]>
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> |
617ba13b | 18 | #include <linux/ext4_fs.h> |
dab291af | 19 | #include <linux/ext4_jbd2.h> |
ac27a0ec DK |
20 | #include <linux/quotaops.h> |
21 | #include <linux/buffer_head.h> | |
22 | ||
717d50e4 | 23 | #include "group.h" |
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 | ||
717d50e4 AD |
46 | /* Initializes an uninitialized block bitmap if given, and returns the |
47 | * number of blocks free in the group. */ | |
48 | unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh, | |
fd2d4291 | 49 | ext4_group_t block_group, struct ext4_group_desc *gdp) |
717d50e4 AD |
50 | { |
51 | unsigned long start; | |
52 | int bit, bit_max; | |
53 | unsigned free_blocks, group_blocks; | |
54 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
55 | ||
56 | if (bh) { | |
57 | J_ASSERT_BH(bh, buffer_locked(bh)); | |
58 | ||
59 | /* If checksum is bad mark all blocks used to prevent allocation | |
60 | * essentially implementing a per-group read-only flag. */ | |
61 | if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) { | |
62 | ext4_error(sb, __FUNCTION__, | |
fd2d4291 | 63 | "Checksum bad for group %lu\n", block_group); |
717d50e4 AD |
64 | gdp->bg_free_blocks_count = 0; |
65 | gdp->bg_free_inodes_count = 0; | |
66 | gdp->bg_itable_unused = 0; | |
67 | memset(bh->b_data, 0xff, sb->s_blocksize); | |
68 | return 0; | |
69 | } | |
70 | memset(bh->b_data, 0, sb->s_blocksize); | |
71 | } | |
72 | ||
73 | /* Check for superblock and gdt backups in this group */ | |
74 | bit_max = ext4_bg_has_super(sb, block_group); | |
75 | ||
76 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) || | |
77 | block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) * | |
78 | sbi->s_desc_per_block) { | |
79 | if (bit_max) { | |
80 | bit_max += ext4_bg_num_gdb(sb, block_group); | |
81 | bit_max += | |
82 | le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); | |
83 | } | |
84 | } else { /* For META_BG_BLOCK_GROUPS */ | |
85 | int group_rel = (block_group - | |
86 | le32_to_cpu(sbi->s_es->s_first_meta_bg)) % | |
87 | EXT4_DESC_PER_BLOCK(sb); | |
88 | if (group_rel == 0 || group_rel == 1 || | |
89 | (group_rel == EXT4_DESC_PER_BLOCK(sb) - 1)) | |
90 | bit_max += 1; | |
91 | } | |
92 | ||
93 | if (block_group == sbi->s_groups_count - 1) { | |
94 | /* | |
95 | * Even though mke2fs always initialize first and last group | |
96 | * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need | |
97 | * to make sure we calculate the right free blocks | |
98 | */ | |
99 | group_blocks = ext4_blocks_count(sbi->s_es) - | |
100 | le32_to_cpu(sbi->s_es->s_first_data_block) - | |
101 | (EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count -1)); | |
102 | } else { | |
103 | group_blocks = EXT4_BLOCKS_PER_GROUP(sb); | |
104 | } | |
105 | ||
106 | free_blocks = group_blocks - bit_max; | |
107 | ||
108 | if (bh) { | |
109 | for (bit = 0; bit < bit_max; bit++) | |
110 | ext4_set_bit(bit, bh->b_data); | |
111 | ||
112 | start = block_group * EXT4_BLOCKS_PER_GROUP(sb) + | |
113 | le32_to_cpu(sbi->s_es->s_first_data_block); | |
114 | ||
115 | /* Set bits for block and inode bitmaps, and inode table */ | |
116 | ext4_set_bit(ext4_block_bitmap(sb, gdp) - start, bh->b_data); | |
117 | ext4_set_bit(ext4_inode_bitmap(sb, gdp) - start, bh->b_data); | |
5272f837 | 118 | for (bit = (ext4_inode_table(sb, gdp) - start), |
717d50e4 AD |
119 | bit_max = bit + sbi->s_itb_per_group; bit < bit_max; bit++) |
120 | ext4_set_bit(bit, bh->b_data); | |
121 | ||
122 | /* | |
123 | * Also if the number of blocks within the group is | |
124 | * less than the blocksize * 8 ( which is the size | |
125 | * of bitmap ), set rest of the block bitmap to 1 | |
126 | */ | |
127 | mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data); | |
128 | } | |
129 | ||
130 | return free_blocks - sbi->s_itb_per_group - 2; | |
131 | } | |
132 | ||
133 | ||
ac27a0ec DK |
134 | /* |
135 | * The free blocks are managed by bitmaps. A file system contains several | |
136 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
137 | * block for inodes, N blocks for the inode table and data blocks. | |
138 | * | |
139 | * The file system contains group descriptors which are located after the | |
140 | * super block. Each descriptor contains the number of the bitmap block and | |
141 | * the free blocks count in the block. The descriptors are loaded in memory | |
e627432c | 142 | * when a file system is mounted (see ext4_fill_super). |
ac27a0ec DK |
143 | */ |
144 | ||
145 | ||
146 | #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1) | |
147 | ||
148 | /** | |
617ba13b | 149 | * ext4_get_group_desc() -- load group descriptor from disk |
ac27a0ec DK |
150 | * @sb: super block |
151 | * @block_group: given block group | |
152 | * @bh: pointer to the buffer head to store the block | |
153 | * group descriptor | |
154 | */ | |
617ba13b | 155 | struct ext4_group_desc * ext4_get_group_desc(struct super_block * sb, |
fd2d4291 | 156 | ext4_group_t block_group, |
ac27a0ec DK |
157 | struct buffer_head ** bh) |
158 | { | |
159 | unsigned long group_desc; | |
160 | unsigned long offset; | |
617ba13b MC |
161 | struct ext4_group_desc * desc; |
162 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
ac27a0ec DK |
163 | |
164 | if (block_group >= sbi->s_groups_count) { | |
617ba13b | 165 | ext4_error (sb, "ext4_get_group_desc", |
ac27a0ec | 166 | "block_group >= groups_count - " |
fd2d4291 | 167 | "block_group = %lu, groups_count = %lu", |
ac27a0ec DK |
168 | block_group, sbi->s_groups_count); |
169 | ||
170 | return NULL; | |
171 | } | |
172 | smp_rmb(); | |
173 | ||
617ba13b MC |
174 | group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); |
175 | offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); | |
ac27a0ec | 176 | if (!sbi->s_group_desc[group_desc]) { |
617ba13b | 177 | ext4_error (sb, "ext4_get_group_desc", |
ac27a0ec | 178 | "Group descriptor not loaded - " |
fd2d4291 | 179 | "block_group = %lu, group_desc = %lu, desc = %lu", |
ac27a0ec DK |
180 | block_group, group_desc, offset); |
181 | return NULL; | |
182 | } | |
183 | ||
0d1ee42f AR |
184 | desc = (struct ext4_group_desc *)( |
185 | (__u8 *)sbi->s_group_desc[group_desc]->b_data + | |
186 | offset * EXT4_DESC_SIZE(sb)); | |
ac27a0ec DK |
187 | if (bh) |
188 | *bh = sbi->s_group_desc[group_desc]; | |
0d1ee42f | 189 | return desc; |
ac27a0ec DK |
190 | } |
191 | ||
192 | /** | |
193 | * read_block_bitmap() | |
194 | * @sb: super block | |
195 | * @block_group: given block group | |
196 | * | |
197 | * Read the bitmap for a given block_group, reading into the specified | |
198 | * slot in the superblock's bitmap cache. | |
199 | * | |
200 | * Return buffer_head on success or NULL in case of failure. | |
201 | */ | |
717d50e4 | 202 | struct buffer_head * |
fd2d4291 | 203 | read_block_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 204 | { |
617ba13b | 205 | struct ext4_group_desc * desc; |
ac27a0ec | 206 | struct buffer_head * bh = NULL; |
7c9e69fa | 207 | ext4_fsblk_t bitmap_blk; |
ac27a0ec | 208 | |
717d50e4 | 209 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 210 | if (!desc) |
7c9e69fa AK |
211 | return NULL; |
212 | bitmap_blk = ext4_block_bitmap(sb, desc); | |
717d50e4 AD |
213 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
214 | bh = sb_getblk(sb, bitmap_blk); | |
215 | if (!buffer_uptodate(bh)) { | |
216 | lock_buffer(bh); | |
217 | if (!buffer_uptodate(bh)) { | |
218 | ext4_init_block_bitmap(sb, bh, block_group, | |
219 | desc); | |
220 | set_buffer_uptodate(bh); | |
221 | } | |
222 | unlock_buffer(bh); | |
223 | } | |
224 | } else { | |
225 | bh = sb_bread(sb, bitmap_blk); | |
226 | } | |
ac27a0ec | 227 | if (!bh) |
7c9e69fa | 228 | ext4_error (sb, __FUNCTION__, |
ac27a0ec | 229 | "Cannot read block bitmap - " |
fd2d4291 | 230 | "block_group = %lu, block_bitmap = %llu", |
7c9e69fa | 231 | block_group, bitmap_blk); |
ac27a0ec DK |
232 | return bh; |
233 | } | |
234 | /* | |
235 | * The reservation window structure operations | |
236 | * -------------------------------------------- | |
237 | * Operations include: | |
238 | * dump, find, add, remove, is_empty, find_next_reservable_window, etc. | |
239 | * | |
240 | * We use a red-black tree to represent per-filesystem reservation | |
241 | * windows. | |
242 | * | |
243 | */ | |
244 | ||
245 | /** | |
246 | * __rsv_window_dump() -- Dump the filesystem block allocation reservation map | |
247 | * @rb_root: root of per-filesystem reservation rb tree | |
248 | * @verbose: verbose mode | |
249 | * @fn: function which wishes to dump the reservation map | |
250 | * | |
251 | * If verbose is turned on, it will print the whole block reservation | |
252 | * windows(start, end). Otherwise, it will only print out the "bad" windows, | |
253 | * those windows that overlap with their immediate neighbors. | |
254 | */ | |
255 | #if 1 | |
256 | static void __rsv_window_dump(struct rb_root *root, int verbose, | |
257 | const char *fn) | |
258 | { | |
259 | struct rb_node *n; | |
617ba13b | 260 | struct ext4_reserve_window_node *rsv, *prev; |
ac27a0ec DK |
261 | int bad; |
262 | ||
263 | restart: | |
264 | n = rb_first(root); | |
265 | bad = 0; | |
266 | prev = NULL; | |
267 | ||
268 | printk("Block Allocation Reservation Windows Map (%s):\n", fn); | |
269 | while (n) { | |
b78a657f | 270 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
271 | if (verbose) |
272 | printk("reservation window 0x%p " | |
2ae02107 | 273 | "start: %llu, end: %llu\n", |
ac27a0ec DK |
274 | rsv, rsv->rsv_start, rsv->rsv_end); |
275 | if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) { | |
276 | printk("Bad reservation %p (start >= end)\n", | |
277 | rsv); | |
278 | bad = 1; | |
279 | } | |
280 | if (prev && prev->rsv_end >= rsv->rsv_start) { | |
281 | printk("Bad reservation %p (prev->end >= start)\n", | |
282 | rsv); | |
283 | bad = 1; | |
284 | } | |
285 | if (bad) { | |
286 | if (!verbose) { | |
287 | printk("Restarting reservation walk in verbose mode\n"); | |
288 | verbose = 1; | |
289 | goto restart; | |
290 | } | |
291 | } | |
292 | n = rb_next(n); | |
293 | prev = rsv; | |
294 | } | |
295 | printk("Window map complete.\n"); | |
296 | if (bad) | |
297 | BUG(); | |
298 | } | |
299 | #define rsv_window_dump(root, verbose) \ | |
300 | __rsv_window_dump((root), (verbose), __FUNCTION__) | |
301 | #else | |
302 | #define rsv_window_dump(root, verbose) do {} while (0) | |
303 | #endif | |
304 | ||
305 | /** | |
306 | * goal_in_my_reservation() | |
307 | * @rsv: inode's reservation window | |
308 | * @grp_goal: given goal block relative to the allocation block group | |
309 | * @group: the current allocation block group | |
310 | * @sb: filesystem super block | |
311 | * | |
312 | * Test if the given goal block (group relative) is within the file's | |
313 | * own block reservation window range. | |
314 | * | |
315 | * If the reservation window is outside the goal allocation group, return 0; | |
316 | * grp_goal (given goal block) could be -1, which means no specific | |
317 | * goal block. In this case, always return 1. | |
318 | * If the goal block is within the reservation window, return 1; | |
319 | * otherwise, return 0; | |
320 | */ | |
321 | static int | |
617ba13b | 322 | goal_in_my_reservation(struct ext4_reserve_window *rsv, ext4_grpblk_t grp_goal, |
fd2d4291 | 323 | ext4_group_t group, struct super_block *sb) |
ac27a0ec | 324 | { |
617ba13b | 325 | ext4_fsblk_t group_first_block, group_last_block; |
ac27a0ec | 326 | |
617ba13b MC |
327 | group_first_block = ext4_group_first_block_no(sb, group); |
328 | group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
329 | |
330 | if ((rsv->_rsv_start > group_last_block) || | |
331 | (rsv->_rsv_end < group_first_block)) | |
332 | return 0; | |
333 | if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start) | |
334 | || (grp_goal + group_first_block > rsv->_rsv_end))) | |
335 | return 0; | |
336 | return 1; | |
337 | } | |
338 | ||
339 | /** | |
340 | * search_reserve_window() | |
341 | * @rb_root: root of reservation tree | |
342 | * @goal: target allocation block | |
343 | * | |
344 | * Find the reserved window which includes the goal, or the previous one | |
345 | * if the goal is not in any window. | |
346 | * Returns NULL if there are no windows or if all windows start after the goal. | |
347 | */ | |
617ba13b MC |
348 | static struct ext4_reserve_window_node * |
349 | search_reserve_window(struct rb_root *root, ext4_fsblk_t goal) | |
ac27a0ec DK |
350 | { |
351 | struct rb_node *n = root->rb_node; | |
617ba13b | 352 | struct ext4_reserve_window_node *rsv; |
ac27a0ec DK |
353 | |
354 | if (!n) | |
355 | return NULL; | |
356 | ||
357 | do { | |
617ba13b | 358 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
359 | |
360 | if (goal < rsv->rsv_start) | |
361 | n = n->rb_left; | |
362 | else if (goal > rsv->rsv_end) | |
363 | n = n->rb_right; | |
364 | else | |
365 | return rsv; | |
366 | } while (n); | |
367 | /* | |
368 | * We've fallen off the end of the tree: the goal wasn't inside | |
369 | * any particular node. OK, the previous node must be to one | |
370 | * side of the interval containing the goal. If it's the RHS, | |
371 | * we need to back up one. | |
372 | */ | |
373 | if (rsv->rsv_start > goal) { | |
374 | n = rb_prev(&rsv->rsv_node); | |
617ba13b | 375 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
376 | } |
377 | return rsv; | |
378 | } | |
379 | ||
380 | /** | |
617ba13b | 381 | * ext4_rsv_window_add() -- Insert a window to the block reservation rb tree. |
ac27a0ec DK |
382 | * @sb: super block |
383 | * @rsv: reservation window to add | |
384 | * | |
385 | * Must be called with rsv_lock hold. | |
386 | */ | |
617ba13b MC |
387 | void ext4_rsv_window_add(struct super_block *sb, |
388 | struct ext4_reserve_window_node *rsv) | |
ac27a0ec | 389 | { |
617ba13b | 390 | struct rb_root *root = &EXT4_SB(sb)->s_rsv_window_root; |
ac27a0ec | 391 | struct rb_node *node = &rsv->rsv_node; |
617ba13b | 392 | ext4_fsblk_t start = rsv->rsv_start; |
ac27a0ec DK |
393 | |
394 | struct rb_node ** p = &root->rb_node; | |
395 | struct rb_node * parent = NULL; | |
617ba13b | 396 | struct ext4_reserve_window_node *this; |
ac27a0ec DK |
397 | |
398 | while (*p) | |
399 | { | |
400 | parent = *p; | |
617ba13b | 401 | this = rb_entry(parent, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
402 | |
403 | if (start < this->rsv_start) | |
404 | p = &(*p)->rb_left; | |
405 | else if (start > this->rsv_end) | |
406 | p = &(*p)->rb_right; | |
407 | else { | |
408 | rsv_window_dump(root, 1); | |
409 | BUG(); | |
410 | } | |
411 | } | |
412 | ||
413 | rb_link_node(node, parent, p); | |
414 | rb_insert_color(node, root); | |
415 | } | |
416 | ||
417 | /** | |
617ba13b | 418 | * ext4_rsv_window_remove() -- unlink a window from the reservation rb tree |
ac27a0ec DK |
419 | * @sb: super block |
420 | * @rsv: reservation window to remove | |
421 | * | |
422 | * Mark the block reservation window as not allocated, and unlink it | |
423 | * from the filesystem reservation window rb tree. Must be called with | |
424 | * rsv_lock hold. | |
425 | */ | |
426 | static void rsv_window_remove(struct super_block *sb, | |
617ba13b | 427 | struct ext4_reserve_window_node *rsv) |
ac27a0ec | 428 | { |
617ba13b MC |
429 | rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
430 | rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; | |
ac27a0ec | 431 | rsv->rsv_alloc_hit = 0; |
617ba13b | 432 | rb_erase(&rsv->rsv_node, &EXT4_SB(sb)->s_rsv_window_root); |
ac27a0ec DK |
433 | } |
434 | ||
435 | /* | |
436 | * rsv_is_empty() -- Check if the reservation window is allocated. | |
437 | * @rsv: given reservation window to check | |
438 | * | |
617ba13b | 439 | * returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED. |
ac27a0ec | 440 | */ |
617ba13b | 441 | static inline int rsv_is_empty(struct ext4_reserve_window *rsv) |
ac27a0ec DK |
442 | { |
443 | /* a valid reservation end block could not be 0 */ | |
617ba13b | 444 | return rsv->_rsv_end == EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
ac27a0ec DK |
445 | } |
446 | ||
447 | /** | |
617ba13b | 448 | * ext4_init_block_alloc_info() |
ac27a0ec DK |
449 | * @inode: file inode structure |
450 | * | |
451 | * Allocate and initialize the reservation window structure, and | |
617ba13b | 452 | * link the window to the ext4 inode structure at last |
ac27a0ec DK |
453 | * |
454 | * The reservation window structure is only dynamically allocated | |
617ba13b MC |
455 | * and linked to ext4 inode the first time the open file |
456 | * needs a new block. So, before every ext4_new_block(s) call, for | |
ac27a0ec DK |
457 | * regular files, we should check whether the reservation window |
458 | * structure exists or not. In the latter case, this function is called. | |
459 | * Fail to do so will result in block reservation being turned off for that | |
460 | * open file. | |
461 | * | |
617ba13b | 462 | * This function is called from ext4_get_blocks_handle(), also called |
ac27a0ec DK |
463 | * when setting the reservation window size through ioctl before the file |
464 | * is open for write (needs block allocation). | |
465 | * | |
466 | * Needs truncate_mutex protection prior to call this function. | |
467 | */ | |
617ba13b | 468 | void ext4_init_block_alloc_info(struct inode *inode) |
ac27a0ec | 469 | { |
617ba13b MC |
470 | struct ext4_inode_info *ei = EXT4_I(inode); |
471 | struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info; | |
ac27a0ec DK |
472 | struct super_block *sb = inode->i_sb; |
473 | ||
474 | block_i = kmalloc(sizeof(*block_i), GFP_NOFS); | |
475 | if (block_i) { | |
617ba13b | 476 | struct ext4_reserve_window_node *rsv = &block_i->rsv_window_node; |
ac27a0ec | 477 | |
617ba13b MC |
478 | rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
479 | rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; | |
ac27a0ec DK |
480 | |
481 | /* | |
482 | * if filesystem is mounted with NORESERVATION, the goal | |
483 | * reservation window size is set to zero to indicate | |
484 | * block reservation is off | |
485 | */ | |
486 | if (!test_opt(sb, RESERVATION)) | |
487 | rsv->rsv_goal_size = 0; | |
488 | else | |
617ba13b | 489 | rsv->rsv_goal_size = EXT4_DEFAULT_RESERVE_BLOCKS; |
ac27a0ec DK |
490 | rsv->rsv_alloc_hit = 0; |
491 | block_i->last_alloc_logical_block = 0; | |
492 | block_i->last_alloc_physical_block = 0; | |
493 | } | |
494 | ei->i_block_alloc_info = block_i; | |
495 | } | |
496 | ||
497 | /** | |
617ba13b | 498 | * ext4_discard_reservation() |
ac27a0ec DK |
499 | * @inode: inode |
500 | * | |
501 | * Discard(free) block reservation window on last file close, or truncate | |
502 | * or at last iput(). | |
503 | * | |
504 | * It is being called in three cases: | |
617ba13b MC |
505 | * ext4_release_file(): last writer close the file |
506 | * ext4_clear_inode(): last iput(), when nobody link to this file. | |
507 | * ext4_truncate(): when the block indirect map is about to change. | |
ac27a0ec DK |
508 | * |
509 | */ | |
617ba13b | 510 | void ext4_discard_reservation(struct inode *inode) |
ac27a0ec | 511 | { |
617ba13b MC |
512 | struct ext4_inode_info *ei = EXT4_I(inode); |
513 | struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info; | |
514 | struct ext4_reserve_window_node *rsv; | |
515 | spinlock_t *rsv_lock = &EXT4_SB(inode->i_sb)->s_rsv_window_lock; | |
ac27a0ec DK |
516 | |
517 | if (!block_i) | |
518 | return; | |
519 | ||
520 | rsv = &block_i->rsv_window_node; | |
521 | if (!rsv_is_empty(&rsv->rsv_window)) { | |
522 | spin_lock(rsv_lock); | |
523 | if (!rsv_is_empty(&rsv->rsv_window)) | |
524 | rsv_window_remove(inode->i_sb, rsv); | |
525 | spin_unlock(rsv_lock); | |
526 | } | |
527 | } | |
528 | ||
529 | /** | |
617ba13b | 530 | * ext4_free_blocks_sb() -- Free given blocks and update quota |
ac27a0ec DK |
531 | * @handle: handle to this transaction |
532 | * @sb: super block | |
533 | * @block: start physcial block to free | |
534 | * @count: number of blocks to free | |
535 | * @pdquot_freed_blocks: pointer to quota | |
536 | */ | |
617ba13b MC |
537 | void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb, |
538 | ext4_fsblk_t block, unsigned long count, | |
ac27a0ec DK |
539 | unsigned long *pdquot_freed_blocks) |
540 | { | |
541 | struct buffer_head *bitmap_bh = NULL; | |
542 | struct buffer_head *gd_bh; | |
fd2d4291 | 543 | ext4_group_t block_group; |
617ba13b | 544 | ext4_grpblk_t bit; |
ac27a0ec DK |
545 | unsigned long i; |
546 | unsigned long overflow; | |
617ba13b MC |
547 | struct ext4_group_desc * desc; |
548 | struct ext4_super_block * es; | |
549 | struct ext4_sb_info *sbi; | |
ac27a0ec | 550 | int err = 0, ret; |
617ba13b | 551 | ext4_grpblk_t group_freed; |
ac27a0ec DK |
552 | |
553 | *pdquot_freed_blocks = 0; | |
617ba13b | 554 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
555 | es = sbi->s_es; |
556 | if (block < le32_to_cpu(es->s_first_data_block) || | |
557 | block + count < block || | |
bd81d8ee | 558 | block + count > ext4_blocks_count(es)) { |
617ba13b | 559 | ext4_error (sb, "ext4_free_blocks", |
ac27a0ec | 560 | "Freeing blocks not in datazone - " |
2ae02107 | 561 | "block = %llu, count = %lu", block, count); |
ac27a0ec DK |
562 | goto error_return; |
563 | } | |
564 | ||
bd81d8ee | 565 | ext4_debug ("freeing block(s) %llu-%llu\n", block, block + count - 1); |
ac27a0ec DK |
566 | |
567 | do_more: | |
568 | overflow = 0; | |
3a5b2ecd | 569 | ext4_get_group_no_and_offset(sb, block, &block_group, &bit); |
ac27a0ec DK |
570 | /* |
571 | * Check to see if we are freeing blocks across a group | |
572 | * boundary. | |
573 | */ | |
617ba13b MC |
574 | if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) { |
575 | overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb); | |
ac27a0ec DK |
576 | count -= overflow; |
577 | } | |
578 | brelse(bitmap_bh); | |
579 | bitmap_bh = read_block_bitmap(sb, block_group); | |
580 | if (!bitmap_bh) | |
581 | goto error_return; | |
617ba13b | 582 | desc = ext4_get_group_desc (sb, block_group, &gd_bh); |
ac27a0ec DK |
583 | if (!desc) |
584 | goto error_return; | |
585 | ||
8fadc143 AR |
586 | if (in_range(ext4_block_bitmap(sb, desc), block, count) || |
587 | in_range(ext4_inode_bitmap(sb, desc), block, count) || | |
588 | in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) || | |
589 | in_range(block + count - 1, ext4_inode_table(sb, desc), | |
bd81d8ee | 590 | sbi->s_itb_per_group)) |
617ba13b | 591 | ext4_error (sb, "ext4_free_blocks", |
ac27a0ec | 592 | "Freeing blocks in system zones - " |
2ae02107 | 593 | "Block = %llu, count = %lu", |
ac27a0ec DK |
594 | block, count); |
595 | ||
596 | /* | |
597 | * We are about to start releasing blocks in the bitmap, | |
598 | * so we need undo access. | |
599 | */ | |
600 | /* @@@ check errors */ | |
601 | BUFFER_TRACE(bitmap_bh, "getting undo access"); | |
617ba13b | 602 | err = ext4_journal_get_undo_access(handle, bitmap_bh); |
ac27a0ec DK |
603 | if (err) |
604 | goto error_return; | |
605 | ||
606 | /* | |
607 | * We are about to modify some metadata. Call the journal APIs | |
608 | * to unshare ->b_data if a currently-committing transaction is | |
609 | * using it | |
610 | */ | |
611 | BUFFER_TRACE(gd_bh, "get_write_access"); | |
617ba13b | 612 | err = ext4_journal_get_write_access(handle, gd_bh); |
ac27a0ec DK |
613 | if (err) |
614 | goto error_return; | |
615 | ||
616 | jbd_lock_bh_state(bitmap_bh); | |
617 | ||
618 | for (i = 0, group_freed = 0; i < count; i++) { | |
619 | /* | |
620 | * An HJ special. This is expensive... | |
621 | */ | |
e23291b9 | 622 | #ifdef CONFIG_JBD2_DEBUG |
ac27a0ec DK |
623 | jbd_unlock_bh_state(bitmap_bh); |
624 | { | |
625 | struct buffer_head *debug_bh; | |
626 | debug_bh = sb_find_get_block(sb, block + i); | |
627 | if (debug_bh) { | |
628 | BUFFER_TRACE(debug_bh, "Deleted!"); | |
629 | if (!bh2jh(bitmap_bh)->b_committed_data) | |
630 | BUFFER_TRACE(debug_bh, | |
631 | "No commited data in bitmap"); | |
632 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap"); | |
633 | __brelse(debug_bh); | |
634 | } | |
635 | } | |
636 | jbd_lock_bh_state(bitmap_bh); | |
637 | #endif | |
638 | if (need_resched()) { | |
639 | jbd_unlock_bh_state(bitmap_bh); | |
640 | cond_resched(); | |
641 | jbd_lock_bh_state(bitmap_bh); | |
642 | } | |
643 | /* @@@ This prevents newly-allocated data from being | |
644 | * freed and then reallocated within the same | |
645 | * transaction. | |
646 | * | |
647 | * Ideally we would want to allow that to happen, but to | |
dab291af | 648 | * do so requires making jbd2_journal_forget() capable of |
ac27a0ec DK |
649 | * revoking the queued write of a data block, which |
650 | * implies blocking on the journal lock. *forget() | |
651 | * cannot block due to truncate races. | |
652 | * | |
dab291af | 653 | * Eventually we can fix this by making jbd2_journal_forget() |
ac27a0ec DK |
654 | * return a status indicating whether or not it was able |
655 | * to revoke the buffer. On successful revoke, it is | |
656 | * safe not to set the allocation bit in the committed | |
657 | * bitmap, because we know that there is no outstanding | |
658 | * activity on the buffer any more and so it is safe to | |
659 | * reallocate it. | |
660 | */ | |
661 | BUFFER_TRACE(bitmap_bh, "set in b_committed_data"); | |
662 | J_ASSERT_BH(bitmap_bh, | |
663 | bh2jh(bitmap_bh)->b_committed_data != NULL); | |
617ba13b | 664 | ext4_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i, |
ac27a0ec DK |
665 | bh2jh(bitmap_bh)->b_committed_data); |
666 | ||
667 | /* | |
668 | * We clear the bit in the bitmap after setting the committed | |
669 | * data bit, because this is the reverse order to that which | |
670 | * the allocator uses. | |
671 | */ | |
672 | BUFFER_TRACE(bitmap_bh, "clear bit"); | |
617ba13b | 673 | if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group), |
ac27a0ec DK |
674 | bit + i, bitmap_bh->b_data)) { |
675 | jbd_unlock_bh_state(bitmap_bh); | |
617ba13b | 676 | ext4_error(sb, __FUNCTION__, |
2ae02107 | 677 | "bit already cleared for block %llu", |
bd81d8ee | 678 | (ext4_fsblk_t)(block + i)); |
ac27a0ec DK |
679 | jbd_lock_bh_state(bitmap_bh); |
680 | BUFFER_TRACE(bitmap_bh, "bit already cleared"); | |
681 | } else { | |
682 | group_freed++; | |
683 | } | |
684 | } | |
685 | jbd_unlock_bh_state(bitmap_bh); | |
686 | ||
687 | spin_lock(sb_bgl_lock(sbi, block_group)); | |
688 | desc->bg_free_blocks_count = | |
689 | cpu_to_le16(le16_to_cpu(desc->bg_free_blocks_count) + | |
690 | group_freed); | |
717d50e4 | 691 | desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc); |
ac27a0ec | 692 | spin_unlock(sb_bgl_lock(sbi, block_group)); |
aa0dff2d | 693 | percpu_counter_add(&sbi->s_freeblocks_counter, count); |
ac27a0ec DK |
694 | |
695 | /* We dirtied the bitmap block */ | |
696 | BUFFER_TRACE(bitmap_bh, "dirtied bitmap block"); | |
617ba13b | 697 | err = ext4_journal_dirty_metadata(handle, bitmap_bh); |
ac27a0ec DK |
698 | |
699 | /* And the group descriptor block */ | |
700 | BUFFER_TRACE(gd_bh, "dirtied group descriptor block"); | |
617ba13b | 701 | ret = ext4_journal_dirty_metadata(handle, gd_bh); |
ac27a0ec DK |
702 | if (!err) err = ret; |
703 | *pdquot_freed_blocks += group_freed; | |
704 | ||
705 | if (overflow && !err) { | |
706 | block += count; | |
707 | count = overflow; | |
708 | goto do_more; | |
709 | } | |
710 | sb->s_dirt = 1; | |
711 | error_return: | |
712 | brelse(bitmap_bh); | |
617ba13b | 713 | ext4_std_error(sb, err); |
ac27a0ec DK |
714 | return; |
715 | } | |
716 | ||
717 | /** | |
617ba13b | 718 | * ext4_free_blocks() -- Free given blocks and update quota |
ac27a0ec DK |
719 | * @handle: handle for this transaction |
720 | * @inode: inode | |
721 | * @block: start physical block to free | |
722 | * @count: number of blocks to count | |
723 | */ | |
617ba13b MC |
724 | void ext4_free_blocks(handle_t *handle, struct inode *inode, |
725 | ext4_fsblk_t block, unsigned long count) | |
ac27a0ec DK |
726 | { |
727 | struct super_block * sb; | |
728 | unsigned long dquot_freed_blocks; | |
729 | ||
730 | sb = inode->i_sb; | |
731 | if (!sb) { | |
617ba13b | 732 | printk ("ext4_free_blocks: nonexistent device"); |
ac27a0ec DK |
733 | return; |
734 | } | |
617ba13b | 735 | ext4_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks); |
ac27a0ec DK |
736 | if (dquot_freed_blocks) |
737 | DQUOT_FREE_BLOCK(inode, dquot_freed_blocks); | |
738 | return; | |
739 | } | |
740 | ||
741 | /** | |
617ba13b | 742 | * ext4_test_allocatable() |
ac27a0ec DK |
743 | * @nr: given allocation block group |
744 | * @bh: bufferhead contains the bitmap of the given block group | |
745 | * | |
617ba13b | 746 | * For ext4 allocations, we must not reuse any blocks which are |
ac27a0ec DK |
747 | * allocated in the bitmap buffer's "last committed data" copy. This |
748 | * prevents deletes from freeing up the page for reuse until we have | |
749 | * committed the delete transaction. | |
750 | * | |
751 | * If we didn't do this, then deleting something and reallocating it as | |
752 | * data would allow the old block to be overwritten before the | |
753 | * transaction committed (because we force data to disk before commit). | |
754 | * This would lead to corruption if we crashed between overwriting the | |
755 | * data and committing the delete. | |
756 | * | |
757 | * @@@ We may want to make this allocation behaviour conditional on | |
758 | * data-writes at some point, and disable it for metadata allocations or | |
759 | * sync-data inodes. | |
760 | */ | |
617ba13b | 761 | static int ext4_test_allocatable(ext4_grpblk_t nr, struct buffer_head *bh) |
ac27a0ec DK |
762 | { |
763 | int ret; | |
764 | struct journal_head *jh = bh2jh(bh); | |
765 | ||
617ba13b | 766 | if (ext4_test_bit(nr, bh->b_data)) |
ac27a0ec DK |
767 | return 0; |
768 | ||
769 | jbd_lock_bh_state(bh); | |
770 | if (!jh->b_committed_data) | |
771 | ret = 1; | |
772 | else | |
617ba13b | 773 | ret = !ext4_test_bit(nr, jh->b_committed_data); |
ac27a0ec DK |
774 | jbd_unlock_bh_state(bh); |
775 | return ret; | |
776 | } | |
777 | ||
778 | /** | |
779 | * bitmap_search_next_usable_block() | |
780 | * @start: the starting block (group relative) of the search | |
781 | * @bh: bufferhead contains the block group bitmap | |
782 | * @maxblocks: the ending block (group relative) of the reservation | |
783 | * | |
784 | * The bitmap search --- search forward alternately through the actual | |
785 | * bitmap on disk and the last-committed copy in journal, until we find a | |
786 | * bit free in both bitmaps. | |
787 | */ | |
617ba13b MC |
788 | static ext4_grpblk_t |
789 | bitmap_search_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh, | |
790 | ext4_grpblk_t maxblocks) | |
ac27a0ec | 791 | { |
617ba13b | 792 | ext4_grpblk_t next; |
ac27a0ec DK |
793 | struct journal_head *jh = bh2jh(bh); |
794 | ||
795 | while (start < maxblocks) { | |
617ba13b | 796 | next = ext4_find_next_zero_bit(bh->b_data, maxblocks, start); |
ac27a0ec DK |
797 | if (next >= maxblocks) |
798 | return -1; | |
617ba13b | 799 | if (ext4_test_allocatable(next, bh)) |
ac27a0ec DK |
800 | return next; |
801 | jbd_lock_bh_state(bh); | |
802 | if (jh->b_committed_data) | |
617ba13b | 803 | start = ext4_find_next_zero_bit(jh->b_committed_data, |
ac27a0ec DK |
804 | maxblocks, next); |
805 | jbd_unlock_bh_state(bh); | |
806 | } | |
807 | return -1; | |
808 | } | |
809 | ||
810 | /** | |
811 | * find_next_usable_block() | |
812 | * @start: the starting block (group relative) to find next | |
813 | * allocatable block in bitmap. | |
814 | * @bh: bufferhead contains the block group bitmap | |
815 | * @maxblocks: the ending block (group relative) for the search | |
816 | * | |
817 | * Find an allocatable block in a bitmap. We honor both the bitmap and | |
818 | * its last-committed copy (if that exists), and perform the "most | |
819 | * appropriate allocation" algorithm of looking for a free block near | |
820 | * the initial goal; then for a free byte somewhere in the bitmap; then | |
821 | * for any free bit in the bitmap. | |
822 | */ | |
617ba13b MC |
823 | static ext4_grpblk_t |
824 | find_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh, | |
825 | ext4_grpblk_t maxblocks) | |
ac27a0ec | 826 | { |
617ba13b | 827 | ext4_grpblk_t here, next; |
ac27a0ec DK |
828 | char *p, *r; |
829 | ||
830 | if (start > 0) { | |
831 | /* | |
832 | * The goal was occupied; search forward for a free | |
833 | * block within the next XX blocks. | |
834 | * | |
835 | * end_goal is more or less random, but it has to be | |
617ba13b | 836 | * less than EXT4_BLOCKS_PER_GROUP. Aligning up to the |
ac27a0ec DK |
837 | * next 64-bit boundary is simple.. |
838 | */ | |
617ba13b | 839 | ext4_grpblk_t end_goal = (start + 63) & ~63; |
ac27a0ec DK |
840 | if (end_goal > maxblocks) |
841 | end_goal = maxblocks; | |
617ba13b MC |
842 | here = ext4_find_next_zero_bit(bh->b_data, end_goal, start); |
843 | if (here < end_goal && ext4_test_allocatable(here, bh)) | |
ac27a0ec | 844 | return here; |
617ba13b | 845 | ext4_debug("Bit not found near goal\n"); |
ac27a0ec DK |
846 | } |
847 | ||
848 | here = start; | |
849 | if (here < 0) | |
850 | here = 0; | |
851 | ||
852 | p = ((char *)bh->b_data) + (here >> 3); | |
ec0837f2 | 853 | r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3)); |
ac27a0ec DK |
854 | next = (r - ((char *)bh->b_data)) << 3; |
855 | ||
617ba13b | 856 | if (next < maxblocks && next >= start && ext4_test_allocatable(next, bh)) |
ac27a0ec DK |
857 | return next; |
858 | ||
859 | /* | |
860 | * The bitmap search --- search forward alternately through the actual | |
861 | * bitmap and the last-committed copy until we find a bit free in | |
862 | * both | |
863 | */ | |
864 | here = bitmap_search_next_usable_block(here, bh, maxblocks); | |
865 | return here; | |
866 | } | |
867 | ||
868 | /** | |
869 | * claim_block() | |
870 | * @block: the free block (group relative) to allocate | |
871 | * @bh: the bufferhead containts the block group bitmap | |
872 | * | |
873 | * We think we can allocate this block in this bitmap. Try to set the bit. | |
874 | * If that succeeds then check that nobody has allocated and then freed the | |
875 | * block since we saw that is was not marked in b_committed_data. If it _was_ | |
876 | * allocated and freed then clear the bit in the bitmap again and return | |
877 | * zero (failure). | |
878 | */ | |
879 | static inline int | |
617ba13b | 880 | claim_block(spinlock_t *lock, ext4_grpblk_t block, struct buffer_head *bh) |
ac27a0ec DK |
881 | { |
882 | struct journal_head *jh = bh2jh(bh); | |
883 | int ret; | |
884 | ||
617ba13b | 885 | if (ext4_set_bit_atomic(lock, block, bh->b_data)) |
ac27a0ec DK |
886 | return 0; |
887 | jbd_lock_bh_state(bh); | |
617ba13b MC |
888 | if (jh->b_committed_data && ext4_test_bit(block,jh->b_committed_data)) { |
889 | ext4_clear_bit_atomic(lock, block, bh->b_data); | |
ac27a0ec DK |
890 | ret = 0; |
891 | } else { | |
892 | ret = 1; | |
893 | } | |
894 | jbd_unlock_bh_state(bh); | |
895 | return ret; | |
896 | } | |
897 | ||
898 | /** | |
617ba13b | 899 | * ext4_try_to_allocate() |
ac27a0ec DK |
900 | * @sb: superblock |
901 | * @handle: handle to this transaction | |
902 | * @group: given allocation block group | |
903 | * @bitmap_bh: bufferhead holds the block bitmap | |
904 | * @grp_goal: given target block within the group | |
905 | * @count: target number of blocks to allocate | |
906 | * @my_rsv: reservation window | |
907 | * | |
908 | * Attempt to allocate blocks within a give range. Set the range of allocation | |
909 | * first, then find the first free bit(s) from the bitmap (within the range), | |
910 | * and at last, allocate the blocks by claiming the found free bit as allocated. | |
911 | * | |
912 | * To set the range of this allocation: | |
913 | * if there is a reservation window, only try to allocate block(s) from the | |
914 | * file's own reservation window; | |
915 | * Otherwise, the allocation range starts from the give goal block, ends at | |
916 | * the block group's last block. | |
917 | * | |
918 | * If we failed to allocate the desired block then we may end up crossing to a | |
919 | * new bitmap. In that case we must release write access to the old one via | |
617ba13b | 920 | * ext4_journal_release_buffer(), else we'll run out of credits. |
ac27a0ec | 921 | */ |
617ba13b | 922 | static ext4_grpblk_t |
fd2d4291 AM |
923 | ext4_try_to_allocate(struct super_block *sb, handle_t *handle, |
924 | ext4_group_t group, struct buffer_head *bitmap_bh, | |
925 | ext4_grpblk_t grp_goal, unsigned long *count, | |
926 | struct ext4_reserve_window *my_rsv) | |
ac27a0ec | 927 | { |
617ba13b MC |
928 | ext4_fsblk_t group_first_block; |
929 | ext4_grpblk_t start, end; | |
ac27a0ec DK |
930 | unsigned long num = 0; |
931 | ||
932 | /* we do allocation within the reservation window if we have a window */ | |
933 | if (my_rsv) { | |
617ba13b | 934 | group_first_block = ext4_group_first_block_no(sb, group); |
ac27a0ec DK |
935 | if (my_rsv->_rsv_start >= group_first_block) |
936 | start = my_rsv->_rsv_start - group_first_block; | |
937 | else | |
938 | /* reservation window cross group boundary */ | |
939 | start = 0; | |
940 | end = my_rsv->_rsv_end - group_first_block + 1; | |
617ba13b | 941 | if (end > EXT4_BLOCKS_PER_GROUP(sb)) |
ac27a0ec | 942 | /* reservation window crosses group boundary */ |
617ba13b | 943 | end = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
944 | if ((start <= grp_goal) && (grp_goal < end)) |
945 | start = grp_goal; | |
946 | else | |
947 | grp_goal = -1; | |
948 | } else { | |
949 | if (grp_goal > 0) | |
950 | start = grp_goal; | |
951 | else | |
952 | start = 0; | |
617ba13b | 953 | end = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
954 | } |
955 | ||
617ba13b | 956 | BUG_ON(start > EXT4_BLOCKS_PER_GROUP(sb)); |
ac27a0ec DK |
957 | |
958 | repeat: | |
617ba13b | 959 | if (grp_goal < 0 || !ext4_test_allocatable(grp_goal, bitmap_bh)) { |
ac27a0ec DK |
960 | grp_goal = find_next_usable_block(start, bitmap_bh, end); |
961 | if (grp_goal < 0) | |
962 | goto fail_access; | |
963 | if (!my_rsv) { | |
964 | int i; | |
965 | ||
966 | for (i = 0; i < 7 && grp_goal > start && | |
617ba13b | 967 | ext4_test_allocatable(grp_goal - 1, |
ac27a0ec DK |
968 | bitmap_bh); |
969 | i++, grp_goal--) | |
970 | ; | |
971 | } | |
972 | } | |
973 | start = grp_goal; | |
974 | ||
617ba13b | 975 | if (!claim_block(sb_bgl_lock(EXT4_SB(sb), group), |
ac27a0ec DK |
976 | grp_goal, bitmap_bh)) { |
977 | /* | |
978 | * The block was allocated by another thread, or it was | |
979 | * allocated and then freed by another thread | |
980 | */ | |
981 | start++; | |
982 | grp_goal++; | |
983 | if (start >= end) | |
984 | goto fail_access; | |
985 | goto repeat; | |
986 | } | |
987 | num++; | |
988 | grp_goal++; | |
989 | while (num < *count && grp_goal < end | |
617ba13b MC |
990 | && ext4_test_allocatable(grp_goal, bitmap_bh) |
991 | && claim_block(sb_bgl_lock(EXT4_SB(sb), group), | |
ac27a0ec DK |
992 | grp_goal, bitmap_bh)) { |
993 | num++; | |
994 | grp_goal++; | |
995 | } | |
996 | *count = num; | |
997 | return grp_goal - num; | |
998 | fail_access: | |
999 | *count = num; | |
1000 | return -1; | |
1001 | } | |
1002 | ||
1003 | /** | |
1004 | * find_next_reservable_window(): | |
1005 | * find a reservable space within the given range. | |
1006 | * It does not allocate the reservation window for now: | |
1007 | * alloc_new_reservation() will do the work later. | |
1008 | * | |
1009 | * @search_head: the head of the searching list; | |
1010 | * This is not necessarily the list head of the whole filesystem | |
1011 | * | |
1012 | * We have both head and start_block to assist the search | |
1013 | * for the reservable space. The list starts from head, | |
1014 | * but we will shift to the place where start_block is, | |
1015 | * then start from there, when looking for a reservable space. | |
1016 | * | |
1017 | * @size: the target new reservation window size | |
1018 | * | |
1019 | * @group_first_block: the first block we consider to start | |
1020 | * the real search from | |
1021 | * | |
1022 | * @last_block: | |
1023 | * the maximum block number that our goal reservable space | |
1024 | * could start from. This is normally the last block in this | |
1025 | * group. The search will end when we found the start of next | |
1026 | * possible reservable space is out of this boundary. | |
1027 | * This could handle the cross boundary reservation window | |
1028 | * request. | |
1029 | * | |
1030 | * basically we search from the given range, rather than the whole | |
1031 | * reservation double linked list, (start_block, last_block) | |
1032 | * to find a free region that is of my size and has not | |
1033 | * been reserved. | |
1034 | * | |
1035 | */ | |
1036 | static int find_next_reservable_window( | |
617ba13b MC |
1037 | struct ext4_reserve_window_node *search_head, |
1038 | struct ext4_reserve_window_node *my_rsv, | |
ac27a0ec | 1039 | struct super_block * sb, |
617ba13b MC |
1040 | ext4_fsblk_t start_block, |
1041 | ext4_fsblk_t last_block) | |
ac27a0ec DK |
1042 | { |
1043 | struct rb_node *next; | |
617ba13b MC |
1044 | struct ext4_reserve_window_node *rsv, *prev; |
1045 | ext4_fsblk_t cur; | |
ac27a0ec DK |
1046 | int size = my_rsv->rsv_goal_size; |
1047 | ||
1048 | /* TODO: make the start of the reservation window byte-aligned */ | |
1049 | /* cur = *start_block & ~7;*/ | |
1050 | cur = start_block; | |
1051 | rsv = search_head; | |
1052 | if (!rsv) | |
1053 | return -1; | |
1054 | ||
1055 | while (1) { | |
1056 | if (cur <= rsv->rsv_end) | |
1057 | cur = rsv->rsv_end + 1; | |
1058 | ||
1059 | /* TODO? | |
1060 | * in the case we could not find a reservable space | |
1061 | * that is what is expected, during the re-search, we could | |
1062 | * remember what's the largest reservable space we could have | |
1063 | * and return that one. | |
1064 | * | |
1065 | * For now it will fail if we could not find the reservable | |
1066 | * space with expected-size (or more)... | |
1067 | */ | |
1068 | if (cur > last_block) | |
1069 | return -1; /* fail */ | |
1070 | ||
1071 | prev = rsv; | |
1072 | next = rb_next(&rsv->rsv_node); | |
b78a657f | 1073 | rsv = rb_entry(next,struct ext4_reserve_window_node,rsv_node); |
ac27a0ec DK |
1074 | |
1075 | /* | |
1076 | * Reached the last reservation, we can just append to the | |
1077 | * previous one. | |
1078 | */ | |
1079 | if (!next) | |
1080 | break; | |
1081 | ||
1082 | if (cur + size <= rsv->rsv_start) { | |
1083 | /* | |
1084 | * Found a reserveable space big enough. We could | |
1085 | * have a reservation across the group boundary here | |
1086 | */ | |
1087 | break; | |
1088 | } | |
1089 | } | |
1090 | /* | |
1091 | * we come here either : | |
1092 | * when we reach the end of the whole list, | |
1093 | * and there is empty reservable space after last entry in the list. | |
1094 | * append it to the end of the list. | |
1095 | * | |
1096 | * or we found one reservable space in the middle of the list, | |
1097 | * return the reservation window that we could append to. | |
1098 | * succeed. | |
1099 | */ | |
1100 | ||
1101 | if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window))) | |
1102 | rsv_window_remove(sb, my_rsv); | |
1103 | ||
1104 | /* | |
1105 | * Let's book the whole avaliable window for now. We will check the | |
1106 | * disk bitmap later and then, if there are free blocks then we adjust | |
1107 | * the window size if it's larger than requested. | |
1108 | * Otherwise, we will remove this node from the tree next time | |
1109 | * call find_next_reservable_window. | |
1110 | */ | |
1111 | my_rsv->rsv_start = cur; | |
1112 | my_rsv->rsv_end = cur + size - 1; | |
1113 | my_rsv->rsv_alloc_hit = 0; | |
1114 | ||
1115 | if (prev != my_rsv) | |
617ba13b | 1116 | ext4_rsv_window_add(sb, my_rsv); |
ac27a0ec DK |
1117 | |
1118 | return 0; | |
1119 | } | |
1120 | ||
1121 | /** | |
1122 | * alloc_new_reservation()--allocate a new reservation window | |
1123 | * | |
1124 | * To make a new reservation, we search part of the filesystem | |
1125 | * reservation list (the list that inside the group). We try to | |
1126 | * allocate a new reservation window near the allocation goal, | |
1127 | * or the beginning of the group, if there is no goal. | |
1128 | * | |
1129 | * We first find a reservable space after the goal, then from | |
1130 | * there, we check the bitmap for the first free block after | |
1131 | * it. If there is no free block until the end of group, then the | |
1132 | * whole group is full, we failed. Otherwise, check if the free | |
1133 | * block is inside the expected reservable space, if so, we | |
1134 | * succeed. | |
1135 | * If the first free block is outside the reservable space, then | |
1136 | * start from the first free block, we search for next available | |
1137 | * space, and go on. | |
1138 | * | |
1139 | * on succeed, a new reservation will be found and inserted into the list | |
1140 | * It contains at least one free block, and it does not overlap with other | |
1141 | * reservation windows. | |
1142 | * | |
1143 | * failed: we failed to find a reservation window in this group | |
1144 | * | |
1145 | * @rsv: the reservation | |
1146 | * | |
1147 | * @grp_goal: The goal (group-relative). It is where the search for a | |
1148 | * free reservable space should start from. | |
1149 | * if we have a grp_goal(grp_goal >0 ), then start from there, | |
1150 | * no grp_goal(grp_goal = -1), we start from the first block | |
1151 | * of the group. | |
1152 | * | |
1153 | * @sb: the super block | |
1154 | * @group: the group we are trying to allocate in | |
1155 | * @bitmap_bh: the block group block bitmap | |
1156 | * | |
1157 | */ | |
617ba13b MC |
1158 | static int alloc_new_reservation(struct ext4_reserve_window_node *my_rsv, |
1159 | ext4_grpblk_t grp_goal, struct super_block *sb, | |
fd2d4291 | 1160 | ext4_group_t group, struct buffer_head *bitmap_bh) |
ac27a0ec | 1161 | { |
617ba13b MC |
1162 | struct ext4_reserve_window_node *search_head; |
1163 | ext4_fsblk_t group_first_block, group_end_block, start_block; | |
1164 | ext4_grpblk_t first_free_block; | |
1165 | struct rb_root *fs_rsv_root = &EXT4_SB(sb)->s_rsv_window_root; | |
ac27a0ec DK |
1166 | unsigned long size; |
1167 | int ret; | |
617ba13b | 1168 | spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock; |
ac27a0ec | 1169 | |
617ba13b MC |
1170 | group_first_block = ext4_group_first_block_no(sb, group); |
1171 | group_end_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
1172 | |
1173 | if (grp_goal < 0) | |
1174 | start_block = group_first_block; | |
1175 | else | |
1176 | start_block = grp_goal + group_first_block; | |
1177 | ||
1178 | size = my_rsv->rsv_goal_size; | |
1179 | ||
1180 | if (!rsv_is_empty(&my_rsv->rsv_window)) { | |
1181 | /* | |
1182 | * if the old reservation is cross group boundary | |
1183 | * and if the goal is inside the old reservation window, | |
1184 | * we will come here when we just failed to allocate from | |
1185 | * the first part of the window. We still have another part | |
1186 | * that belongs to the next group. In this case, there is no | |
1187 | * point to discard our window and try to allocate a new one | |
1188 | * in this group(which will fail). we should | |
1189 | * keep the reservation window, just simply move on. | |
1190 | * | |
1191 | * Maybe we could shift the start block of the reservation | |
1192 | * window to the first block of next group. | |
1193 | */ | |
1194 | ||
1195 | if ((my_rsv->rsv_start <= group_end_block) && | |
1196 | (my_rsv->rsv_end > group_end_block) && | |
1197 | (start_block >= my_rsv->rsv_start)) | |
1198 | return -1; | |
1199 | ||
1200 | if ((my_rsv->rsv_alloc_hit > | |
1201 | (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) { | |
1202 | /* | |
1203 | * if the previously allocation hit ratio is | |
1204 | * greater than 1/2, then we double the size of | |
1205 | * the reservation window the next time, | |
1206 | * otherwise we keep the same size window | |
1207 | */ | |
1208 | size = size * 2; | |
617ba13b MC |
1209 | if (size > EXT4_MAX_RESERVE_BLOCKS) |
1210 | size = EXT4_MAX_RESERVE_BLOCKS; | |
ac27a0ec DK |
1211 | my_rsv->rsv_goal_size= size; |
1212 | } | |
1213 | } | |
1214 | ||
1215 | spin_lock(rsv_lock); | |
1216 | /* | |
1217 | * shift the search start to the window near the goal block | |
1218 | */ | |
1219 | search_head = search_reserve_window(fs_rsv_root, start_block); | |
1220 | ||
1221 | /* | |
1222 | * find_next_reservable_window() simply finds a reservable window | |
1223 | * inside the given range(start_block, group_end_block). | |
1224 | * | |
1225 | * To make sure the reservation window has a free bit inside it, we | |
1226 | * need to check the bitmap after we found a reservable window. | |
1227 | */ | |
1228 | retry: | |
1229 | ret = find_next_reservable_window(search_head, my_rsv, sb, | |
1230 | start_block, group_end_block); | |
1231 | ||
1232 | if (ret == -1) { | |
1233 | if (!rsv_is_empty(&my_rsv->rsv_window)) | |
1234 | rsv_window_remove(sb, my_rsv); | |
1235 | spin_unlock(rsv_lock); | |
1236 | return -1; | |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * On success, find_next_reservable_window() returns the | |
1241 | * reservation window where there is a reservable space after it. | |
1242 | * Before we reserve this reservable space, we need | |
1243 | * to make sure there is at least a free block inside this region. | |
1244 | * | |
1245 | * searching the first free bit on the block bitmap and copy of | |
1246 | * last committed bitmap alternatively, until we found a allocatable | |
1247 | * block. Search start from the start block of the reservable space | |
1248 | * we just found. | |
1249 | */ | |
1250 | spin_unlock(rsv_lock); | |
1251 | first_free_block = bitmap_search_next_usable_block( | |
1252 | my_rsv->rsv_start - group_first_block, | |
1253 | bitmap_bh, group_end_block - group_first_block + 1); | |
1254 | ||
1255 | if (first_free_block < 0) { | |
1256 | /* | |
1257 | * no free block left on the bitmap, no point | |
1258 | * to reserve the space. return failed. | |
1259 | */ | |
1260 | spin_lock(rsv_lock); | |
1261 | if (!rsv_is_empty(&my_rsv->rsv_window)) | |
1262 | rsv_window_remove(sb, my_rsv); | |
1263 | spin_unlock(rsv_lock); | |
1264 | return -1; /* failed */ | |
1265 | } | |
1266 | ||
1267 | start_block = first_free_block + group_first_block; | |
1268 | /* | |
1269 | * check if the first free block is within the | |
1270 | * free space we just reserved | |
1271 | */ | |
b2f2c76d | 1272 | if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end) |
ac27a0ec DK |
1273 | return 0; /* success */ |
1274 | /* | |
1275 | * if the first free bit we found is out of the reservable space | |
1276 | * continue search for next reservable space, | |
1277 | * start from where the free block is, | |
1278 | * we also shift the list head to where we stopped last time | |
1279 | */ | |
1280 | search_head = my_rsv; | |
1281 | spin_lock(rsv_lock); | |
1282 | goto retry; | |
1283 | } | |
1284 | ||
1285 | /** | |
1286 | * try_to_extend_reservation() | |
1287 | * @my_rsv: given reservation window | |
1288 | * @sb: super block | |
1289 | * @size: the delta to extend | |
1290 | * | |
1291 | * Attempt to expand the reservation window large enough to have | |
1292 | * required number of free blocks | |
1293 | * | |
617ba13b | 1294 | * Since ext4_try_to_allocate() will always allocate blocks within |
ac27a0ec DK |
1295 | * the reservation window range, if the window size is too small, |
1296 | * multiple blocks allocation has to stop at the end of the reservation | |
1297 | * window. To make this more efficient, given the total number of | |
1298 | * blocks needed and the current size of the window, we try to | |
1299 | * expand the reservation window size if necessary on a best-effort | |
617ba13b | 1300 | * basis before ext4_new_blocks() tries to allocate blocks, |
ac27a0ec | 1301 | */ |
617ba13b | 1302 | static void try_to_extend_reservation(struct ext4_reserve_window_node *my_rsv, |
ac27a0ec DK |
1303 | struct super_block *sb, int size) |
1304 | { | |
617ba13b | 1305 | struct ext4_reserve_window_node *next_rsv; |
ac27a0ec | 1306 | struct rb_node *next; |
617ba13b | 1307 | spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock; |
ac27a0ec DK |
1308 | |
1309 | if (!spin_trylock(rsv_lock)) | |
1310 | return; | |
1311 | ||
1312 | next = rb_next(&my_rsv->rsv_node); | |
1313 | ||
1314 | if (!next) | |
1315 | my_rsv->rsv_end += size; | |
1316 | else { | |
b78a657f | 1317 | next_rsv = rb_entry(next, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
1318 | |
1319 | if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size) | |
1320 | my_rsv->rsv_end += size; | |
1321 | else | |
1322 | my_rsv->rsv_end = next_rsv->rsv_start - 1; | |
1323 | } | |
1324 | spin_unlock(rsv_lock); | |
1325 | } | |
1326 | ||
1327 | /** | |
617ba13b | 1328 | * ext4_try_to_allocate_with_rsv() |
ac27a0ec DK |
1329 | * @sb: superblock |
1330 | * @handle: handle to this transaction | |
1331 | * @group: given allocation block group | |
1332 | * @bitmap_bh: bufferhead holds the block bitmap | |
1333 | * @grp_goal: given target block within the group | |
1334 | * @count: target number of blocks to allocate | |
1335 | * @my_rsv: reservation window | |
1336 | * @errp: pointer to store the error code | |
1337 | * | |
1338 | * This is the main function used to allocate a new block and its reservation | |
1339 | * window. | |
1340 | * | |
1341 | * Each time when a new block allocation is need, first try to allocate from | |
1342 | * its own reservation. If it does not have a reservation window, instead of | |
1343 | * looking for a free bit on bitmap first, then look up the reservation list to | |
1344 | * see if it is inside somebody else's reservation window, we try to allocate a | |
1345 | * reservation window for it starting from the goal first. Then do the block | |
1346 | * allocation within the reservation window. | |
1347 | * | |
1348 | * This will avoid keeping on searching the reservation list again and | |
1349 | * again when somebody is looking for a free block (without | |
1350 | * reservation), and there are lots of free blocks, but they are all | |
1351 | * being reserved. | |
1352 | * | |
1353 | * We use a red-black tree for the per-filesystem reservation list. | |
1354 | * | |
1355 | */ | |
617ba13b MC |
1356 | static ext4_grpblk_t |
1357 | ext4_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle, | |
fd2d4291 | 1358 | ext4_group_t group, struct buffer_head *bitmap_bh, |
617ba13b MC |
1359 | ext4_grpblk_t grp_goal, |
1360 | struct ext4_reserve_window_node * my_rsv, | |
ac27a0ec DK |
1361 | unsigned long *count, int *errp) |
1362 | { | |
617ba13b MC |
1363 | ext4_fsblk_t group_first_block, group_last_block; |
1364 | ext4_grpblk_t ret = 0; | |
ac27a0ec DK |
1365 | int fatal; |
1366 | unsigned long num = *count; | |
1367 | ||
1368 | *errp = 0; | |
1369 | ||
1370 | /* | |
1371 | * Make sure we use undo access for the bitmap, because it is critical | |
1372 | * that we do the frozen_data COW on bitmap buffers in all cases even | |
1373 | * if the buffer is in BJ_Forget state in the committing transaction. | |
1374 | */ | |
1375 | BUFFER_TRACE(bitmap_bh, "get undo access for new block"); | |
617ba13b | 1376 | fatal = ext4_journal_get_undo_access(handle, bitmap_bh); |
ac27a0ec DK |
1377 | if (fatal) { |
1378 | *errp = fatal; | |
1379 | return -1; | |
1380 | } | |
1381 | ||
1382 | /* | |
1383 | * we don't deal with reservation when | |
1384 | * filesystem is mounted without reservation | |
1385 | * or the file is not a regular file | |
1386 | * or last attempt to allocate a block with reservation turned on failed | |
1387 | */ | |
1388 | if (my_rsv == NULL ) { | |
617ba13b | 1389 | ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh, |
ac27a0ec DK |
1390 | grp_goal, count, NULL); |
1391 | goto out; | |
1392 | } | |
1393 | /* | |
1394 | * grp_goal is a group relative block number (if there is a goal) | |
e7dc95db | 1395 | * 0 <= grp_goal < EXT4_BLOCKS_PER_GROUP(sb) |
ac27a0ec DK |
1396 | * first block is a filesystem wide block number |
1397 | * first block is the block number of the first block in this group | |
1398 | */ | |
617ba13b MC |
1399 | group_first_block = ext4_group_first_block_no(sb, group); |
1400 | group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
1401 | |
1402 | /* | |
1403 | * Basically we will allocate a new block from inode's reservation | |
1404 | * window. | |
1405 | * | |
1406 | * We need to allocate a new reservation window, if: | |
1407 | * a) inode does not have a reservation window; or | |
1408 | * b) last attempt to allocate a block from existing reservation | |
1409 | * failed; or | |
1410 | * c) we come here with a goal and with a reservation window | |
1411 | * | |
1412 | * We do not need to allocate a new reservation window if we come here | |
1413 | * at the beginning with a goal and the goal is inside the window, or | |
1414 | * we don't have a goal but already have a reservation window. | |
1415 | * then we could go to allocate from the reservation window directly. | |
1416 | */ | |
1417 | while (1) { | |
1418 | if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) || | |
1419 | !goal_in_my_reservation(&my_rsv->rsv_window, | |
1420 | grp_goal, group, sb)) { | |
1421 | if (my_rsv->rsv_goal_size < *count) | |
1422 | my_rsv->rsv_goal_size = *count; | |
1423 | ret = alloc_new_reservation(my_rsv, grp_goal, sb, | |
1424 | group, bitmap_bh); | |
1425 | if (ret < 0) | |
1426 | break; /* failed */ | |
1427 | ||
1428 | if (!goal_in_my_reservation(&my_rsv->rsv_window, | |
1429 | grp_goal, group, sb)) | |
1430 | grp_goal = -1; | |
e7dc95db | 1431 | } else if (grp_goal >= 0) { |
1df1e63b MC |
1432 | int curr = my_rsv->rsv_end - |
1433 | (grp_goal + group_first_block) + 1; | |
1434 | ||
1435 | if (curr < *count) | |
1436 | try_to_extend_reservation(my_rsv, sb, | |
1437 | *count - curr); | |
1438 | } | |
ac27a0ec DK |
1439 | |
1440 | if ((my_rsv->rsv_start > group_last_block) || | |
1441 | (my_rsv->rsv_end < group_first_block)) { | |
617ba13b | 1442 | rsv_window_dump(&EXT4_SB(sb)->s_rsv_window_root, 1); |
ac27a0ec DK |
1443 | BUG(); |
1444 | } | |
617ba13b | 1445 | ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh, |
ac27a0ec DK |
1446 | grp_goal, &num, &my_rsv->rsv_window); |
1447 | if (ret >= 0) { | |
1448 | my_rsv->rsv_alloc_hit += num; | |
1449 | *count = num; | |
1450 | break; /* succeed */ | |
1451 | } | |
1452 | num = *count; | |
1453 | } | |
1454 | out: | |
1455 | if (ret >= 0) { | |
1456 | BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for " | |
1457 | "bitmap block"); | |
617ba13b | 1458 | fatal = ext4_journal_dirty_metadata(handle, bitmap_bh); |
ac27a0ec DK |
1459 | if (fatal) { |
1460 | *errp = fatal; | |
1461 | return -1; | |
1462 | } | |
1463 | return ret; | |
1464 | } | |
1465 | ||
1466 | BUFFER_TRACE(bitmap_bh, "journal_release_buffer"); | |
617ba13b | 1467 | ext4_journal_release_buffer(handle, bitmap_bh); |
ac27a0ec DK |
1468 | return ret; |
1469 | } | |
1470 | ||
1471 | /** | |
617ba13b | 1472 | * ext4_has_free_blocks() |
ac27a0ec DK |
1473 | * @sbi: in-core super block structure. |
1474 | * | |
1475 | * Check if filesystem has at least 1 free block available for allocation. | |
1476 | */ | |
617ba13b | 1477 | static int ext4_has_free_blocks(struct ext4_sb_info *sbi) |
ac27a0ec | 1478 | { |
617ba13b | 1479 | ext4_fsblk_t free_blocks, root_blocks; |
ac27a0ec DK |
1480 | |
1481 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
bd81d8ee | 1482 | root_blocks = ext4_r_blocks_count(sbi->s_es); |
ac27a0ec DK |
1483 | if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) && |
1484 | sbi->s_resuid != current->fsuid && | |
1485 | (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) { | |
1486 | return 0; | |
1487 | } | |
1488 | return 1; | |
1489 | } | |
1490 | ||
1491 | /** | |
617ba13b | 1492 | * ext4_should_retry_alloc() |
ac27a0ec DK |
1493 | * @sb: super block |
1494 | * @retries number of attemps has been made | |
1495 | * | |
617ba13b | 1496 | * ext4_should_retry_alloc() is called when ENOSPC is returned, and if |
ac27a0ec DK |
1497 | * it is profitable to retry the operation, this function will wait |
1498 | * for the current or commiting transaction to complete, and then | |
1499 | * return TRUE. | |
1500 | * | |
1501 | * if the total number of retries exceed three times, return FALSE. | |
1502 | */ | |
617ba13b | 1503 | int ext4_should_retry_alloc(struct super_block *sb, int *retries) |
ac27a0ec | 1504 | { |
617ba13b | 1505 | if (!ext4_has_free_blocks(EXT4_SB(sb)) || (*retries)++ > 3) |
ac27a0ec DK |
1506 | return 0; |
1507 | ||
1508 | jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id); | |
1509 | ||
dab291af | 1510 | return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal); |
ac27a0ec DK |
1511 | } |
1512 | ||
1513 | /** | |
617ba13b | 1514 | * ext4_new_blocks() -- core block(s) allocation function |
ac27a0ec DK |
1515 | * @handle: handle to this transaction |
1516 | * @inode: file inode | |
1517 | * @goal: given target block(filesystem wide) | |
1518 | * @count: target number of blocks to allocate | |
1519 | * @errp: error code | |
1520 | * | |
617ba13b | 1521 | * ext4_new_blocks uses a goal block to assist allocation. It tries to |
ac27a0ec DK |
1522 | * allocate block(s) from the block group contains the goal block first. If that |
1523 | * fails, it will try to allocate block(s) from other block groups without | |
1524 | * any specific goal block. | |
1525 | * | |
1526 | */ | |
617ba13b MC |
1527 | ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode, |
1528 | ext4_fsblk_t goal, unsigned long *count, int *errp) | |
ac27a0ec DK |
1529 | { |
1530 | struct buffer_head *bitmap_bh = NULL; | |
1531 | struct buffer_head *gdp_bh; | |
fd2d4291 AM |
1532 | ext4_group_t group_no; |
1533 | ext4_group_t goal_group; | |
617ba13b MC |
1534 | ext4_grpblk_t grp_target_blk; /* blockgroup relative goal block */ |
1535 | ext4_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/ | |
1536 | ext4_fsblk_t ret_block; /* filesyetem-wide allocated block */ | |
fd2d4291 | 1537 | ext4_group_t bgi; /* blockgroup iteration index */ |
ac27a0ec DK |
1538 | int fatal = 0, err; |
1539 | int performed_allocation = 0; | |
617ba13b | 1540 | ext4_grpblk_t free_blocks; /* number of free blocks in a group */ |
ac27a0ec | 1541 | struct super_block *sb; |
617ba13b MC |
1542 | struct ext4_group_desc *gdp; |
1543 | struct ext4_super_block *es; | |
1544 | struct ext4_sb_info *sbi; | |
1545 | struct ext4_reserve_window_node *my_rsv = NULL; | |
1546 | struct ext4_block_alloc_info *block_i; | |
ac27a0ec | 1547 | unsigned short windowsz = 0; |
fd2d4291 | 1548 | ext4_group_t ngroups; |
ac27a0ec DK |
1549 | unsigned long num = *count; |
1550 | ||
1551 | *errp = -ENOSPC; | |
1552 | sb = inode->i_sb; | |
1553 | if (!sb) { | |
617ba13b | 1554 | printk("ext4_new_block: nonexistent device"); |
ac27a0ec DK |
1555 | return 0; |
1556 | } | |
1557 | ||
1558 | /* | |
1559 | * Check quota for allocation of this block. | |
1560 | */ | |
1561 | if (DQUOT_ALLOC_BLOCK(inode, num)) { | |
1562 | *errp = -EDQUOT; | |
1563 | return 0; | |
1564 | } | |
1565 | ||
617ba13b MC |
1566 | sbi = EXT4_SB(sb); |
1567 | es = EXT4_SB(sb)->s_es; | |
1568 | ext4_debug("goal=%lu.\n", goal); | |
ac27a0ec DK |
1569 | /* |
1570 | * Allocate a block from reservation only when | |
1571 | * filesystem is mounted with reservation(default,-o reservation), and | |
1572 | * it's a regular file, and | |
1573 | * the desired window size is greater than 0 (One could use ioctl | |
617ba13b | 1574 | * command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off |
ac27a0ec DK |
1575 | * reservation on that particular file) |
1576 | */ | |
617ba13b | 1577 | block_i = EXT4_I(inode)->i_block_alloc_info; |
ac27a0ec DK |
1578 | if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0)) |
1579 | my_rsv = &block_i->rsv_window_node; | |
1580 | ||
617ba13b | 1581 | if (!ext4_has_free_blocks(sbi)) { |
ac27a0ec DK |
1582 | *errp = -ENOSPC; |
1583 | goto out; | |
1584 | } | |
1585 | ||
1586 | /* | |
1587 | * First, test whether the goal block is free. | |
1588 | */ | |
1589 | if (goal < le32_to_cpu(es->s_first_data_block) || | |
bd81d8ee | 1590 | goal >= ext4_blocks_count(es)) |
ac27a0ec | 1591 | goal = le32_to_cpu(es->s_first_data_block); |
3a5b2ecd | 1592 | ext4_get_group_no_and_offset(sb, goal, &group_no, &grp_target_blk); |
ac27a0ec DK |
1593 | goal_group = group_no; |
1594 | retry_alloc: | |
617ba13b | 1595 | gdp = ext4_get_group_desc(sb, group_no, &gdp_bh); |
ac27a0ec DK |
1596 | if (!gdp) |
1597 | goto io_error; | |
1598 | ||
1599 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); | |
1600 | /* | |
1601 | * if there is not enough free blocks to make a new resevation | |
1602 | * turn off reservation for this allocation | |
1603 | */ | |
1604 | if (my_rsv && (free_blocks < windowsz) | |
1605 | && (rsv_is_empty(&my_rsv->rsv_window))) | |
1606 | my_rsv = NULL; | |
1607 | ||
1608 | if (free_blocks > 0) { | |
ac27a0ec DK |
1609 | bitmap_bh = read_block_bitmap(sb, group_no); |
1610 | if (!bitmap_bh) | |
1611 | goto io_error; | |
617ba13b | 1612 | grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, |
ac27a0ec DK |
1613 | group_no, bitmap_bh, grp_target_blk, |
1614 | my_rsv, &num, &fatal); | |
1615 | if (fatal) | |
1616 | goto out; | |
1617 | if (grp_alloc_blk >= 0) | |
1618 | goto allocated; | |
1619 | } | |
1620 | ||
617ba13b | 1621 | ngroups = EXT4_SB(sb)->s_groups_count; |
ac27a0ec DK |
1622 | smp_rmb(); |
1623 | ||
1624 | /* | |
1625 | * Now search the rest of the groups. We assume that | |
1626 | * i and gdp correctly point to the last group visited. | |
1627 | */ | |
1628 | for (bgi = 0; bgi < ngroups; bgi++) { | |
1629 | group_no++; | |
1630 | if (group_no >= ngroups) | |
1631 | group_no = 0; | |
617ba13b | 1632 | gdp = ext4_get_group_desc(sb, group_no, &gdp_bh); |
341cee43 HD |
1633 | if (!gdp) |
1634 | goto io_error; | |
ac27a0ec DK |
1635 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); |
1636 | /* | |
1637 | * skip this group if the number of | |
1638 | * free blocks is less than half of the reservation | |
1639 | * window size. | |
1640 | */ | |
1641 | if (free_blocks <= (windowsz/2)) | |
1642 | continue; | |
1643 | ||
1644 | brelse(bitmap_bh); | |
1645 | bitmap_bh = read_block_bitmap(sb, group_no); | |
1646 | if (!bitmap_bh) | |
1647 | goto io_error; | |
1648 | /* | |
1649 | * try to allocate block(s) from this group, without a goal(-1). | |
1650 | */ | |
617ba13b | 1651 | grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, |
ac27a0ec DK |
1652 | group_no, bitmap_bh, -1, my_rsv, |
1653 | &num, &fatal); | |
1654 | if (fatal) | |
1655 | goto out; | |
1656 | if (grp_alloc_blk >= 0) | |
1657 | goto allocated; | |
1658 | } | |
1659 | /* | |
1660 | * We may end up a bogus ealier ENOSPC error due to | |
1661 | * filesystem is "full" of reservations, but | |
1662 | * there maybe indeed free blocks avaliable on disk | |
1663 | * In this case, we just forget about the reservations | |
1664 | * just do block allocation as without reservations. | |
1665 | */ | |
1666 | if (my_rsv) { | |
1667 | my_rsv = NULL; | |
cd16c8f7 | 1668 | windowsz = 0; |
ac27a0ec DK |
1669 | group_no = goal_group; |
1670 | goto retry_alloc; | |
1671 | } | |
1672 | /* No space left on the device */ | |
1673 | *errp = -ENOSPC; | |
1674 | goto out; | |
1675 | ||
1676 | allocated: | |
1677 | ||
617ba13b | 1678 | ext4_debug("using block group %d(%d)\n", |
ac27a0ec DK |
1679 | group_no, gdp->bg_free_blocks_count); |
1680 | ||
1681 | BUFFER_TRACE(gdp_bh, "get_write_access"); | |
617ba13b | 1682 | fatal = ext4_journal_get_write_access(handle, gdp_bh); |
ac27a0ec DK |
1683 | if (fatal) |
1684 | goto out; | |
1685 | ||
617ba13b | 1686 | ret_block = grp_alloc_blk + ext4_group_first_block_no(sb, group_no); |
ac27a0ec | 1687 | |
8fadc143 | 1688 | if (in_range(ext4_block_bitmap(sb, gdp), ret_block, num) || |
29bc5b4f | 1689 | in_range(ext4_inode_bitmap(sb, gdp), ret_block, num) || |
8fadc143 | 1690 | in_range(ret_block, ext4_inode_table(sb, gdp), |
bd81d8ee | 1691 | EXT4_SB(sb)->s_itb_per_group) || |
8fadc143 | 1692 | in_range(ret_block + num - 1, ext4_inode_table(sb, gdp), |
bd81d8ee | 1693 | EXT4_SB(sb)->s_itb_per_group)) |
617ba13b | 1694 | ext4_error(sb, "ext4_new_block", |
ac27a0ec | 1695 | "Allocating block in system zone - " |
2ae02107 | 1696 | "blocks from %llu, length %lu", |
ac27a0ec DK |
1697 | ret_block, num); |
1698 | ||
1699 | performed_allocation = 1; | |
1700 | ||
e23291b9 | 1701 | #ifdef CONFIG_JBD2_DEBUG |
ac27a0ec DK |
1702 | { |
1703 | struct buffer_head *debug_bh; | |
1704 | ||
1705 | /* Record bitmap buffer state in the newly allocated block */ | |
1706 | debug_bh = sb_find_get_block(sb, ret_block); | |
1707 | if (debug_bh) { | |
1708 | BUFFER_TRACE(debug_bh, "state when allocated"); | |
1709 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state"); | |
1710 | brelse(debug_bh); | |
1711 | } | |
1712 | } | |
1713 | jbd_lock_bh_state(bitmap_bh); | |
1714 | spin_lock(sb_bgl_lock(sbi, group_no)); | |
1715 | if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) { | |
1716 | int i; | |
1717 | ||
1718 | for (i = 0; i < num; i++) { | |
617ba13b | 1719 | if (ext4_test_bit(grp_alloc_blk+i, |
ac27a0ec DK |
1720 | bh2jh(bitmap_bh)->b_committed_data)) { |
1721 | printk("%s: block was unexpectedly set in " | |
1722 | "b_committed_data\n", __FUNCTION__); | |
1723 | } | |
1724 | } | |
1725 | } | |
617ba13b | 1726 | ext4_debug("found bit %d\n", grp_alloc_blk); |
ac27a0ec DK |
1727 | spin_unlock(sb_bgl_lock(sbi, group_no)); |
1728 | jbd_unlock_bh_state(bitmap_bh); | |
1729 | #endif | |
1730 | ||
bd81d8ee | 1731 | if (ret_block + num - 1 >= ext4_blocks_count(es)) { |
617ba13b | 1732 | ext4_error(sb, "ext4_new_block", |
2ae02107 | 1733 | "block(%llu) >= blocks count(%llu) - " |
3a5b2ecd | 1734 | "block_group = %lu, es == %p ", ret_block, |
bd81d8ee | 1735 | ext4_blocks_count(es), group_no, es); |
ac27a0ec DK |
1736 | goto out; |
1737 | } | |
1738 | ||
1739 | /* | |
1740 | * It is up to the caller to add the new buffer to a journal | |
1741 | * list of some description. We don't know in advance whether | |
1742 | * the caller wants to use it as metadata or data. | |
1743 | */ | |
ac27a0ec | 1744 | spin_lock(sb_bgl_lock(sbi, group_no)); |
717d50e4 AD |
1745 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) |
1746 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); | |
ac27a0ec DK |
1747 | gdp->bg_free_blocks_count = |
1748 | cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count)-num); | |
717d50e4 | 1749 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group_no, gdp); |
ac27a0ec | 1750 | spin_unlock(sb_bgl_lock(sbi, group_no)); |
3cb4f9fa | 1751 | percpu_counter_sub(&sbi->s_freeblocks_counter, num); |
ac27a0ec DK |
1752 | |
1753 | BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor"); | |
617ba13b | 1754 | err = ext4_journal_dirty_metadata(handle, gdp_bh); |
ac27a0ec DK |
1755 | if (!fatal) |
1756 | fatal = err; | |
1757 | ||
1758 | sb->s_dirt = 1; | |
1759 | if (fatal) | |
1760 | goto out; | |
1761 | ||
1762 | *errp = 0; | |
1763 | brelse(bitmap_bh); | |
1764 | DQUOT_FREE_BLOCK(inode, *count-num); | |
1765 | *count = num; | |
1766 | return ret_block; | |
1767 | ||
1768 | io_error: | |
1769 | *errp = -EIO; | |
1770 | out: | |
1771 | if (fatal) { | |
1772 | *errp = fatal; | |
617ba13b | 1773 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
1774 | } |
1775 | /* | |
1776 | * Undo the block allocation | |
1777 | */ | |
1778 | if (!performed_allocation) | |
1779 | DQUOT_FREE_BLOCK(inode, *count); | |
1780 | brelse(bitmap_bh); | |
1781 | return 0; | |
1782 | } | |
1783 | ||
617ba13b MC |
1784 | ext4_fsblk_t ext4_new_block(handle_t *handle, struct inode *inode, |
1785 | ext4_fsblk_t goal, int *errp) | |
ac27a0ec DK |
1786 | { |
1787 | unsigned long count = 1; | |
1788 | ||
617ba13b | 1789 | return ext4_new_blocks(handle, inode, goal, &count, errp); |
ac27a0ec DK |
1790 | } |
1791 | ||
1792 | /** | |
617ba13b | 1793 | * ext4_count_free_blocks() -- count filesystem free blocks |
ac27a0ec DK |
1794 | * @sb: superblock |
1795 | * | |
1796 | * Adds up the number of free blocks from each block group. | |
1797 | */ | |
617ba13b | 1798 | ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb) |
ac27a0ec | 1799 | { |
617ba13b MC |
1800 | ext4_fsblk_t desc_count; |
1801 | struct ext4_group_desc *gdp; | |
fd2d4291 AM |
1802 | ext4_group_t i; |
1803 | ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; | |
617ba13b MC |
1804 | #ifdef EXT4FS_DEBUG |
1805 | struct ext4_super_block *es; | |
1806 | ext4_fsblk_t bitmap_count; | |
ac27a0ec DK |
1807 | unsigned long x; |
1808 | struct buffer_head *bitmap_bh = NULL; | |
1809 | ||
617ba13b | 1810 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1811 | desc_count = 0; |
1812 | bitmap_count = 0; | |
1813 | gdp = NULL; | |
1814 | ||
1815 | smp_rmb(); | |
1816 | for (i = 0; i < ngroups; i++) { | |
617ba13b | 1817 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1818 | if (!gdp) |
1819 | continue; | |
1820 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); | |
1821 | brelse(bitmap_bh); | |
1822 | bitmap_bh = read_block_bitmap(sb, i); | |
1823 | if (bitmap_bh == NULL) | |
1824 | continue; | |
1825 | ||
617ba13b | 1826 | x = ext4_count_free(bitmap_bh, sb->s_blocksize); |
fd2d4291 | 1827 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
ac27a0ec DK |
1828 | i, le16_to_cpu(gdp->bg_free_blocks_count), x); |
1829 | bitmap_count += x; | |
1830 | } | |
1831 | brelse(bitmap_bh); | |
2ae02107 MC |
1832 | printk("ext4_count_free_blocks: stored = %llu" |
1833 | ", computed = %llu, %llu\n", | |
bd81d8ee | 1834 | EXT4_FREE_BLOCKS_COUNT(es), |
ac27a0ec DK |
1835 | desc_count, bitmap_count); |
1836 | return bitmap_count; | |
1837 | #else | |
1838 | desc_count = 0; | |
1839 | smp_rmb(); | |
1840 | for (i = 0; i < ngroups; i++) { | |
617ba13b | 1841 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1842 | if (!gdp) |
1843 | continue; | |
1844 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); | |
1845 | } | |
1846 | ||
1847 | return desc_count; | |
1848 | #endif | |
1849 | } | |
1850 | ||
fd2d4291 | 1851 | static inline int test_root(ext4_group_t a, int b) |
ac27a0ec DK |
1852 | { |
1853 | int num = b; | |
1854 | ||
1855 | while (a > num) | |
1856 | num *= b; | |
1857 | return num == a; | |
1858 | } | |
1859 | ||
fd2d4291 | 1860 | static int ext4_group_sparse(ext4_group_t group) |
ac27a0ec DK |
1861 | { |
1862 | if (group <= 1) | |
1863 | return 1; | |
1864 | if (!(group & 1)) | |
1865 | return 0; | |
1866 | return (test_root(group, 7) || test_root(group, 5) || | |
1867 | test_root(group, 3)); | |
1868 | } | |
1869 | ||
1870 | /** | |
617ba13b | 1871 | * ext4_bg_has_super - number of blocks used by the superblock in group |
ac27a0ec DK |
1872 | * @sb: superblock for filesystem |
1873 | * @group: group number to check | |
1874 | * | |
1875 | * Return the number of blocks used by the superblock (primary or backup) | |
1876 | * in this group. Currently this will be only 0 or 1. | |
1877 | */ | |
fd2d4291 | 1878 | int ext4_bg_has_super(struct super_block *sb, ext4_group_t group) |
ac27a0ec | 1879 | { |
617ba13b MC |
1880 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
1881 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) && | |
1882 | !ext4_group_sparse(group)) | |
ac27a0ec DK |
1883 | return 0; |
1884 | return 1; | |
1885 | } | |
1886 | ||
fd2d4291 AM |
1887 | static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, |
1888 | ext4_group_t group) | |
ac27a0ec | 1889 | { |
617ba13b | 1890 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); |
fd2d4291 AM |
1891 | ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb); |
1892 | ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1; | |
ac27a0ec DK |
1893 | |
1894 | if (group == first || group == first + 1 || group == last) | |
1895 | return 1; | |
1896 | return 0; | |
1897 | } | |
1898 | ||
fd2d4291 AM |
1899 | static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, |
1900 | ext4_group_t group) | |
ac27a0ec | 1901 | { |
617ba13b MC |
1902 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
1903 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) && | |
1904 | !ext4_group_sparse(group)) | |
ac27a0ec | 1905 | return 0; |
617ba13b | 1906 | return EXT4_SB(sb)->s_gdb_count; |
ac27a0ec DK |
1907 | } |
1908 | ||
1909 | /** | |
617ba13b | 1910 | * ext4_bg_num_gdb - number of blocks used by the group table in group |
ac27a0ec DK |
1911 | * @sb: superblock for filesystem |
1912 | * @group: group number to check | |
1913 | * | |
1914 | * Return the number of blocks used by the group descriptor table | |
1915 | * (primary or backup) in this group. In the future there may be a | |
1916 | * different number of descriptor blocks in each group. | |
1917 | */ | |
fd2d4291 | 1918 | unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group) |
ac27a0ec DK |
1919 | { |
1920 | unsigned long first_meta_bg = | |
617ba13b MC |
1921 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); |
1922 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); | |
ac27a0ec | 1923 | |
617ba13b | 1924 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) || |
ac27a0ec | 1925 | metagroup < first_meta_bg) |
617ba13b | 1926 | return ext4_bg_num_gdb_nometa(sb,group); |
ac27a0ec | 1927 | |
617ba13b | 1928 | return ext4_bg_num_gdb_meta(sb,group); |
ac27a0ec DK |
1929 | |
1930 | } |