356 | * find_next_reservable_window() simply finds a reservable window | |
1357 | * inside the given range(start_block, group_end_block). | |
1358 | * | |
1359 | * To make sure the reservation window has a free bit inside it, we | |
1360 | * need to check the bitmap after we found a reservable window. | |
1361 | */ | |
1362 | retry: | |
1363 | ret = find_next_reservable_window(search_head, my_rsv, sb, | |
1364 | start_block, group_end_block); | |
1365 | ||
1366 | if (ret == -1) { | |
1367 | if (!rsv_is_empty(&my_rsv->rsv_window)) | |
1368 | rsv_window_remove(sb, my_rsv); | |
1369 | spin_unlock(rsv_lock); | |
1370 | return -1; | |
1371 | } | |
1372 | ||
1373 | /* | |
1374 | * On success, find_next_reservable_window() returns the | |
1375 | * reservation window where there is a reservable space after it. | |
1376 | * Before we reserve this reservable space, we need | |
1377 | * to make sure there is at least a free block inside this region. | |
1378 | * | |
1379 | * searching the first free bit on the block bitmap and copy of | |
1380 | * last committed bitmap alternatively, until we found a allocatable | |
1381 | * block. Search start from the start block of the reservable space | |
1382 | * we just found. | |
1383 | */ | |
1384 | spin_unlock(rsv_lock); | |
1385 | first_free_block = bitmap_search_next_usable_block( | |
1386 | my_rsv->rsv_start - group_first_block, | |
1387 | bitmap_bh, group_end_block - group_first_block + 1); | |
1388 | ||
1389 | if (first_free_block < 0) { | |
1390 | /* | |
1391 | * no free block left on the bitmap, no point | |
1392 | * to reserve the space. return failed. | |
1393 | */ | |
1394 | spin_lock(rsv_lock); | |
1395 | if (!rsv_is_empty(&my_rsv->rsv_window)) | |
1396 | rsv_window_remove(sb, my_rsv); | |
1397 | spin_unlock(rsv_lock); | |
1398 | return -1; /* failed */ | |
1399 | } | |
1400 | ||
1401 | start_block = first_free_block + group_first_block; | |
1402 | /* | |
1403 | * check if the first free block is within the | |
1404 | * free space we just reserved | |
1405 | */ | |
b2f2c76d | 1406 | if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end) |
ac27a0ec DK |
1407 | return 0; /* success */ |
1408 | /* | |
1409 | * if the first free bit we found is out of the reservable space | |
1410 | * continue search for next reservable space, | |
1411 | * start from where the free block is, | |
1412 | * we also shift the list head to where we stopped last time | |
1413 | */ | |
1414 | search_head = my_rsv; | |
1415 | spin_lock(rsv_lock); | |
1416 | goto retry; | |
1417 | } | |
1418 | ||
1419 | /** | |
1420 | * try_to_extend_reservation() | |
1421 | * @my_rsv: given reservation window | |
1422 | * @sb: super block | |
1423 | * @size: the delta to extend | |
1424 | * | |
1425 | * Attempt to expand the reservation window large enough to have | |
1426 | * required number of free blocks | |
1427 | * | |
617ba13b | 1428 | * Since ext4_try_to_allocate() will always allocate blocks within |
ac27a0ec DK |
1429 | * the reservation window range, if the window size is too small, |
1430 | * multiple blocks allocation has to stop at the end of the reservation | |
1431 | * window. To make this more efficient, given the total number of | |
1432 | * blocks needed and the current size of the window, we try to | |
1433 | * expand the reservation window size if necessary on a best-effort | |
617ba13b | 1434 | * basis before ext4_new_blocks() tries to allocate blocks, |
ac27a0ec | 1435 | */ |
617ba13b | 1436 | static void try_to_extend_reservation(struct ext4_reserve_window_node *my_rsv, |
ac27a0ec DK |
1437 | struct super_block *sb, int size) |
1438 | { | |
617ba13b | 1439 | struct ext4_reserve_window_node *next_rsv; |
ac27a0ec | 1440 | struct rb_node *next; |
617ba13b | 1441 | spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock; |
ac27a0ec DK |
1442 | |
1443 | if (!spin_trylock(rsv_lock)) | |
1444 | return; | |
1445 | ||
1446 | next = rb_next(&my_rsv->rsv_node); | |
1447 | ||
1448 | if (!next) | |
1449 | my_rsv->rsv_end += size; | |
1450 | else { | |
b78a657f | 1451 | next_rsv = rb_entry(next, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
1452 | |
1453 | if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size) | |
1454 | my_rsv->rsv_end += size; | |
1455 | else | |
1456 | my_rsv->rsv_end = next_rsv->rsv_start - 1; | |
1457 | } | |
1458 | spin_unlock(rsv_lock); | |
1459 | } | |
1460 | ||
1461 | /** | |
617ba13b | 1462 | * ext4_try_to_allocate_with_rsv() |
ac27a0ec DK |
1463 | * @sb: superblock |
1464 | * @handle: handle to this transaction | |
1465 | * @group: given allocation block group | |
1466 | * @bitmap_bh: bufferhead holds the block bitmap | |
1467 | * @grp_goal: given target block within the group | |
1468 | * @count: target number of blocks to allocate | |
1469 | * @my_rsv: reservation window | |
1470 | * @errp: pointer to store the error code | |
1471 | * | |
1472 | * This is the main function used to allocate a new block and its reservation | |
1473 | * window. | |
1474 | * | |
1475 | * Each time when a new block allocation is need, first try to allocate from | |
1476 | * its own reservation. If it does not have a reservation window, instead of | |
1477 | * looking for a free bit on bitmap first, then look up the reservation list to | |
1478 | * see if it is inside somebody else's reservation window, we try to allocate a | |
1479 | * reservation window for it starting from the goal first. Then do the block | |
1480 | * allocation within the reservation window. | |
1481 | * | |
1482 | * This will avoid keeping on searching the reservation list again and | |
1483 | * again when somebody is looking for a free block (without | |
1484 | * reservation), and there are lots of free blocks, but they are all | |
1485 | * being reserved. | |
1486 | * | |
1487 | * We use a red-black tree for the per-filesystem reservation list. | |
1488 | * | |
1489 | */ | |
617ba13b MC |
1490 | static ext4_grpblk_t |
1491 | ext4_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle, | |
fd2d4291 | 1492 | ext4_group_t group, struct buffer_head *bitmap_bh, |
617ba13b | 1493 | ext4_grpblk_t grp_goal, |
af5bc92d | 1494 | struct ext4_reserve_window_node *my_rsv, |
ac27a0ec DK |
1495 | unsigned long *count, int *errp) |
1496 | { | |
617ba13b MC |
1497 | ext4_fsblk_t group_first_block, group_last_block; |
1498 | ext4_grpblk_t ret = 0; | |
ac27a0ec DK |
1499 | int fatal; |
1500 | unsigned long num = *count; | |
1501 | ||
1502 | *errp = 0; | |
1503 | ||
1504 | /* | |
1505 | * Make sure we use undo access for the bitmap, because it is critical | |
1506 | * that we do the frozen_data COW on bitmap buffers in all cases even | |
1507 | * if the buffer is in BJ_Forget state in the committing transaction. | |
1508 | */ | |
1509 | BUFFER_TRACE(bitmap_bh, "get undo access for new block"); | |
617ba13b | 1510 | fatal = ext4_journal_get_undo_access(handle, bitmap_bh); |
ac27a0ec DK |
1511 | if (fatal) { |
1512 | *errp = fatal; | |
1513 | return -1; | |
1514 | } | |
1515 | ||
1516 | /* | |
1517 | * we don't deal with reservation when | |
1518 | * filesystem is mounted without reservation | |
1519 | * or the file is not a regular file | |
1520 | * or last attempt to allocate a block with reservation turned on failed | |
1521 | */ | |
af5bc92d | 1522 | if (my_rsv == NULL) { |
617ba13b | 1523 | ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh, |
ac27a0ec DK |
1524 | grp_goal, count, NULL); |
1525 | goto out; | |
1526 | } | |
1527 | /* | |
1528 | * grp_goal is a group relative block number (if there is a goal) | |
e7dc95db | 1529 | * 0 <= grp_goal < EXT4_BLOCKS_PER_GROUP(sb) |
ac27a0ec DK |
1530 | * first block is a filesystem wide block number |
1531 | * first block is the block number of the first block in this group | |
1532 | */ | |
617ba13b MC |
1533 | group_first_block = ext4_group_first_block_no(sb, group); |
1534 | group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
1535 | |
1536 | /* | |
1537 | * Basically we will allocate a new block from inode's reservation | |
1538 | * window. | |
1539 | * | |
1540 | * We need to allocate a new reservation window, if: | |
1541 | * a) inode does not have a reservation window; or | |
1542 | * b) last attempt to allocate a block from existing reservation | |
1543 | * failed; or | |
1544 | * c) we come here with a goal and with a reservation window | |
1545 | * | |
1546 | * We do not need to allocate a new reservation window if we come here | |
1547 | * at the beginning with a goal and the goal is inside the window, or | |
1548 | * we don't have a goal but already have a reservation window. | |
1549 | * then we could go to allocate from the reservation window directly. | |
1550 | */ | |
1551 | while (1) { | |
1552 | if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) || | |
1553 | !goal_in_my_reservation(&my_rsv->rsv_window, | |
1554 | grp_goal, group, sb)) { | |
1555 | if (my_rsv->rsv_goal_size < *count) | |
1556 | my_rsv->rsv_goal_size = *count; | |
1557 | ret = alloc_new_reservation(my_rsv, grp_goal, sb, | |
1558 | group, bitmap_bh); | |
1559 | if (ret < 0) | |
1560 | break; /* failed */ | |
1561 | ||
1562 | if (!goal_in_my_reservation(&my_rsv->rsv_window, | |
1563 | grp_goal, group, sb)) | |
1564 | grp_goal = -1; | |
e7dc95db | 1565 | } else if (grp_goal >= 0) { |
1df1e63b MC |
1566 | int curr = my_rsv->rsv_end - |
1567 | (grp_goal + group_first_block) + 1; | |
1568 | ||
1569 | if (curr < *count) | |
1570 | try_to_extend_reservation(my_rsv, sb, | |
1571 | *count - curr); | |
1572 | } | |
ac27a0ec DK |
1573 | |
1574 | if ((my_rsv->rsv_start > group_last_block) || | |
1575 | (my_rsv->rsv_end < group_first_block)) { | |
617ba13b | 1576 | rsv_window_dump(&EXT4_SB(sb)->s_rsv_window_root, 1); |
ac27a0ec DK |
1577 | BUG(); |
1578 | } | |
617ba13b | 1579 | ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh, |
ac27a0ec DK |
1580 | grp_goal, &num, &my_rsv->rsv_window); |
1581 | if (ret >= 0) { | |
1582 | my_rsv->rsv_alloc_hit += num; | |
1583 | *count = num; | |
1584 | break; /* succeed */ | |
1585 | } | |
1586 | num = *count; | |
1587 | } | |
1588 | out: | |
1589 | if (ret >= 0) { | |
1590 | BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for " | |
1591 | "bitmap block"); | |
617ba13b | 1592 | fatal = ext4_journal_dirty_metadata(handle, bitmap_bh); |
ac27a0ec DK |
1593 | if (fatal) { |
1594 | *errp = fatal; | |
1595 | return -1; | |
1596 | } | |
1597 | return ret; | |
1598 | } | |
1599 | ||
1600 | BUFFER_TRACE(bitmap_bh, "journal_release_buffer"); | |
617ba13b | 1601 | ext4_journal_release_buffer(handle, bitmap_bh); |
ac27a0ec DK |
1602 | return ret; |
1603 | } | |
1604 | ||
a30d542a | 1605 | int ext4_claim_free_blocks(struct ext4_sb_info *sbi, |
5c791616 | 1606 | s64 nblocks) |
a30d542a | 1607 | { |
6bc6e63f | 1608 | s64 free_blocks, dirty_blocks; |
5c791616 | 1609 | s64 root_blocks = 0; |
a30d542a | 1610 | struct percpu_counter *fbc = &sbi->s_freeblocks_counter; |
6bc6e63f | 1611 | struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter; |
a30d542a | 1612 | |
6bc6e63f AK |
1613 | free_blocks = percpu_counter_read_positive(fbc); |
1614 | dirty_blocks = percpu_counter_read_positive(dbc); | |
a30d542a AK |
1615 | |
1616 | if (!capable(CAP_SYS_RESOURCE) && | |
1617 | sbi->s_resuid != current->fsuid && | |
1618 | (sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid))) | |
1619 | root_blocks = ext4_r_blocks_count(sbi->s_es); | |
1620 | ||
6bc6e63f AK |
1621 | if (free_blocks - (nblocks + root_blocks + dirty_blocks) < |
1622 | EXT4_FREEBLOCKS_WATERMARK) { | |
1623 | free_blocks = percpu_counter_sum(fbc); | |
1624 | dirty_blocks = percpu_counter_sum(dbc); | |
1625 | if (dirty_blocks < 0) { | |
1626 | printk(KERN_CRIT "Dirty block accounting " | |
1627 | "went wrong %lld\n", | |
1628 | dirty_blocks); | |
1629 | } | |
1630 | } | |
1631 | /* Check whether we have space after | |
1632 | * accounting for current dirty blocks | |
1633 | */ | |
5c791616 | 1634 | if (free_blocks < ((root_blocks + nblocks) + dirty_blocks)) |
a30d542a AK |
1635 | /* we don't have free space */ |
1636 | return -ENOSPC; | |
1637 | ||
6bc6e63f AK |
1638 | /* Add the blocks to nblocks */ |
1639 | percpu_counter_add(dbc, nblocks); | |
a30d542a AK |
1640 | return 0; |
1641 | } | |
1642 | ||
ac27a0ec | 1643 | /** |
617ba13b | 1644 | * ext4_has_free_blocks() |
07031431 MC |
1645 | * @sbi: in-core super block structure. |
1646 | * @nblocks: number of neeed blocks | |
ac27a0ec | 1647 | * |
07031431 MC |
1648 | * Check if filesystem has free blocks available for allocation. |
1649 | * Return the number of blocks avaible for allocation for this request | |
1650 | * On success, return nblocks | |
ac27a0ec | 1651 | */ |
07031431 | 1652 | ext4_fsblk_t ext4_has_free_blocks(struct ext4_sb_info *sbi, |
5c791616 | 1653 | s64 nblocks) |
ac27a0ec | 1654 | { |
5c791616 AK |
1655 | s64 free_blocks, dirty_blocks; |
1656 | s64 root_blocks = 0; | |
6bc6e63f AK |
1657 | struct percpu_counter *fbc = &sbi->s_freeblocks_counter; |
1658 | struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter; | |
ac27a0ec | 1659 | |
6bc6e63f AK |
1660 | free_blocks = percpu_counter_read_positive(fbc); |
1661 | dirty_blocks = percpu_counter_read_positive(dbc); | |
07031431 MC |
1662 | |
1663 | if (!capable(CAP_SYS_RESOURCE) && | |
ac27a0ec | 1664 | sbi->s_resuid != current->fsuid && |
07031431 MC |
1665 | (sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid))) |
1666 | root_blocks = ext4_r_blocks_count(sbi->s_es); | |
a30d542a | 1667 | |
6bc6e63f AK |
1668 | if (free_blocks - (nblocks + root_blocks + dirty_blocks) < |
1669 | EXT4_FREEBLOCKS_WATERMARK) { | |
5c791616 AK |
1670 | free_blocks = percpu_counter_sum(fbc); |
1671 | dirty_blocks = percpu_counter_sum(dbc); | |
6bc6e63f AK |
1672 | } |
1673 | if (free_blocks <= (root_blocks + dirty_blocks)) | |
16eb7295 AK |
1674 | /* we don't have free space */ |
1675 | return 0; | |
5c791616 | 1676 | |
6bc6e63f | 1677 | if (free_blocks - (root_blocks + dirty_blocks) < nblocks) |
5c791616 | 1678 | return free_blocks - (root_blocks + dirty_blocks); |
07031431 | 1679 | return nblocks; |
a30d542a | 1680 | } |
07031431 | 1681 | |
ac27a0ec DK |
1682 | |
1683 | /** | |
617ba13b | 1684 | * ext4_should_retry_alloc() |
ac27a0ec DK |
1685 | * @sb: super block |
1686 | * @retries number of attemps has been made | |
1687 | * | |
617ba13b | 1688 | * ext4_should_retry_alloc() is called when ENOSPC is returned, and if |
ac27a0ec DK |
1689 | * it is profitable to retry the operation, this function will wait |
1690 | * for the current or commiting transaction to complete, and then | |
1691 | * return TRUE. | |
1692 | * | |
1693 | * if the total number of retries exceed three times, return FALSE. | |
1694 | */ | |
617ba13b | 1695 | int ext4_should_retry_alloc(struct super_block *sb, int *retries) |
ac27a0ec | 1696 | { |
07031431 | 1697 | if (!ext4_has_free_blocks(EXT4_SB(sb), 1) || (*retries)++ > 3) |
ac27a0ec DK |
1698 | return 0; |
1699 | ||
1700 | jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id); | |
1701 | ||
dab291af | 1702 | return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal); |
ac27a0ec DK |
1703 | } |
1704 | ||
1705 | /** | |
654b4908 AK |
1706 | * ext4_old_new_blocks() -- core block bitmap based block allocation function |
1707 | * | |
ac27a0ec DK |
1708 | * @handle: handle to this transaction |
1709 | * @inode: file inode | |
1710 | * @goal: given target block(filesystem wide) | |
1711 | * @count: target number of blocks to allocate | |
1712 | * @errp: error code | |
1713 | * | |
654b4908 AK |
1714 | * ext4_old_new_blocks uses a goal block to assist allocation and look up |
1715 | * the block bitmap directly to do block allocation. It tries to | |
1716 | * allocate block(s) from the block group contains the goal block first. If | |
1717 | * that fails, it will try to allocate block(s) from other block groups | |
1718 | * without any specific goal block. | |
1719 | * | |
1720 | * This function is called when -o nomballoc mount option is enabled | |
ac27a0ec DK |
1721 | * |
1722 | */ | |
654b4908 | 1723 | ext4_fsblk_t ext4_old_new_blocks(handle_t *handle, struct inode *inode, |
617ba13b | 1724 | ext4_fsblk_t goal, unsigned long *count, int *errp) |
ac27a0ec DK |
1725 | { |
1726 | struct buffer_head *bitmap_bh = NULL; | |
1727 | struct buffer_head *gdp_bh; | |
fd2d4291 AM |
1728 | ext4_group_t group_no; |
1729 | ext4_group_t goal_group; | |
617ba13b MC |
1730 | ext4_grpblk_t grp_target_blk; /* blockgroup relative goal block */ |
1731 | ext4_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/ | |
1732 | ext4_fsblk_t ret_block; /* filesyetem-wide allocated block */ | |
fd2d4291 | 1733 | ext4_group_t bgi; /* blockgroup iteration index */ |
ac27a0ec DK |
1734 | int fatal = 0, err; |
1735 | int performed_allocation = 0; | |
617ba13b | 1736 | ext4_grpblk_t free_blocks; /* number of free blocks in a group */ |
ac27a0ec | 1737 | struct super_block *sb; |
617ba13b MC |
1738 | struct ext4_group_desc *gdp; |
1739 | struct ext4_super_block *es; | |
1740 | struct ext4_sb_info *sbi; | |
1741 | struct ext4_reserve_window_node *my_rsv = NULL; | |
1742 | struct ext4_block_alloc_info *block_i; | |
ac27a0ec | 1743 | unsigned short windowsz = 0; |
fd2d4291 | 1744 | ext4_group_t ngroups; |
ac27a0ec DK |
1745 | unsigned long num = *count; |
1746 | ||
ac27a0ec DK |
1747 | sb = inode->i_sb; |
1748 | if (!sb) { | |
07031431 | 1749 | *errp = -ENODEV; |
4776004f | 1750 | printk(KERN_ERR "ext4_new_block: nonexistent superblock"); |
ac27a0ec DK |
1751 | return 0; |
1752 | } | |
1753 | ||
07031431 | 1754 | sbi = EXT4_SB(sb); |
d2a17637 MC |
1755 | if (!EXT4_I(inode)->i_delalloc_reserved_flag) { |
1756 | /* | |
1757 | * With delalloc we already reserved the blocks | |
1758 | */ | |
030ba6bc AK |
1759 | while (*count && ext4_claim_free_blocks(sbi, *count)) { |
1760 | /* let others to free the space */ | |
1761 | yield(); | |
1762 | *count = *count >> 1; | |
1763 | } | |
1764 | if (!*count) { | |
a30d542a AK |
1765 | *errp = -ENOSPC; |
1766 | return 0; /*return with ENOSPC error */ | |
1767 | } | |
030ba6bc | 1768 | num = *count; |
07031431 | 1769 | } |
ac27a0ec DK |
1770 | /* |
1771 | * Check quota for allocation of this block. | |
1772 | */ | |
1773 | if (DQUOT_ALLOC_BLOCK(inode, num)) { | |
1774 | *errp = -EDQUOT; | |
1775 | return 0; | |
1776 | } | |
1777 | ||
617ba13b MC |
1778 | sbi = EXT4_SB(sb); |
1779 | es = EXT4_SB(sb)->s_es; | |
c549a95d | 1780 | ext4_debug("goal=%llu.\n", goal); |
ac27a0ec DK |
1781 | /* |
1782 | * Allocate a block from reservation only when | |
1783 | * filesystem is mounted with reservation(default,-o reservation), and | |
1784 | * it's a regular file, and | |
1785 | * the desired window size is greater than 0 (One could use ioctl | |
617ba13b | 1786 | * command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off |
ac27a0ec DK |
1787 | * reservation on that particular file) |
1788 | */ | |
617ba13b | 1789 | block_i = EXT4_I(inode)->i_block_alloc_info; |
ac27a0ec DK |
1790 | if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0)) |
1791 | my_rsv = &block_i->rsv_window_node; | |
1792 | ||
ac27a0ec DK |
1793 | /* |
1794 | * First, test whether the goal block is free. | |
1795 | */ | |
1796 | if (goal < le32_to_cpu(es->s_first_data_block) || | |
bd81d8ee | 1797 | goal >= ext4_blocks_count(es)) |
ac27a0ec | 1798 | goal = le32_to_cpu(es->s_first_data_block); |
3a5b2ecd | 1799 | ext4_get_group_no_and_offset(sb, goal, &group_no, &grp_target_blk); |
ac27a0ec DK |
1800 | goal_group = group_no; |
1801 | retry_alloc: | |
617ba13b | 1802 | gdp = ext4_get_group_desc(sb, group_no, &gdp_bh); |
ac27a0ec DK |
1803 | if (!gdp) |
1804 | goto io_error; | |
1805 | ||
1806 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); | |
ac27a0ec DK |
1807 | |
1808 | if (free_blocks > 0) { | |
68629f29 AK |
1809 | /* |
1810 | * try to allocate with group target block | |
1811 | * in the goal group. If we have low free_blocks | |
1812 | * count turn off reservation | |
1813 | */ | |
1814 | if (my_rsv && (free_blocks < windowsz) | |
1815 | && (rsv_is_empty(&my_rsv->rsv_window))) | |
1816 | my_rsv = NULL; | |
1817 | ||
574ca174 | 1818 | bitmap_bh = ext4_read_block_bitmap(sb, group_no); |
ac27a0ec DK |
1819 | if (!bitmap_bh) |
1820 | goto io_error; | |
617ba13b | 1821 | grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, |
ac27a0ec DK |
1822 | group_no, bitmap_bh, grp_target_blk, |
1823 | my_rsv, &num, &fatal); | |
1824 | if (fatal) | |
1825 | goto out; | |
1826 | if (grp_alloc_blk >= 0) | |
1827 | goto allocated; | |
1828 | } | |
1829 | ||
617ba13b | 1830 | ngroups = EXT4_SB(sb)->s_groups_count; |
ac27a0ec DK |
1831 | smp_rmb(); |
1832 | ||
1833 | /* | |
1834 | * Now search the rest of the groups. We assume that | |
144704e5 | 1835 | * group_no and gdp correctly point to the last group visited. |
ac27a0ec DK |
1836 | */ |
1837 | for (bgi = 0; bgi < ngroups; bgi++) { | |
1838 | group_no++; | |
1839 | if (group_no >= ngroups) | |
1840 | group_no = 0; | |
617ba13b | 1841 | gdp = ext4_get_group_desc(sb, group_no, &gdp_bh); |
341cee43 HD |
1842 | if (!gdp) |
1843 | goto io_error; | |
ac27a0ec DK |
1844 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); |
1845 | /* | |
1846 | * skip this group if the number of | |
1847 | * free blocks is less than half of the reservation | |
1848 | * window size. | |
1849 | */ | |
68629f29 | 1850 | if (my_rsv && (free_blocks <= (windowsz/2))) |
ac27a0ec DK |
1851 | continue; |
1852 | ||
1853 | brelse(bitmap_bh); | |
574ca174 | 1854 | bitmap_bh = ext4_read_block_bitmap(sb, group_no); |
ac27a0ec DK |
1855 | if (!bitmap_bh) |
1856 | goto io_error; | |
1857 | /* | |
1858 | * try to allocate block(s) from this group, without a goal(-1). | |
1859 | */ | |
617ba13b | 1860 | grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, |
ac27a0ec DK |
1861 | group_no, bitmap_bh, -1, my_rsv, |
1862 | &num, &fatal); | |
1863 | if (fatal) | |
1864 | goto out; | |
1865 | if (grp_alloc_blk >= 0) | |
1866 | goto allocated; | |
1867 | } | |
1868 | /* | |
1869 | * We may end up a bogus ealier ENOSPC error due to | |
1870 | * filesystem is "full" of reservations, but | |
1871 | * there maybe indeed free blocks avaliable on disk | |
1872 | * In this case, we just forget about the reservations | |
1873 | * just do block allocation as without reservations. | |
1874 | */ | |
1875 | if (my_rsv) { | |
1876 | my_rsv = NULL; | |
cd16c8f7 | 1877 | windowsz = 0; |
ac27a0ec DK |
1878 | group_no = goal_group; |
1879 | goto retry_alloc; | |
1880 | } | |
1881 | /* No space left on the device */ | |
1882 | *errp = -ENOSPC; | |
1883 | goto out; | |
1884 | ||
1885 | allocated: | |
1886 | ||
c549a95d | 1887 | ext4_debug("using block group %lu(%d)\n", |
ac27a0ec DK |
1888 | group_no, gdp->bg_free_blocks_count); |
1889 | ||
1890 | BUFFER_TRACE(gdp_bh, "get_write_access"); | |
617ba13b | 1891 | fatal = ext4_journal_get_write_access(handle, gdp_bh); |
ac27a0ec DK |
1892 | if (fatal) |
1893 | goto out; | |
1894 | ||
617ba13b | 1895 | ret_block = grp_alloc_blk + ext4_group_first_block_no(sb, group_no); |
ac27a0ec | 1896 | |
8fadc143 | 1897 | if (in_range(ext4_block_bitmap(sb, gdp), ret_block, num) || |
29bc5b4f | 1898 | in_range(ext4_inode_bitmap(sb, gdp), ret_block, num) || |
8fadc143 | 1899 | in_range(ret_block, ext4_inode_table(sb, gdp), |
bd81d8ee | 1900 | EXT4_SB(sb)->s_itb_per_group) || |
8fadc143 | 1901 | in_range(ret_block + num - 1, ext4_inode_table(sb, gdp), |
cb47dce7 | 1902 | EXT4_SB(sb)->s_itb_per_group)) { |
617ba13b | 1903 | ext4_error(sb, "ext4_new_block", |
ac27a0ec | 1904 | "Allocating block in system zone - " |
2ae02107 | 1905 | "blocks from %llu, length %lu", |
ac27a0ec | 1906 | ret_block, num); |
519deca0 AK |
1907 | /* |
1908 | * claim_block marked the blocks we allocated | |
1909 | * as in use. So we may want to selectively | |
1910 | * mark some of the blocks as free | |
1911 | */ | |
1912 | goto retry_alloc; | |
cb47dce7 | 1913 | } |
ac27a0ec DK |
1914 | |
1915 | performed_allocation = 1; | |
1916 | ||
e23291b9 | 1917 | #ifdef CONFIG_JBD2_DEBUG |
ac27a0ec DK |
1918 | { |
1919 | struct buffer_head *debug_bh; | |
1920 | ||
1921 | /* Record bitmap buffer state in the newly allocated block */ | |
1922 | debug_bh = sb_find_get_block(sb, ret_block); | |
1923 | if (debug_bh) { | |
1924 | BUFFER_TRACE(debug_bh, "state when allocated"); | |
1925 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state"); | |
1926 | brelse(debug_bh); | |
1927 | } | |
1928 | } | |
1929 | jbd_lock_bh_state(bitmap_bh); | |
1930 | spin_lock(sb_bgl_lock(sbi, group_no)); | |
1931 | if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) { | |
1932 | int i; | |
1933 | ||
1934 | for (i = 0; i < num; i++) { | |
617ba13b | 1935 | if (ext4_test_bit(grp_alloc_blk+i, |
ac27a0ec | 1936 | bh2jh(bitmap_bh)->b_committed_data)) { |
4776004f TT |
1937 | printk(KERN_ERR "%s: block was unexpectedly " |
1938 | "set in b_committed_data\n", __func__); | |
ac27a0ec DK |
1939 | } |
1940 | } | |
1941 | } | |
617ba13b | 1942 | ext4_debug("found bit %d\n", grp_alloc_blk); |
ac27a0ec DK |
1943 | spin_unlock(sb_bgl_lock(sbi, group_no)); |
1944 | jbd_unlock_bh_state(bitmap_bh); | |
1945 | #endif | |
1946 | ||
bd81d8ee | 1947 | if (ret_block + num - 1 >= ext4_blocks_count(es)) { |
617ba13b | 1948 | ext4_error(sb, "ext4_new_block", |
2ae02107 | 1949 | "block(%llu) >= blocks count(%llu) - " |
3a5b2ecd | 1950 | "block_group = %lu, es == %p ", ret_block, |
bd81d8ee | 1951 | ext4_blocks_count(es), group_no, es); |
ac27a0ec DK |
1952 | goto out; |
1953 | } | |
1954 | ||
1955 | /* | |
1956 | * It is up to the caller to add the new buffer to a journal | |
1957 | * list of some description. We don't know in advance whether | |
1958 | * the caller wants to use it as metadata or data. | |
1959 | */ | |
ac27a0ec | 1960 | spin_lock(sb_bgl_lock(sbi, group_no)); |
717d50e4 AD |
1961 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) |
1962 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); | |
e8546d06 | 1963 | le16_add_cpu(&gdp->bg_free_blocks_count, -num); |
717d50e4 | 1964 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group_no, gdp); |
ac27a0ec | 1965 | spin_unlock(sb_bgl_lock(sbi, group_no)); |
6bc6e63f AK |
1966 | percpu_counter_sub(&sbi->s_freeblocks_counter, num); |
1967 | /* | |
1968 | * Now reduce the dirty block count also. Should not go negative | |
1969 | */ | |
1970 | if (!EXT4_I(inode)->i_delalloc_reserved_flag) | |
1971 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, *count); | |
1972 | else | |
1973 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, num); | |
772cb7c8 JS |
1974 | if (sbi->s_log_groups_per_flex) { |
1975 | ext4_group_t flex_group = ext4_flex_group(sbi, group_no); | |
1976 | spin_lock(sb_bgl_lock(sbi, flex_group)); | |
1977 | sbi->s_flex_groups[flex_group].free_blocks -= num; | |
1978 | spin_unlock(sb_bgl_lock(sbi, flex_group)); | |
1979 | } | |
1980 | ||
ac27a0ec | 1981 | BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor"); |
617ba13b | 1982 | err = ext4_journal_dirty_metadata(handle, gdp_bh); |
ac27a0ec DK |
1983 | if (!fatal) |
1984 | fatal = err; | |
1985 | ||
1986 | sb->s_dirt = 1; | |
1987 | if (fatal) | |
1988 | goto out; | |
1989 | ||
1990 | *errp = 0; | |
1991 | brelse(bitmap_bh); | |
1992 | DQUOT_FREE_BLOCK(inode, *count-num); | |
1993 | *count = num; | |
1994 | return ret_block; | |
1995 | ||
1996 | io_error: | |
1997 | *errp = -EIO; | |
1998 | out: | |
1999 | if (fatal) { | |
2000 | *errp = fatal; | |
617ba13b | 2001 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
2002 | } |
2003 | /* | |
2004 | * Undo the block allocation | |
2005 | */ | |
2006 | if (!performed_allocation) | |
2007 | DQUOT_FREE_BLOCK(inode, *count); | |
2008 | brelse(bitmap_bh); | |
2009 | return 0; | |
2010 | } | |
2011 | ||
654b4908 | 2012 | #define EXT4_META_BLOCK 0x1 |
c9de560d | 2013 | |
654b4908 | 2014 | static ext4_fsblk_t do_blk_alloc(handle_t *handle, struct inode *inode, |
7061eba7 | 2015 | ext4_lblk_t iblock, ext4_fsblk_t goal, |
654b4908 | 2016 | unsigned long *count, int *errp, int flags) |
ac27a0ec | 2017 | { |
c9de560d AT |
2018 | struct ext4_allocation_request ar; |
2019 | ext4_fsblk_t ret; | |
ac27a0ec | 2020 | |
c9de560d | 2021 | if (!test_opt(inode->i_sb, MBALLOC)) { |
654b4908 | 2022 | return ext4_old_new_blocks(handle, inode, goal, count, errp); |
c9de560d AT |
2023 | } |
2024 | ||
2025 | memset(&ar, 0, sizeof(ar)); | |
7061eba7 | 2026 | /* Fill with neighbour allocated blocks */ |
7061eba7 | 2027 | |
c9de560d AT |
2028 | ar.inode = inode; |
2029 | ar.goal = goal; | |
2030 | ar.len = *count; | |
7061eba7 | 2031 | ar.logical = iblock; |
654b4908 AK |
2032 | |
2033 | if (S_ISREG(inode->i_mode) && !(flags & EXT4_META_BLOCK)) | |
2034 | /* enable in-core preallocation for data block allocation */ | |
7061eba7 AK |
2035 | ar.flags = EXT4_MB_HINT_DATA; |
2036 | else | |
2037 | /* disable in-core preallocation for non-regular files */ | |
2038 | ar.flags = 0; | |
654b4908 | 2039 | |
c9de560d AT |
2040 | ret = ext4_mb_new_blocks(handle, &ar, errp); |
2041 | *count = ar.len; | |
2042 | return ret; | |
ac27a0ec DK |
2043 | } |
2044 | ||
654b4908 | 2045 | /* |
d2a17637 | 2046 | * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks |
654b4908 AK |
2047 | * |
2048 | * @handle: handle to this transaction | |
2049 | * @inode: file inode | |
2050 | * @goal: given target block(filesystem wide) | |
d2a17637 | 2051 | * @count: total number of blocks need |
654b4908 AK |
2052 | * @errp: error code |
2053 | * | |
d2a17637 MC |
2054 | * Return 1st allocated block numberon success, *count stores total account |
2055 | * error stores in errp pointer | |
654b4908 | 2056 | */ |
d2a17637 MC |
2057 | ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode, |
2058 | ext4_fsblk_t goal, unsigned long *count, int *errp) | |
654b4908 | 2059 | { |
d2a17637 MC |
2060 | ext4_fsblk_t ret; |
2061 | ret = do_blk_alloc(handle, inode, 0, goal, | |
2062 | count, errp, EXT4_META_BLOCK); | |
2063 | /* | |
2064 | * Account for the allocated meta blocks | |
2065 | */ | |
166348dd | 2066 | if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) { |
d2a17637 MC |
2067 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
2068 | EXT4_I(inode)->i_allocated_meta_blocks += *count; | |
2069 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
2070 | } | |
2071 | return ret; | |
654b4908 AK |
2072 | } |
2073 | ||
2074 | /* | |
d2a17637 | 2075 | * ext4_new_meta_block() -- allocate block for meta data (indexing) blocks |
654b4908 AK |
2076 | * |
2077 | * @handle: handle to this transaction | |
2078 | * @inode: file inode | |
2079 | * @goal: given target block(filesystem wide) | |
654b4908 AK |
2080 | * @errp: error code |
2081 | * | |
d2a17637 | 2082 | * Return allocated block number on success |
654b4908 | 2083 | */ |
d2a17637 MC |
2084 | ext4_fsblk_t ext4_new_meta_block(handle_t *handle, struct inode *inode, |
2085 | ext4_fsblk_t goal, int *errp) | |
654b4908 | 2086 | { |
d2a17637 MC |
2087 | unsigned long count = 1; |
2088 | return ext4_new_meta_blocks(handle, inode, goal, &count, errp); | |
654b4908 AK |
2089 | } |
2090 | ||
2091 | /* | |
2092 | * ext4_new_blocks() -- allocate data blocks | |
2093 | * | |
2094 | * @handle: handle to this transaction | |
2095 | * @inode: file inode | |
2096 | * @goal: given target block(filesystem wide) | |
2097 | * @count: total number of blocks need | |
2098 | * @errp: error code | |
2099 | * | |
2100 | * Return 1st allocated block numberon success, *count stores total account | |
2101 | * error stores in errp pointer | |
2102 | */ | |
2103 | ||
2104 | ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode, | |
2105 | ext4_lblk_t iblock, ext4_fsblk_t goal, | |
2106 | unsigned long *count, int *errp) | |
2107 | { | |
2108 | return do_blk_alloc(handle, inode, iblock, goal, count, errp, 0); | |
2109 | } | |
c9de560d | 2110 | |
ac27a0ec | 2111 | /** |
617ba13b | 2112 | * ext4_count_free_blocks() -- count filesystem free blocks |
ac27a0ec DK |
2113 | * @sb: superblock |
2114 | * | |
2115 | * Adds up the number of free blocks from each block group. | |
2116 | */ | |
617ba13b | 2117 | ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb) |
ac27a0ec | 2118 | { |
617ba13b MC |
2119 | ext4_fsblk_t desc_count; |
2120 | struct ext4_group_desc *gdp; | |
fd2d4291 AM |
2121 | ext4_group_t i; |
2122 | ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; | |
617ba13b MC |
2123 | #ifdef EXT4FS_DEBUG |
2124 | struct ext4_super_block *es; | |
2125 | ext4_fsblk_t bitmap_count; | |
ac27a0ec DK |
2126 | unsigned long x; |
2127 | struct buffer_head *bitmap_bh = NULL; | |
2128 | ||
617ba13b | 2129 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
2130 | desc_count = 0; |
2131 | bitmap_count = 0; | |
2132 | gdp = NULL; | |
2133 | ||
2134 | smp_rmb(); | |
2135 | for (i = 0; i < ngroups; i++) { | |
617ba13b | 2136 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
2137 | if (!gdp) |
2138 | continue; | |
2139 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); | |
2140 | brelse(bitmap_bh); | |
574ca174 | 2141 | bitmap_bh = ext4_read_block_bitmap(sb, i); |
ac27a0ec DK |
2142 | if (bitmap_bh == NULL) |
2143 | continue; | |
2144 | ||
617ba13b | 2145 | x = ext4_count_free(bitmap_bh, sb->s_blocksize); |
fd2d4291 | 2146 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
ac27a0ec DK |
2147 | i, le16_to_cpu(gdp->bg_free_blocks_count), x); |
2148 | bitmap_count += x; | |
2149 | } | |
2150 | brelse(bitmap_bh); | |
4776004f TT |
2151 | printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu" |
2152 | ", computed = %llu, %llu\n", ext4_free_blocks_count(es), | |
2153 | desc_count, bitmap_count); | |
ac27a0ec DK |
2154 | return bitmap_count; |
2155 | #else | |
2156 | desc_count = 0; | |
2157 | smp_rmb(); | |
2158 | for (i = 0; i < ngroups; i++) { | |
617ba13b | 2159 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
2160 | if (!gdp) |
2161 | continue; | |
2162 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); | |
2163 | } | |
2164 | ||
2165 | return desc_count; | |
2166 | #endif | |
2167 | } | |
2168 | ||
fd2d4291 | 2169 | static inline int test_root(ext4_group_t a, int b) |
ac27a0ec DK |
2170 | { |
2171 | int num = b; | |
2172 | ||
2173 | while (a > num) | |
2174 | num *= b; | |
2175 | return num == a; | |
2176 | } | |
2177 | ||
fd2d4291 | 2178 | static int ext4_group_sparse(ext4_group_t group) |
ac27a0ec DK |
2179 | { |
2180 | if (group <= 1) | |
2181 | return 1; | |
2182 | if (!(group & 1)) | |
2183 | return 0; | |
2184 | return (test_root(group, 7) || test_root(group, 5) || | |
2185 | test_root(group, 3)); | |
2186 | } | |
2187 | ||
2188 | /** | |
617ba13b | 2189 | * ext4_bg_has_super - number of blocks used by the superblock in group |
ac27a0ec DK |
2190 | * @sb: superblock for filesystem |
2191 | * @group: group number to check | |
2192 | * | |
2193 | * Return the number of blocks used by the superblock (primary or backup) | |
2194 | * in this group. Currently this will be only 0 or 1. | |
2195 | */ | |
fd2d4291 | 2196 | int ext4_bg_has_super(struct super_block *sb, ext4_group_t group) |
ac27a0ec | 2197 | { |
617ba13b MC |
2198 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
2199 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) && | |
2200 | !ext4_group_sparse(group)) | |
ac27a0ec DK |
2201 | return 0; |
2202 | return 1; | |
2203 | } | |
2204 | ||
fd2d4291 AM |
2205 | static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, |
2206 | ext4_group_t group) | |
ac27a0ec | 2207 | { |
617ba13b | 2208 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); |
fd2d4291 AM |
2209 | ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb); |
2210 | ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1; | |
ac27a0ec DK |
2211 | |
2212 | if (group == first || group == first + 1 || group == last) | |
2213 | return 1; | |
2214 | return 0; | |
2215 | } | |
2216 | ||
fd2d4291 AM |
2217 | static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, |
2218 | ext4_group_t group) | |
ac27a0ec | 2219 | { |
859cb936 | 2220 | return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0; |
ac27a0ec DK |
2221 | } |
2222 | ||
2223 | /** | |
617ba13b | 2224 | * ext4_bg_num_gdb - number of blocks used by the group table in group |
ac27a0ec DK |
2225 | * @sb: superblock for filesystem |
2226 | * @group: group number to check | |
2227 | * | |
2228 | * Return the number of blocks used by the group descriptor table | |
2229 | * (primary or backup) in this group. In the future there may be a | |
2230 | * different number of descriptor blocks in each group. | |
2231 | */ | |
fd2d4291 | 2232 | unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group) |
ac27a0ec DK |
2233 | { |
2234 | unsigned long first_meta_bg = | |
617ba13b MC |
2235 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); |
2236 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); | |
ac27a0ec | 2237 | |
617ba13b | 2238 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) || |
ac27a0ec | 2239 | metagroup < first_meta_bg) |
af5bc92d | 2240 | return ext4_bg_num_gdb_nometa(sb, group); |
ac27a0ec | 2241 | |
617ba13b | 2242 | return ext4_bg_num_gdb_meta(sb,group); |
ac27a0ec DK |
2243 | |
2244 | } |