]>
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 | ||
23 | /* | |
24 | * balloc.c contains the blocks allocation and deallocation routines | |
25 | */ | |
26 | ||
72b64b59 AM |
27 | /* |
28 | * Calculate the block group number and offset, given a block number | |
29 | */ | |
30 | void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr, | |
31 | unsigned long *blockgrpp, ext4_grpblk_t *offsetp) | |
32 | { | |
8c55e204 | 33 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
72b64b59 AM |
34 | ext4_grpblk_t offset; |
35 | ||
8c55e204 | 36 | blocknr = blocknr - le32_to_cpu(es->s_first_data_block); |
f4e5bc24 | 37 | offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)); |
72b64b59 AM |
38 | if (offsetp) |
39 | *offsetp = offset; | |
40 | if (blockgrpp) | |
8c55e204 | 41 | *blockgrpp = blocknr; |
72b64b59 AM |
42 | |
43 | } | |
44 | ||
ac27a0ec DK |
45 | /* |
46 | * The free blocks are managed by bitmaps. A file system contains several | |
47 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
48 | * block for inodes, N blocks for the inode table and data blocks. | |
49 | * | |
50 | * The file system contains group descriptors which are located after the | |
51 | * super block. Each descriptor contains the number of the bitmap block and | |
52 | * the free blocks count in the block. The descriptors are loaded in memory | |
e627432c | 53 | * when a file system is mounted (see ext4_fill_super). |
ac27a0ec DK |
54 | */ |
55 | ||
56 | ||
57 | #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1) | |
58 | ||
59 | /** | |
617ba13b | 60 | * ext4_get_group_desc() -- load group descriptor from disk |
ac27a0ec DK |
61 | * @sb: super block |
62 | * @block_group: given block group | |
63 | * @bh: pointer to the buffer head to store the block | |
64 | * group descriptor | |
65 | */ | |
617ba13b | 66 | struct ext4_group_desc * ext4_get_group_desc(struct super_block * sb, |
ac27a0ec DK |
67 | unsigned int block_group, |
68 | struct buffer_head ** bh) | |
69 | { | |
70 | unsigned long group_desc; | |
71 | unsigned long offset; | |
617ba13b MC |
72 | struct ext4_group_desc * desc; |
73 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
ac27a0ec DK |
74 | |
75 | if (block_group >= sbi->s_groups_count) { | |
617ba13b | 76 | ext4_error (sb, "ext4_get_group_desc", |
ac27a0ec DK |
77 | "block_group >= groups_count - " |
78 | "block_group = %d, groups_count = %lu", | |
79 | block_group, sbi->s_groups_count); | |
80 | ||
81 | return NULL; | |
82 | } | |
83 | smp_rmb(); | |
84 | ||
617ba13b MC |
85 | group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); |
86 | offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); | |
ac27a0ec | 87 | if (!sbi->s_group_desc[group_desc]) { |
617ba13b | 88 | ext4_error (sb, "ext4_get_group_desc", |
ac27a0ec DK |
89 | "Group descriptor not loaded - " |
90 | "block_group = %d, group_desc = %lu, desc = %lu", | |
91 | block_group, group_desc, offset); | |
92 | return NULL; | |
93 | } | |
94 | ||
0d1ee42f AR |
95 | desc = (struct ext4_group_desc *)( |
96 | (__u8 *)sbi->s_group_desc[group_desc]->b_data + | |
97 | offset * EXT4_DESC_SIZE(sb)); | |
ac27a0ec DK |
98 | if (bh) |
99 | *bh = sbi->s_group_desc[group_desc]; | |
0d1ee42f | 100 | return desc; |
ac27a0ec DK |
101 | } |
102 | ||
7c9e69fa AK |
103 | static inline int |
104 | block_in_use(ext4_fsblk_t block, struct super_block *sb, unsigned char *map) | |
105 | { | |
106 | ext4_grpblk_t offset; | |
107 | ||
108 | ext4_get_group_no_and_offset(sb, block, NULL, &offset); | |
109 | return ext4_test_bit (offset, map); | |
110 | } | |
111 | ||
ac27a0ec DK |
112 | /** |
113 | * read_block_bitmap() | |
114 | * @sb: super block | |
115 | * @block_group: given block group | |
116 | * | |
117 | * Read the bitmap for a given block_group, reading into the specified | |
118 | * slot in the superblock's bitmap cache. | |
119 | * | |
120 | * Return buffer_head on success or NULL in case of failure. | |
121 | */ | |
122 | static struct buffer_head * | |
123 | read_block_bitmap(struct super_block *sb, unsigned int block_group) | |
124 | { | |
7c9e69fa | 125 | int i; |
617ba13b | 126 | struct ext4_group_desc * desc; |
ac27a0ec | 127 | struct buffer_head * bh = NULL; |
7c9e69fa | 128 | ext4_fsblk_t bitmap_blk; |
ac27a0ec | 129 | |
617ba13b | 130 | desc = ext4_get_group_desc (sb, block_group, NULL); |
ac27a0ec | 131 | if (!desc) |
7c9e69fa AK |
132 | return NULL; |
133 | bitmap_blk = ext4_block_bitmap(sb, desc); | |
134 | bh = sb_bread(sb, bitmap_blk); | |
ac27a0ec | 135 | if (!bh) |
7c9e69fa | 136 | ext4_error (sb, __FUNCTION__, |
ac27a0ec | 137 | "Cannot read block bitmap - " |
2ae02107 | 138 | "block_group = %d, block_bitmap = %llu", |
7c9e69fa AK |
139 | block_group, bitmap_blk); |
140 | ||
141 | /* check whether block bitmap block number is set */ | |
142 | if (!block_in_use(bitmap_blk, sb, bh->b_data)) { | |
143 | /* bad block bitmap */ | |
144 | goto error_out; | |
145 | } | |
146 | ||
147 | /* check whether the inode bitmap block number is set */ | |
148 | bitmap_blk = ext4_inode_bitmap(sb, desc); | |
149 | if (!block_in_use(bitmap_blk, sb, bh->b_data)) { | |
150 | /* bad block bitmap */ | |
151 | goto error_out; | |
152 | } | |
153 | /* check whether the inode table block number is set */ | |
154 | bitmap_blk = ext4_inode_table(sb, desc); | |
155 | for (i = 0; i < EXT4_SB(sb)->s_itb_per_group; i++, bitmap_blk++) { | |
156 | if (!block_in_use(bitmap_blk, sb, bh->b_data)) { | |
157 | /* bad block bitmap */ | |
158 | goto error_out; | |
159 | } | |
160 | } | |
161 | ||
ac27a0ec | 162 | return bh; |
7c9e69fa AK |
163 | |
164 | error_out: | |
165 | brelse(bh); | |
166 | ext4_error(sb, __FUNCTION__, | |
167 | "Invalid block bitmap - " | |
168 | "block_group = %d, block = %llu", | |
169 | block_group, bitmap_blk); | |
170 | return NULL; | |
171 | ||
ac27a0ec DK |
172 | } |
173 | /* | |
174 | * The reservation window structure operations | |
175 | * -------------------------------------------- | |
176 | * Operations include: | |
177 | * dump, find, add, remove, is_empty, find_next_reservable_window, etc. | |
178 | * | |
179 | * We use a red-black tree to represent per-filesystem reservation | |
180 | * windows. | |
181 | * | |
182 | */ | |
183 | ||
184 | /** | |
185 | * __rsv_window_dump() -- Dump the filesystem block allocation reservation map | |
186 | * @rb_root: root of per-filesystem reservation rb tree | |
187 | * @verbose: verbose mode | |
188 | * @fn: function which wishes to dump the reservation map | |
189 | * | |
190 | * If verbose is turned on, it will print the whole block reservation | |
191 | * windows(start, end). Otherwise, it will only print out the "bad" windows, | |
192 | * those windows that overlap with their immediate neighbors. | |
193 | */ | |
194 | #if 1 | |
195 | static void __rsv_window_dump(struct rb_root *root, int verbose, | |
196 | const char *fn) | |
197 | { | |
198 | struct rb_node *n; | |
617ba13b | 199 | struct ext4_reserve_window_node *rsv, *prev; |
ac27a0ec DK |
200 | int bad; |
201 | ||
202 | restart: | |
203 | n = rb_first(root); | |
204 | bad = 0; | |
205 | prev = NULL; | |
206 | ||
207 | printk("Block Allocation Reservation Windows Map (%s):\n", fn); | |
208 | while (n) { | |
b78a657f | 209 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
210 | if (verbose) |
211 | printk("reservation window 0x%p " | |
2ae02107 | 212 | "start: %llu, end: %llu\n", |
ac27a0ec DK |
213 | rsv, rsv->rsv_start, rsv->rsv_end); |
214 | if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) { | |
215 | printk("Bad reservation %p (start >= end)\n", | |
216 | rsv); | |
217 | bad = 1; | |
218 | } | |
219 | if (prev && prev->rsv_end >= rsv->rsv_start) { | |
220 | printk("Bad reservation %p (prev->end >= start)\n", | |
221 | rsv); | |
222 | bad = 1; | |
223 | } | |
224 | if (bad) { | |
225 | if (!verbose) { | |
226 | printk("Restarting reservation walk in verbose mode\n"); | |
227 | verbose = 1; | |
228 | goto restart; | |
229 | } | |
230 | } | |
231 | n = rb_next(n); | |
232 | prev = rsv; | |
233 | } | |
234 | printk("Window map complete.\n"); | |
235 | if (bad) | |
236 | BUG(); | |
237 | } | |
238 | #define rsv_window_dump(root, verbose) \ | |
239 | __rsv_window_dump((root), (verbose), __FUNCTION__) | |
240 | #else | |
241 | #define rsv_window_dump(root, verbose) do {} while (0) | |
242 | #endif | |
243 | ||
244 | /** | |
245 | * goal_in_my_reservation() | |
246 | * @rsv: inode's reservation window | |
247 | * @grp_goal: given goal block relative to the allocation block group | |
248 | * @group: the current allocation block group | |
249 | * @sb: filesystem super block | |
250 | * | |
251 | * Test if the given goal block (group relative) is within the file's | |
252 | * own block reservation window range. | |
253 | * | |
254 | * If the reservation window is outside the goal allocation group, return 0; | |
255 | * grp_goal (given goal block) could be -1, which means no specific | |
256 | * goal block. In this case, always return 1. | |
257 | * If the goal block is within the reservation window, return 1; | |
258 | * otherwise, return 0; | |
259 | */ | |
260 | static int | |
617ba13b | 261 | goal_in_my_reservation(struct ext4_reserve_window *rsv, ext4_grpblk_t grp_goal, |
ac27a0ec DK |
262 | unsigned int group, struct super_block * sb) |
263 | { | |
617ba13b | 264 | ext4_fsblk_t group_first_block, group_last_block; |
ac27a0ec | 265 | |
617ba13b MC |
266 | group_first_block = ext4_group_first_block_no(sb, group); |
267 | group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
268 | |
269 | if ((rsv->_rsv_start > group_last_block) || | |
270 | (rsv->_rsv_end < group_first_block)) | |
271 | return 0; | |
272 | if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start) | |
273 | || (grp_goal + group_first_block > rsv->_rsv_end))) | |
274 | return 0; | |
275 | return 1; | |
276 | } | |
277 | ||
278 | /** | |
279 | * search_reserve_window() | |
280 | * @rb_root: root of reservation tree | |
281 | * @goal: target allocation block | |
282 | * | |
283 | * Find the reserved window which includes the goal, or the previous one | |
284 | * if the goal is not in any window. | |
285 | * Returns NULL if there are no windows or if all windows start after the goal. | |
286 | */ | |
617ba13b MC |
287 | static struct ext4_reserve_window_node * |
288 | search_reserve_window(struct rb_root *root, ext4_fsblk_t goal) | |
ac27a0ec DK |
289 | { |
290 | struct rb_node *n = root->rb_node; | |
617ba13b | 291 | struct ext4_reserve_window_node *rsv; |
ac27a0ec DK |
292 | |
293 | if (!n) | |
294 | return NULL; | |
295 | ||
296 | do { | |
617ba13b | 297 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
298 | |
299 | if (goal < rsv->rsv_start) | |
300 | n = n->rb_left; | |
301 | else if (goal > rsv->rsv_end) | |
302 | n = n->rb_right; | |
303 | else | |
304 | return rsv; | |
305 | } while (n); | |
306 | /* | |
307 | * We've fallen off the end of the tree: the goal wasn't inside | |
308 | * any particular node. OK, the previous node must be to one | |
309 | * side of the interval containing the goal. If it's the RHS, | |
310 | * we need to back up one. | |
311 | */ | |
312 | if (rsv->rsv_start > goal) { | |
313 | n = rb_prev(&rsv->rsv_node); | |
617ba13b | 314 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
315 | } |
316 | return rsv; | |
317 | } | |
318 | ||
319 | /** | |
617ba13b | 320 | * ext4_rsv_window_add() -- Insert a window to the block reservation rb tree. |
ac27a0ec DK |
321 | * @sb: super block |
322 | * @rsv: reservation window to add | |
323 | * | |
324 | * Must be called with rsv_lock hold. | |
325 | */ | |
617ba13b MC |
326 | void ext4_rsv_window_add(struct super_block *sb, |
327 | struct ext4_reserve_window_node *rsv) | |
ac27a0ec | 328 | { |
617ba13b | 329 | struct rb_root *root = &EXT4_SB(sb)->s_rsv_window_root; |
ac27a0ec | 330 | struct rb_node *node = &rsv->rsv_node; |
617ba13b | 331 | ext4_fsblk_t start = rsv->rsv_start; |
ac27a0ec DK |
332 | |
333 | struct rb_node ** p = &root->rb_node; | |
334 | struct rb_node * parent = NULL; | |
617ba13b | 335 | struct ext4_reserve_window_node *this; |
ac27a0ec DK |
336 | |
337 | while (*p) | |
338 | { | |
339 | parent = *p; | |
617ba13b | 340 | this = rb_entry(parent, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
341 | |
342 | if (start < this->rsv_start) | |
343 | p = &(*p)->rb_left; | |
344 | else if (start > this->rsv_end) | |
345 | p = &(*p)->rb_right; | |
346 | else { | |
347 | rsv_window_dump(root, 1); | |
348 | BUG(); | |
349 | } | |
350 | } | |
351 | ||
352 | rb_link_node(node, parent, p); | |
353 | rb_insert_color(node, root); | |
354 | } | |
355 | ||
356 | /** | |
617ba13b | 357 | * ext4_rsv_window_remove() -- unlink a window from the reservation rb tree |
ac27a0ec DK |
358 | * @sb: super block |
359 | * @rsv: reservation window to remove | |
360 | * | |
361 | * Mark the block reservation window as not allocated, and unlink it | |
362 | * from the filesystem reservation window rb tree. Must be called with | |
363 | * rsv_lock hold. | |
364 | */ | |
365 | static void rsv_window_remove(struct super_block *sb, | |
617ba13b | 366 | struct ext4_reserve_window_node *rsv) |
ac27a0ec | 367 | { |
617ba13b MC |
368 | rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
369 | rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; | |
ac27a0ec | 370 | rsv->rsv_alloc_hit = 0; |
617ba13b | 371 | rb_erase(&rsv->rsv_node, &EXT4_SB(sb)->s_rsv_window_root); |
ac27a0ec DK |
372 | } |
373 | ||
374 | /* | |
375 | * rsv_is_empty() -- Check if the reservation window is allocated. | |
376 | * @rsv: given reservation window to check | |
377 | * | |
617ba13b | 378 | * returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED. |
ac27a0ec | 379 | */ |
617ba13b | 380 | static inline int rsv_is_empty(struct ext4_reserve_window *rsv) |
ac27a0ec DK |
381 | { |
382 | /* a valid reservation end block could not be 0 */ | |
617ba13b | 383 | return rsv->_rsv_end == EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
ac27a0ec DK |
384 | } |
385 | ||
386 | /** | |
617ba13b | 387 | * ext4_init_block_alloc_info() |
ac27a0ec DK |
388 | * @inode: file inode structure |
389 | * | |
390 | * Allocate and initialize the reservation window structure, and | |
617ba13b | 391 | * link the window to the ext4 inode structure at last |
ac27a0ec DK |
392 | * |
393 | * The reservation window structure is only dynamically allocated | |
617ba13b MC |
394 | * and linked to ext4 inode the first time the open file |
395 | * needs a new block. So, before every ext4_new_block(s) call, for | |
ac27a0ec DK |
396 | * regular files, we should check whether the reservation window |
397 | * structure exists or not. In the latter case, this function is called. | |
398 | * Fail to do so will result in block reservation being turned off for that | |
399 | * open file. | |
400 | * | |
617ba13b | 401 | * This function is called from ext4_get_blocks_handle(), also called |
ac27a0ec DK |
402 | * when setting the reservation window size through ioctl before the file |
403 | * is open for write (needs block allocation). | |
404 | * | |
405 | * Needs truncate_mutex protection prior to call this function. | |
406 | */ | |
617ba13b | 407 | void ext4_init_block_alloc_info(struct inode *inode) |
ac27a0ec | 408 | { |
617ba13b MC |
409 | struct ext4_inode_info *ei = EXT4_I(inode); |
410 | struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info; | |
ac27a0ec DK |
411 | struct super_block *sb = inode->i_sb; |
412 | ||
413 | block_i = kmalloc(sizeof(*block_i), GFP_NOFS); | |
414 | if (block_i) { | |
617ba13b | 415 | struct ext4_reserve_window_node *rsv = &block_i->rsv_window_node; |
ac27a0ec | 416 | |
617ba13b MC |
417 | rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
418 | rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; | |
ac27a0ec DK |
419 | |
420 | /* | |
421 | * if filesystem is mounted with NORESERVATION, the goal | |
422 | * reservation window size is set to zero to indicate | |
423 | * block reservation is off | |
424 | */ | |
425 | if (!test_opt(sb, RESERVATION)) | |
426 | rsv->rsv_goal_size = 0; | |
427 | else | |
617ba13b | 428 | rsv->rsv_goal_size = EXT4_DEFAULT_RESERVE_BLOCKS; |
ac27a0ec DK |
429 | rsv->rsv_alloc_hit = 0; |
430 | block_i->last_alloc_logical_block = 0; | |
431 | block_i->last_alloc_physical_block = 0; | |
432 | } | |
433 | ei->i_block_alloc_info = block_i; | |
434 | } | |
435 | ||
436 | /** | |
617ba13b | 437 | * ext4_discard_reservation() |
ac27a0ec DK |
438 | * @inode: inode |
439 | * | |
440 | * Discard(free) block reservation window on last file close, or truncate | |
441 | * or at last iput(). | |
442 | * | |
443 | * It is being called in three cases: | |
617ba13b MC |
444 | * ext4_release_file(): last writer close the file |
445 | * ext4_clear_inode(): last iput(), when nobody link to this file. | |
446 | * ext4_truncate(): when the block indirect map is about to change. | |
ac27a0ec DK |
447 | * |
448 | */ | |
617ba13b | 449 | void ext4_discard_reservation(struct inode *inode) |
ac27a0ec | 450 | { |
617ba13b MC |
451 | struct ext4_inode_info *ei = EXT4_I(inode); |
452 | struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info; | |
453 | struct ext4_reserve_window_node *rsv; | |
454 | spinlock_t *rsv_lock = &EXT4_SB(inode->i_sb)->s_rsv_window_lock; | |
ac27a0ec DK |
455 | |
456 | if (!block_i) | |
457 | return; | |
458 | ||
459 | rsv = &block_i->rsv_window_node; | |
460 | if (!rsv_is_empty(&rsv->rsv_window)) { | |
461 | spin_lock(rsv_lock); | |
462 | if (!rsv_is_empty(&rsv->rsv_window)) | |
463 | rsv_window_remove(inode->i_sb, rsv); | |
464 | spin_unlock(rsv_lock); | |
465 | } | |
466 | } | |
467 | ||
468 | /** | |
617ba13b | 469 | * ext4_free_blocks_sb() -- Free given blocks and update quota |
ac27a0ec DK |
470 | * @handle: handle to this transaction |
471 | * @sb: super block | |
472 | * @block: start physcial block to free | |
473 | * @count: number of blocks to free | |
474 | * @pdquot_freed_blocks: pointer to quota | |
475 | */ | |
617ba13b MC |
476 | void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb, |
477 | ext4_fsblk_t block, unsigned long count, | |
ac27a0ec DK |
478 | unsigned long *pdquot_freed_blocks) |
479 | { | |
480 | struct buffer_head *bitmap_bh = NULL; | |
481 | struct buffer_head *gd_bh; | |
482 | unsigned long block_group; | |
617ba13b | 483 | ext4_grpblk_t bit; |
ac27a0ec DK |
484 | unsigned long i; |
485 | unsigned long overflow; | |
617ba13b MC |
486 | struct ext4_group_desc * desc; |
487 | struct ext4_super_block * es; | |
488 | struct ext4_sb_info *sbi; | |
ac27a0ec | 489 | int err = 0, ret; |
617ba13b | 490 | ext4_grpblk_t group_freed; |
ac27a0ec DK |
491 | |
492 | *pdquot_freed_blocks = 0; | |
617ba13b | 493 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
494 | es = sbi->s_es; |
495 | if (block < le32_to_cpu(es->s_first_data_block) || | |
496 | block + count < block || | |
bd81d8ee | 497 | block + count > ext4_blocks_count(es)) { |
617ba13b | 498 | ext4_error (sb, "ext4_free_blocks", |
ac27a0ec | 499 | "Freeing blocks not in datazone - " |
2ae02107 | 500 | "block = %llu, count = %lu", block, count); |
ac27a0ec DK |
501 | goto error_return; |
502 | } | |
503 | ||
bd81d8ee | 504 | ext4_debug ("freeing block(s) %llu-%llu\n", block, block + count - 1); |
ac27a0ec DK |
505 | |
506 | do_more: | |
507 | overflow = 0; | |
3a5b2ecd | 508 | ext4_get_group_no_and_offset(sb, block, &block_group, &bit); |
ac27a0ec DK |
509 | /* |
510 | * Check to see if we are freeing blocks across a group | |
511 | * boundary. | |
512 | */ | |
617ba13b MC |
513 | if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) { |
514 | overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb); | |
ac27a0ec DK |
515 | count -= overflow; |
516 | } | |
517 | brelse(bitmap_bh); | |
518 | bitmap_bh = read_block_bitmap(sb, block_group); | |
519 | if (!bitmap_bh) | |
520 | goto error_return; | |
617ba13b | 521 | desc = ext4_get_group_desc (sb, block_group, &gd_bh); |
ac27a0ec DK |
522 | if (!desc) |
523 | goto error_return; | |
524 | ||
8fadc143 AR |
525 | if (in_range(ext4_block_bitmap(sb, desc), block, count) || |
526 | in_range(ext4_inode_bitmap(sb, desc), block, count) || | |
527 | in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) || | |
528 | in_range(block + count - 1, ext4_inode_table(sb, desc), | |
bd81d8ee | 529 | sbi->s_itb_per_group)) |
617ba13b | 530 | ext4_error (sb, "ext4_free_blocks", |
ac27a0ec | 531 | "Freeing blocks in system zones - " |
2ae02107 | 532 | "Block = %llu, count = %lu", |
ac27a0ec DK |
533 | block, count); |
534 | ||
535 | /* | |
536 | * We are about to start releasing blocks in the bitmap, | |
537 | * so we need undo access. | |
538 | */ | |
539 | /* @@@ check errors */ | |
540 | BUFFER_TRACE(bitmap_bh, "getting undo access"); | |
617ba13b | 541 | err = ext4_journal_get_undo_access(handle, bitmap_bh); |
ac27a0ec DK |
542 | if (err) |
543 | goto error_return; | |
544 | ||
545 | /* | |
546 | * We are about to modify some metadata. Call the journal APIs | |
547 | * to unshare ->b_data if a currently-committing transaction is | |
548 | * using it | |
549 | */ | |
550 | BUFFER_TRACE(gd_bh, "get_write_access"); | |
617ba13b | 551 | err = ext4_journal_get_write_access(handle, gd_bh); |
ac27a0ec DK |
552 | if (err) |
553 | goto error_return; | |
554 | ||
555 | jbd_lock_bh_state(bitmap_bh); | |
556 | ||
557 | for (i = 0, group_freed = 0; i < count; i++) { | |
558 | /* | |
559 | * An HJ special. This is expensive... | |
560 | */ | |
e23291b9 | 561 | #ifdef CONFIG_JBD2_DEBUG |
ac27a0ec DK |
562 | jbd_unlock_bh_state(bitmap_bh); |
563 | { | |
564 | struct buffer_head *debug_bh; | |
565 | debug_bh = sb_find_get_block(sb, block + i); | |
566 | if (debug_bh) { | |
567 | BUFFER_TRACE(debug_bh, "Deleted!"); | |
568 | if (!bh2jh(bitmap_bh)->b_committed_data) | |
569 | BUFFER_TRACE(debug_bh, | |
570 | "No commited data in bitmap"); | |
571 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap"); | |
572 | __brelse(debug_bh); | |
573 | } | |
574 | } | |
575 | jbd_lock_bh_state(bitmap_bh); | |
576 | #endif | |
577 | if (need_resched()) { | |
578 | jbd_unlock_bh_state(bitmap_bh); | |
579 | cond_resched(); | |
580 | jbd_lock_bh_state(bitmap_bh); | |
581 | } | |
582 | /* @@@ This prevents newly-allocated data from being | |
583 | * freed and then reallocated within the same | |
584 | * transaction. | |
585 | * | |
586 | * Ideally we would want to allow that to happen, but to | |
dab291af | 587 | * do so requires making jbd2_journal_forget() capable of |
ac27a0ec DK |
588 | * revoking the queued write of a data block, which |
589 | * implies blocking on the journal lock. *forget() | |
590 | * cannot block due to truncate races. | |
591 | * | |
dab291af | 592 | * Eventually we can fix this by making jbd2_journal_forget() |
ac27a0ec DK |
593 | * return a status indicating whether or not it was able |
594 | * to revoke the buffer. On successful revoke, it is | |
595 | * safe not to set the allocation bit in the committed | |
596 | * bitmap, because we know that there is no outstanding | |
597 | * activity on the buffer any more and so it is safe to | |
598 | * reallocate it. | |
599 | */ | |
600 | BUFFER_TRACE(bitmap_bh, "set in b_committed_data"); | |
601 | J_ASSERT_BH(bitmap_bh, | |
602 | bh2jh(bitmap_bh)->b_committed_data != NULL); | |
617ba13b | 603 | ext4_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i, |
ac27a0ec DK |
604 | bh2jh(bitmap_bh)->b_committed_data); |
605 | ||
606 | /* | |
607 | * We clear the bit in the bitmap after setting the committed | |
608 | * data bit, because this is the reverse order to that which | |
609 | * the allocator uses. | |
610 | */ | |
611 | BUFFER_TRACE(bitmap_bh, "clear bit"); | |
617ba13b | 612 | if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group), |
ac27a0ec DK |
613 | bit + i, bitmap_bh->b_data)) { |
614 | jbd_unlock_bh_state(bitmap_bh); | |
617ba13b | 615 | ext4_error(sb, __FUNCTION__, |
2ae02107 | 616 | "bit already cleared for block %llu", |
bd81d8ee | 617 | (ext4_fsblk_t)(block + i)); |
ac27a0ec DK |
618 | jbd_lock_bh_state(bitmap_bh); |
619 | BUFFER_TRACE(bitmap_bh, "bit already cleared"); | |
620 | } else { | |
621 | group_freed++; | |
622 | } | |
623 | } | |
624 | jbd_unlock_bh_state(bitmap_bh); | |
625 | ||
626 | spin_lock(sb_bgl_lock(sbi, block_group)); | |
627 | desc->bg_free_blocks_count = | |
628 | cpu_to_le16(le16_to_cpu(desc->bg_free_blocks_count) + | |
629 | group_freed); | |
630 | spin_unlock(sb_bgl_lock(sbi, block_group)); | |
aa0dff2d | 631 | percpu_counter_add(&sbi->s_freeblocks_counter, count); |
ac27a0ec DK |
632 | |
633 | /* We dirtied the bitmap block */ | |
634 | BUFFER_TRACE(bitmap_bh, "dirtied bitmap block"); | |
617ba13b | 635 | err = ext4_journal_dirty_metadata(handle, bitmap_bh); |
ac27a0ec DK |
636 | |
637 | /* And the group descriptor block */ | |
638 | BUFFER_TRACE(gd_bh, "dirtied group descriptor block"); | |
617ba13b | 639 | ret = ext4_journal_dirty_metadata(handle, gd_bh); |
ac27a0ec DK |
640 | if (!err) err = ret; |
641 | *pdquot_freed_blocks += group_freed; | |
642 | ||
643 | if (overflow && !err) { | |
644 | block += count; | |
645 | count = overflow; | |
646 | goto do_more; | |
647 | } | |
648 | sb->s_dirt = 1; | |
649 | error_return: | |
650 | brelse(bitmap_bh); | |
617ba13b | 651 | ext4_std_error(sb, err); |
ac27a0ec DK |
652 | return; |
653 | } | |
654 | ||
655 | /** | |
617ba13b | 656 | * ext4_free_blocks() -- Free given blocks and update quota |
ac27a0ec DK |
657 | * @handle: handle for this transaction |
658 | * @inode: inode | |
659 | * @block: start physical block to free | |
660 | * @count: number of blocks to count | |
661 | */ | |
617ba13b MC |
662 | void ext4_free_blocks(handle_t *handle, struct inode *inode, |
663 | ext4_fsblk_t block, unsigned long count) | |
ac27a0ec DK |
664 | { |
665 | struct super_block * sb; | |
666 | unsigned long dquot_freed_blocks; | |
667 | ||
668 | sb = inode->i_sb; | |
669 | if (!sb) { | |
617ba13b | 670 | printk ("ext4_free_blocks: nonexistent device"); |
ac27a0ec DK |
671 | return; |
672 | } | |
617ba13b | 673 | ext4_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks); |
ac27a0ec DK |
674 | if (dquot_freed_blocks) |
675 | DQUOT_FREE_BLOCK(inode, dquot_freed_blocks); | |
676 | return; | |
677 | } | |
678 | ||
679 | /** | |
617ba13b | 680 | * ext4_test_allocatable() |
ac27a0ec DK |
681 | * @nr: given allocation block group |
682 | * @bh: bufferhead contains the bitmap of the given block group | |
683 | * | |
617ba13b | 684 | * For ext4 allocations, we must not reuse any blocks which are |
ac27a0ec DK |
685 | * allocated in the bitmap buffer's "last committed data" copy. This |
686 | * prevents deletes from freeing up the page for reuse until we have | |
687 | * committed the delete transaction. | |
688 | * | |
689 | * If we didn't do this, then deleting something and reallocating it as | |
690 | * data would allow the old block to be overwritten before the | |
691 | * transaction committed (because we force data to disk before commit). | |
692 | * This would lead to corruption if we crashed between overwriting the | |
693 | * data and committing the delete. | |
694 | * | |
695 | * @@@ We may want to make this allocation behaviour conditional on | |
696 | * data-writes at some point, and disable it for metadata allocations or | |
697 | * sync-data inodes. | |
698 | */ | |
617ba13b | 699 | static int ext4_test_allocatable(ext4_grpblk_t nr, struct buffer_head *bh) |
ac27a0ec DK |
700 | { |
701 | int ret; | |
702 | struct journal_head *jh = bh2jh(bh); | |
703 | ||
617ba13b | 704 | if (ext4_test_bit(nr, bh->b_data)) |
ac27a0ec DK |
705 | return 0; |
706 | ||
707 | jbd_lock_bh_state(bh); | |
708 | if (!jh->b_committed_data) | |
709 | ret = 1; | |
710 | else | |
617ba13b | 711 | ret = !ext4_test_bit(nr, jh->b_committed_data); |
ac27a0ec DK |
712 | jbd_unlock_bh_state(bh); |
713 | return ret; | |
714 | } | |
715 | ||
716 | /** | |
717 | * bitmap_search_next_usable_block() | |
718 | * @start: the starting block (group relative) of the search | |
719 | * @bh: bufferhead contains the block group bitmap | |
720 | * @maxblocks: the ending block (group relative) of the reservation | |
721 | * | |
722 | * The bitmap search --- search forward alternately through the actual | |
723 | * bitmap on disk and the last-committed copy in journal, until we find a | |
724 | * bit free in both bitmaps. | |
725 | */ | |
617ba13b MC |
726 | static ext4_grpblk_t |
727 | bitmap_search_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh, | |
728 | ext4_grpblk_t maxblocks) | |
ac27a0ec | 729 | { |
617ba13b | 730 | ext4_grpblk_t next; |
ac27a0ec DK |
731 | struct journal_head *jh = bh2jh(bh); |
732 | ||
733 | while (start < maxblocks) { | |
617ba13b | 734 | next = ext4_find_next_zero_bit(bh->b_data, maxblocks, start); |
ac27a0ec DK |
735 | if (next >= maxblocks) |
736 | return -1; | |
617ba13b | 737 | if (ext4_test_allocatable(next, bh)) |
ac27a0ec DK |
738 | return next; |
739 | jbd_lock_bh_state(bh); | |
740 | if (jh->b_committed_data) | |
617ba13b | 741 | start = ext4_find_next_zero_bit(jh->b_committed_data, |
ac27a0ec DK |
742 | maxblocks, next); |
743 | jbd_unlock_bh_state(bh); | |
744 | } | |
745 | return -1; | |
746 | } | |
747 | ||
748 | /** | |
749 | * find_next_usable_block() | |
750 | * @start: the starting block (group relative) to find next | |
751 | * allocatable block in bitmap. | |
752 | * @bh: bufferhead contains the block group bitmap | |
753 | * @maxblocks: the ending block (group relative) for the search | |
754 | * | |
755 | * Find an allocatable block in a bitmap. We honor both the bitmap and | |
756 | * its last-committed copy (if that exists), and perform the "most | |
757 | * appropriate allocation" algorithm of looking for a free block near | |
758 | * the initial goal; then for a free byte somewhere in the bitmap; then | |
759 | * for any free bit in the bitmap. | |
760 | */ | |
617ba13b MC |
761 | static ext4_grpblk_t |
762 | find_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh, | |
763 | ext4_grpblk_t maxblocks) | |
ac27a0ec | 764 | { |
617ba13b | 765 | ext4_grpblk_t here, next; |
ac27a0ec DK |
766 | char *p, *r; |
767 | ||
768 | if (start > 0) { | |
769 | /* | |
770 | * The goal was occupied; search forward for a free | |
771 | * block within the next XX blocks. | |
772 | * | |
773 | * end_goal is more or less random, but it has to be | |
617ba13b | 774 | * less than EXT4_BLOCKS_PER_GROUP. Aligning up to the |
ac27a0ec DK |
775 | * next 64-bit boundary is simple.. |
776 | */ | |
617ba13b | 777 | ext4_grpblk_t end_goal = (start + 63) & ~63; |
ac27a0ec DK |
778 | if (end_goal > maxblocks) |
779 | end_goal = maxblocks; | |
617ba13b MC |
780 | here = ext4_find_next_zero_bit(bh->b_data, end_goal, start); |
781 | if (here < end_goal && ext4_test_allocatable(here, bh)) | |
ac27a0ec | 782 | return here; |
617ba13b | 783 | ext4_debug("Bit not found near goal\n"); |
ac27a0ec DK |
784 | } |
785 | ||
786 | here = start; | |
787 | if (here < 0) | |
788 | here = 0; | |
789 | ||
790 | p = ((char *)bh->b_data) + (here >> 3); | |
ec0837f2 | 791 | r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3)); |
ac27a0ec DK |
792 | next = (r - ((char *)bh->b_data)) << 3; |
793 | ||
617ba13b | 794 | if (next < maxblocks && next >= start && ext4_test_allocatable(next, bh)) |
ac27a0ec DK |
795 | return next; |
796 | ||
797 | /* | |
798 | * The bitmap search --- search forward alternately through the actual | |
799 | * bitmap and the last-committed copy until we find a bit free in | |
800 | * both | |
801 | */ | |
802 | here = bitmap_search_next_usable_block(here, bh, maxblocks); | |
803 | return here; | |
804 | } | |
805 | ||
806 | /** | |
807 | * claim_block() | |
808 | * @block: the free block (group relative) to allocate | |
809 | * @bh: the bufferhead containts the block group bitmap | |
810 | * | |
811 | * We think we can allocate this block in this bitmap. Try to set the bit. | |
812 | * If that succeeds then check that nobody has allocated and then freed the | |
813 | * block since we saw that is was not marked in b_committed_data. If it _was_ | |
814 | * allocated and freed then clear the bit in the bitmap again and return | |
815 | * zero (failure). | |
816 | */ | |
817 | static inline int | |
617ba13b | 818 | claim_block(spinlock_t *lock, ext4_grpblk_t block, struct buffer_head *bh) |
ac27a0ec DK |
819 | { |
820 | struct journal_head *jh = bh2jh(bh); | |
821 | int ret; | |
822 | ||
617ba13b | 823 | if (ext4_set_bit_atomic(lock, block, bh->b_data)) |
ac27a0ec DK |
824 | return 0; |
825 | jbd_lock_bh_state(bh); | |
617ba13b MC |
826 | if (jh->b_committed_data && ext4_test_bit(block,jh->b_committed_data)) { |
827 | ext4_clear_bit_atomic(lock, block, bh->b_data); | |
ac27a0ec DK |
828 | ret = 0; |
829 | } else { | |
830 | ret = 1; | |
831 | } | |
832 | jbd_unlock_bh_state(bh); | |
833 | return ret; | |
834 | } | |
835 | ||
836 | /** | |
617ba13b | 837 | * ext4_try_to_allocate() |
ac27a0ec DK |
838 | * @sb: superblock |
839 | * @handle: handle to this transaction | |
840 | * @group: given allocation block group | |
841 | * @bitmap_bh: bufferhead holds the block bitmap | |
842 | * @grp_goal: given target block within the group | |
843 | * @count: target number of blocks to allocate | |
844 | * @my_rsv: reservation window | |
845 | * | |
846 | * Attempt to allocate blocks within a give range. Set the range of allocation | |
847 | * first, then find the first free bit(s) from the bitmap (within the range), | |
848 | * and at last, allocate the blocks by claiming the found free bit as allocated. | |
849 | * | |
850 | * To set the range of this allocation: | |
851 | * if there is a reservation window, only try to allocate block(s) from the | |
852 | * file's own reservation window; | |
853 | * Otherwise, the allocation range starts from the give goal block, ends at | |
854 | * the block group's last block. | |
855 | * | |
856 | * If we failed to allocate the desired block then we may end up crossing to a | |
857 | * new bitmap. In that case we must release write access to the old one via | |
617ba13b | 858 | * ext4_journal_release_buffer(), else we'll run out of credits. |
ac27a0ec | 859 | */ |
617ba13b MC |
860 | static ext4_grpblk_t |
861 | ext4_try_to_allocate(struct super_block *sb, handle_t *handle, int group, | |
862 | struct buffer_head *bitmap_bh, ext4_grpblk_t grp_goal, | |
863 | unsigned long *count, struct ext4_reserve_window *my_rsv) | |
ac27a0ec | 864 | { |
617ba13b MC |
865 | ext4_fsblk_t group_first_block; |
866 | ext4_grpblk_t start, end; | |
ac27a0ec DK |
867 | unsigned long num = 0; |
868 | ||
869 | /* we do allocation within the reservation window if we have a window */ | |
870 | if (my_rsv) { | |
617ba13b | 871 | group_first_block = ext4_group_first_block_no(sb, group); |
ac27a0ec DK |
872 | if (my_rsv->_rsv_start >= group_first_block) |
873 | start = my_rsv->_rsv_start - group_first_block; | |
874 | else | |
875 | /* reservation window cross group boundary */ | |
876 | start = 0; | |
877 | end = my_rsv->_rsv_end - group_first_block + 1; | |
617ba13b | 878 | if (end > EXT4_BLOCKS_PER_GROUP(sb)) |
ac27a0ec | 879 | /* reservation window crosses group boundary */ |
617ba13b | 880 | end = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
881 | if ((start <= grp_goal) && (grp_goal < end)) |
882 | start = grp_goal; | |
883 | else | |
884 | grp_goal = -1; | |
885 | } else { | |
886 | if (grp_goal > 0) | |
887 | start = grp_goal; | |
888 | else | |
889 | start = 0; | |
617ba13b | 890 | end = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
891 | } |
892 | ||
617ba13b | 893 | BUG_ON(start > EXT4_BLOCKS_PER_GROUP(sb)); |
ac27a0ec DK |
894 | |
895 | repeat: | |
617ba13b | 896 | if (grp_goal < 0 || !ext4_test_allocatable(grp_goal, bitmap_bh)) { |
ac27a0ec DK |
897 | grp_goal = find_next_usable_block(start, bitmap_bh, end); |
898 | if (grp_goal < 0) | |
899 | goto fail_access; | |
900 | if (!my_rsv) { | |
901 | int i; | |
902 | ||
903 | for (i = 0; i < 7 && grp_goal > start && | |
617ba13b | 904 | ext4_test_allocatable(grp_goal - 1, |
ac27a0ec DK |
905 | bitmap_bh); |
906 | i++, grp_goal--) | |
907 | ; | |
908 | } | |
909 | } | |
910 | start = grp_goal; | |
911 | ||
617ba13b | 912 | if (!claim_block(sb_bgl_lock(EXT4_SB(sb), group), |
ac27a0ec DK |
913 | grp_goal, bitmap_bh)) { |
914 | /* | |
915 | * The block was allocated by another thread, or it was | |
916 | * allocated and then freed by another thread | |
917 | */ | |
918 | start++; | |
919 | grp_goal++; | |
920 | if (start >= end) | |
921 | goto fail_access; | |
922 | goto repeat; | |
923 | } | |
924 | num++; | |
925 | grp_goal++; | |
926 | while (num < *count && grp_goal < end | |
617ba13b MC |
927 | && ext4_test_allocatable(grp_goal, bitmap_bh) |
928 | && claim_block(sb_bgl_lock(EXT4_SB(sb), group), | |
ac27a0ec DK |
929 | grp_goal, bitmap_bh)) { |
930 | num++; | |
931 | grp_goal++; | |
932 | } | |
933 | *count = num; | |
934 | return grp_goal - num; | |
935 | fail_access: | |
936 | *count = num; | |
937 | return -1; | |
938 | } | |
939 | ||
940 | /** | |
941 | * find_next_reservable_window(): | |
942 | * find a reservable space within the given range. | |
943 | * It does not allocate the reservation window for now: | |
944 | * alloc_new_reservation() will do the work later. | |
945 | * | |
946 | * @search_head: the head of the searching list; | |
947 | * This is not necessarily the list head of the whole filesystem | |
948 | * | |
949 | * We have both head and start_block to assist the search | |
950 | * for the reservable space. The list starts from head, | |
951 | * but we will shift to the place where start_block is, | |
952 | * then start from there, when looking for a reservable space. | |
953 | * | |
954 | * @size: the target new reservation window size | |
955 | * | |
956 | * @group_first_block: the first block we consider to start | |
957 | * the real search from | |
958 | * | |
959 | * @last_block: | |
960 | * the maximum block number that our goal reservable space | |
961 | * could start from. This is normally the last block in this | |
962 | * group. The search will end when we found the start of next | |
963 | * possible reservable space is out of this boundary. | |
964 | * This could handle the cross boundary reservation window | |
965 | * request. | |
966 | * | |
967 | * basically we search from the given range, rather than the whole | |
968 | * reservation double linked list, (start_block, last_block) | |
969 | * to find a free region that is of my size and has not | |
970 | * been reserved. | |
971 | * | |
972 | */ | |
973 | static int find_next_reservable_window( | |
617ba13b MC |
974 | struct ext4_reserve_window_node *search_head, |
975 | struct ext4_reserve_window_node *my_rsv, | |
ac27a0ec | 976 | struct super_block * sb, |
617ba13b MC |
977 | ext4_fsblk_t start_block, |
978 | ext4_fsblk_t last_block) | |
ac27a0ec DK |
979 | { |
980 | struct rb_node *next; | |
617ba13b MC |
981 | struct ext4_reserve_window_node *rsv, *prev; |
982 | ext4_fsblk_t cur; | |
ac27a0ec DK |
983 | int size = my_rsv->rsv_goal_size; |
984 | ||
985 | /* TODO: make the start of the reservation window byte-aligned */ | |
986 | /* cur = *start_block & ~7;*/ | |
987 | cur = start_block; | |
988 | rsv = search_head; | |
989 | if (!rsv) | |
990 | return -1; | |
991 | ||
992 | while (1) { | |
993 | if (cur <= rsv->rsv_end) | |
994 | cur = rsv->rsv_end + 1; | |
995 | ||
996 | /* TODO? | |
997 | * in the case we could not find a reservable space | |
998 | * that is what is expected, during the re-search, we could | |
999 | * remember what's the largest reservable space we could have | |
1000 | * and return that one. | |
1001 | * | |
1002 | * For now it will fail if we could not find the reservable | |
1003 | * space with expected-size (or more)... | |
1004 | */ | |
1005 | if (cur > last_block) | |
1006 | return -1; /* fail */ | |
1007 | ||
1008 | prev = rsv; | |
1009 | next = rb_next(&rsv->rsv_node); | |
b78a657f | 1010 | rsv = rb_entry(next,struct ext4_reserve_window_node,rsv_node); |
ac27a0ec DK |
1011 | |
1012 | /* | |
1013 | * Reached the last reservation, we can just append to the | |
1014 | * previous one. | |
1015 | */ | |
1016 | if (!next) | |
1017 | break; | |
1018 | ||
1019 | if (cur + size <= rsv->rsv_start) { | |
1020 | /* | |
1021 | * Found a reserveable space big enough. We could | |
1022 | * have a reservation across the group boundary here | |
1023 | */ | |
1024 | break; | |
1025 | } | |
1026 | } | |
1027 | /* | |
1028 | * we come here either : | |
1029 | * when we reach the end of the whole list, | |
1030 | * and there is empty reservable space after last entry in the list. | |
1031 | * append it to the end of the list. | |
1032 | * | |
1033 | * or we found one reservable space in the middle of the list, | |
1034 | * return the reservation window that we could append to. | |
1035 | * succeed. | |
1036 | */ | |
1037 | ||
1038 | if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window))) | |
1039 | rsv_window_remove(sb, my_rsv); | |
1040 | ||
1041 | /* | |
1042 | * Let's book the whole avaliable window for now. We will check the | |
1043 | * disk bitmap later and then, if there are free blocks then we adjust | |
1044 | * the window size if it's larger than requested. | |
1045 | * Otherwise, we will remove this node from the tree next time | |
1046 | * call find_next_reservable_window. | |
1047 | */ | |
1048 | my_rsv->rsv_start = cur; | |
1049 | my_rsv->rsv_end = cur + size - 1; | |
1050 | my_rsv->rsv_alloc_hit = 0; | |
1051 | ||
1052 | if (prev != my_rsv) | |
617ba13b | 1053 | ext4_rsv_window_add(sb, my_rsv); |
ac27a0ec DK |
1054 | |
1055 | return 0; | |
1056 | } | |
1057 | ||
1058 | /** | |
1059 | * alloc_new_reservation()--allocate a new reservation window | |
1060 | * | |
1061 | * To make a new reservation, we search part of the filesystem | |
1062 | * reservation list (the list that inside the group). We try to | |
1063 | * allocate a new reservation window near the allocation goal, | |
1064 | * or the beginning of the group, if there is no goal. | |
1065 | * | |
1066 | * We first find a reservable space after the goal, then from | |
1067 | * there, we check the bitmap for the first free block after | |
1068 | * it. If there is no free block until the end of group, then the | |
1069 | * whole group is full, we failed. Otherwise, check if the free | |
1070 | * block is inside the expected reservable space, if so, we | |
1071 | * succeed. | |
1072 | * If the first free block is outside the reservable space, then | |
1073 | * start from the first free block, we search for next available | |
1074 | * space, and go on. | |
1075 | * | |
1076 | * on succeed, a new reservation will be found and inserted into the list | |
1077 | * It contains at least one free block, and it does not overlap with other | |
1078 | * reservation windows. | |
1079 | * | |
1080 | * failed: we failed to find a reservation window in this group | |
1081 | * | |
1082 | * @rsv: the reservation | |
1083 | * | |
1084 | * @grp_goal: The goal (group-relative). It is where the search for a | |
1085 | * free reservable space should start from. | |
1086 | * if we have a grp_goal(grp_goal >0 ), then start from there, | |
1087 | * no grp_goal(grp_goal = -1), we start from the first block | |
1088 | * of the group. | |
1089 | * | |
1090 | * @sb: the super block | |
1091 | * @group: the group we are trying to allocate in | |
1092 | * @bitmap_bh: the block group block bitmap | |
1093 | * | |
1094 | */ | |
617ba13b MC |
1095 | static int alloc_new_reservation(struct ext4_reserve_window_node *my_rsv, |
1096 | ext4_grpblk_t grp_goal, struct super_block *sb, | |
ac27a0ec DK |
1097 | unsigned int group, struct buffer_head *bitmap_bh) |
1098 | { | |
617ba13b MC |
1099 | struct ext4_reserve_window_node *search_head; |
1100 | ext4_fsblk_t group_first_block, group_end_block, start_block; | |
1101 | ext4_grpblk_t first_free_block; | |
1102 | struct rb_root *fs_rsv_root = &EXT4_SB(sb)->s_rsv_window_root; | |
ac27a0ec DK |
1103 | unsigned long size; |
1104 | int ret; | |
617ba13b | 1105 | spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock; |
ac27a0ec | 1106 | |
617ba13b MC |
1107 | group_first_block = ext4_group_first_block_no(sb, group); |
1108 | group_end_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
1109 | |
1110 | if (grp_goal < 0) | |
1111 | start_block = group_first_block; | |
1112 | else | |
1113 | start_block = grp_goal + group_first_block; | |
1114 | ||
1115 | size = my_rsv->rsv_goal_size; | |
1116 | ||
1117 | if (!rsv_is_empty(&my_rsv->rsv_window)) { | |
1118 | /* | |
1119 | * if the old reservation is cross group boundary | |
1120 | * and if the goal is inside the old reservation window, | |
1121 | * we will come here when we just failed to allocate from | |
1122 | * the first part of the window. We still have another part | |
1123 | * that belongs to the next group. In this case, there is no | |
1124 | * point to discard our window and try to allocate a new one | |
1125 | * in this group(which will fail). we should | |
1126 | * keep the reservation window, just simply move on. | |
1127 | * | |
1128 | * Maybe we could shift the start block of the reservation | |
1129 | * window to the first block of next group. | |
1130 | */ | |
1131 | ||
1132 | if ((my_rsv->rsv_start <= group_end_block) && | |
1133 | (my_rsv->rsv_end > group_end_block) && | |
1134 | (start_block >= my_rsv->rsv_start)) | |
1135 | return -1; | |
1136 | ||
1137 | if ((my_rsv->rsv_alloc_hit > | |
1138 | (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) { | |
1139 | /* | |
1140 | * if the previously allocation hit ratio is | |
1141 | * greater than 1/2, then we double the size of | |
1142 | * the reservation window the next time, | |
1143 | * otherwise we keep the same size window | |
1144 | */ | |
1145 | size = size * 2; | |
617ba13b MC |
1146 | if (size > EXT4_MAX_RESERVE_BLOCKS) |
1147 | size = EXT4_MAX_RESERVE_BLOCKS; | |
ac27a0ec DK |
1148 | my_rsv->rsv_goal_size= size; |
1149 | } | |
1150 | } | |
1151 | ||
1152 | spin_lock(rsv_lock); | |
1153 | /* | |
1154 | * shift the search start to the window near the goal block | |
1155 | */ | |
1156 | search_head = search_reserve_window(fs_rsv_root, start_block); | |
1157 | ||
1158 | /* | |
1159 | * find_next_reservable_window() simply finds a reservable window | |
1160 | * inside the given range(start_block, group_end_block). | |
1161 | * | |
1162 | * To make sure the reservation window has a free bit inside it, we | |
1163 | * need to check the bitmap after we found a reservable window. | |
1164 | */ | |
1165 | retry: | |
1166 | ret = find_next_reservable_window(search_head, my_rsv, sb, | |
1167 | start_block, group_end_block); | |
1168 | ||
1169 | if (ret == -1) { | |
1170 | if (!rsv_is_empty(&my_rsv->rsv_window)) | |
1171 | rsv_window_remove(sb, my_rsv); | |
1172 | spin_unlock(rsv_lock); | |
1173 | return -1; | |
1174 | } | |
1175 | ||
1176 | /* | |
1177 | * On success, find_next_reservable_window() returns the | |
1178 | * reservation window where there is a reservable space after it. | |
1179 | * Before we reserve this reservable space, we need | |
1180 | * to make sure there is at least a free block inside this region. | |
1181 | * | |
1182 | * searching the first free bit on the block bitmap and copy of | |
1183 | * last committed bitmap alternatively, until we found a allocatable | |
1184 | * block. Search start from the start block of the reservable space | |
1185 | * we just found. | |
1186 | */ | |
1187 | spin_unlock(rsv_lock); | |
1188 | first_free_block = bitmap_search_next_usable_block( | |
1189 | my_rsv->rsv_start - group_first_block, | |
1190 | bitmap_bh, group_end_block - group_first_block + 1); | |
1191 | ||
1192 | if (first_free_block < 0) { | |
1193 | /* | |
1194 | * no free block left on the bitmap, no point | |
1195 | * to reserve the space. return failed. | |
1196 | */ | |
1197 | spin_lock(rsv_lock); | |
1198 | if (!rsv_is_empty(&my_rsv->rsv_window)) | |
1199 | rsv_window_remove(sb, my_rsv); | |
1200 | spin_unlock(rsv_lock); | |
1201 | return -1; /* failed */ | |
1202 | } | |
1203 | ||
1204 | start_block = first_free_block + group_first_block; | |
1205 | /* | |
1206 | * check if the first free block is within the | |
1207 | * free space we just reserved | |
1208 | */ | |
b2f2c76d | 1209 | if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end) |
ac27a0ec DK |
1210 | return 0; /* success */ |
1211 | /* | |
1212 | * if the first free bit we found is out of the reservable space | |
1213 | * continue search for next reservable space, | |
1214 | * start from where the free block is, | |
1215 | * we also shift the list head to where we stopped last time | |
1216 | */ | |
1217 | search_head = my_rsv; | |
1218 | spin_lock(rsv_lock); | |
1219 | goto retry; | |
1220 | } | |
1221 | ||
1222 | /** | |
1223 | * try_to_extend_reservation() | |
1224 | * @my_rsv: given reservation window | |
1225 | * @sb: super block | |
1226 | * @size: the delta to extend | |
1227 | * | |
1228 | * Attempt to expand the reservation window large enough to have | |
1229 | * required number of free blocks | |
1230 | * | |
617ba13b | 1231 | * Since ext4_try_to_allocate() will always allocate blocks within |
ac27a0ec DK |
1232 | * the reservation window range, if the window size is too small, |
1233 | * multiple blocks allocation has to stop at the end of the reservation | |
1234 | * window. To make this more efficient, given the total number of | |
1235 | * blocks needed and the current size of the window, we try to | |
1236 | * expand the reservation window size if necessary on a best-effort | |
617ba13b | 1237 | * basis before ext4_new_blocks() tries to allocate blocks, |
ac27a0ec | 1238 | */ |
617ba13b | 1239 | static void try_to_extend_reservation(struct ext4_reserve_window_node *my_rsv, |
ac27a0ec DK |
1240 | struct super_block *sb, int size) |
1241 | { | |
617ba13b | 1242 | struct ext4_reserve_window_node *next_rsv; |
ac27a0ec | 1243 | struct rb_node *next; |
617ba13b | 1244 | spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock; |
ac27a0ec DK |
1245 | |
1246 | if (!spin_trylock(rsv_lock)) | |
1247 | return; | |
1248 | ||
1249 | next = rb_next(&my_rsv->rsv_node); | |
1250 | ||
1251 | if (!next) | |
1252 | my_rsv->rsv_end += size; | |
1253 | else { | |
b78a657f | 1254 | next_rsv = rb_entry(next, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
1255 | |
1256 | if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size) | |
1257 | my_rsv->rsv_end += size; | |
1258 | else | |
1259 | my_rsv->rsv_end = next_rsv->rsv_start - 1; | |
1260 | } | |
1261 | spin_unlock(rsv_lock); | |
1262 | } | |
1263 | ||
1264 | /** | |
617ba13b | 1265 | * ext4_try_to_allocate_with_rsv() |
ac27a0ec DK |
1266 | * @sb: superblock |
1267 | * @handle: handle to this transaction | |
1268 | * @group: given allocation block group | |
1269 | * @bitmap_bh: bufferhead holds the block bitmap | |
1270 | * @grp_goal: given target block within the group | |
1271 | * @count: target number of blocks to allocate | |
1272 | * @my_rsv: reservation window | |
1273 | * @errp: pointer to store the error code | |
1274 | * | |
1275 | * This is the main function used to allocate a new block and its reservation | |
1276 | * window. | |
1277 | * | |
1278 | * Each time when a new block allocation is need, first try to allocate from | |
1279 | * its own reservation. If it does not have a reservation window, instead of | |
1280 | * looking for a free bit on bitmap first, then look up the reservation list to | |
1281 | * see if it is inside somebody else's reservation window, we try to allocate a | |
1282 | * reservation window for it starting from the goal first. Then do the block | |
1283 | * allocation within the reservation window. | |
1284 | * | |
1285 | * This will avoid keeping on searching the reservation list again and | |
1286 | * again when somebody is looking for a free block (without | |
1287 | * reservation), and there are lots of free blocks, but they are all | |
1288 | * being reserved. | |
1289 | * | |
1290 | * We use a red-black tree for the per-filesystem reservation list. | |
1291 | * | |
1292 | */ | |
617ba13b MC |
1293 | static ext4_grpblk_t |
1294 | ext4_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle, | |
ac27a0ec | 1295 | unsigned int group, struct buffer_head *bitmap_bh, |
617ba13b MC |
1296 | ext4_grpblk_t grp_goal, |
1297 | struct ext4_reserve_window_node * my_rsv, | |
ac27a0ec DK |
1298 | unsigned long *count, int *errp) |
1299 | { | |
617ba13b MC |
1300 | ext4_fsblk_t group_first_block, group_last_block; |
1301 | ext4_grpblk_t ret = 0; | |
ac27a0ec DK |
1302 | int fatal; |
1303 | unsigned long num = *count; | |
1304 | ||
1305 | *errp = 0; | |
1306 | ||
1307 | /* | |
1308 | * Make sure we use undo access for the bitmap, because it is critical | |
1309 | * that we do the frozen_data COW on bitmap buffers in all cases even | |
1310 | * if the buffer is in BJ_Forget state in the committing transaction. | |
1311 | */ | |
1312 | BUFFER_TRACE(bitmap_bh, "get undo access for new block"); | |
617ba13b | 1313 | fatal = ext4_journal_get_undo_access(handle, bitmap_bh); |
ac27a0ec DK |
1314 | if (fatal) { |
1315 | *errp = fatal; | |
1316 | return -1; | |
1317 | } | |
1318 | ||
1319 | /* | |
1320 | * we don't deal with reservation when | |
1321 | * filesystem is mounted without reservation | |
1322 | * or the file is not a regular file | |
1323 | * or last attempt to allocate a block with reservation turned on failed | |
1324 | */ | |
1325 | if (my_rsv == NULL ) { | |
617ba13b | 1326 | ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh, |
ac27a0ec DK |
1327 | grp_goal, count, NULL); |
1328 | goto out; | |
1329 | } | |
1330 | /* | |
1331 | * grp_goal is a group relative block number (if there is a goal) | |
e7dc95db | 1332 | * 0 <= grp_goal < EXT4_BLOCKS_PER_GROUP(sb) |
ac27a0ec DK |
1333 | * first block is a filesystem wide block number |
1334 | * first block is the block number of the first block in this group | |
1335 | */ | |
617ba13b MC |
1336 | group_first_block = ext4_group_first_block_no(sb, group); |
1337 | group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
1338 | |
1339 | /* | |
1340 | * Basically we will allocate a new block from inode's reservation | |
1341 | * window. | |
1342 | * | |
1343 | * We need to allocate a new reservation window, if: | |
1344 | * a) inode does not have a reservation window; or | |
1345 | * b) last attempt to allocate a block from existing reservation | |
1346 | * failed; or | |
1347 | * c) we come here with a goal and with a reservation window | |
1348 | * | |
1349 | * We do not need to allocate a new reservation window if we come here | |
1350 | * at the beginning with a goal and the goal is inside the window, or | |
1351 | * we don't have a goal but already have a reservation window. | |
1352 | * then we could go to allocate from the reservation window directly. | |
1353 | */ | |
1354 | while (1) { | |
1355 | if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) || | |
1356 | !goal_in_my_reservation(&my_rsv->rsv_window, | |
1357 | grp_goal, group, sb)) { | |
1358 | if (my_rsv->rsv_goal_size < *count) | |
1359 | my_rsv->rsv_goal_size = *count; | |
1360 | ret = alloc_new_reservation(my_rsv, grp_goal, sb, | |
1361 | group, bitmap_bh); | |
1362 | if (ret < 0) | |
1363 | break; /* failed */ | |
1364 | ||
1365 | if (!goal_in_my_reservation(&my_rsv->rsv_window, | |
1366 | grp_goal, group, sb)) | |
1367 | grp_goal = -1; | |
e7dc95db | 1368 | } else if (grp_goal >= 0) { |
1df1e63b MC |
1369 | int curr = my_rsv->rsv_end - |
1370 | (grp_goal + group_first_block) + 1; | |
1371 | ||
1372 | if (curr < *count) | |
1373 | try_to_extend_reservation(my_rsv, sb, | |
1374 | *count - curr); | |
1375 | } | |
ac27a0ec DK |
1376 | |
1377 | if ((my_rsv->rsv_start > group_last_block) || | |
1378 | (my_rsv->rsv_end < group_first_block)) { | |
617ba13b | 1379 | rsv_window_dump(&EXT4_SB(sb)->s_rsv_window_root, 1); |
ac27a0ec DK |
1380 | BUG(); |
1381 | } | |
617ba13b | 1382 | ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh, |
ac27a0ec DK |
1383 | grp_goal, &num, &my_rsv->rsv_window); |
1384 | if (ret >= 0) { | |
1385 | my_rsv->rsv_alloc_hit += num; | |
1386 | *count = num; | |
1387 | break; /* succeed */ | |
1388 | } | |
1389 | num = *count; | |
1390 | } | |
1391 | out: | |
1392 | if (ret >= 0) { | |
1393 | BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for " | |
1394 | "bitmap block"); | |
617ba13b | 1395 | fatal = ext4_journal_dirty_metadata(handle, bitmap_bh); |
ac27a0ec DK |
1396 | if (fatal) { |
1397 | *errp = fatal; | |
1398 | return -1; | |
1399 | } | |
1400 | return ret; | |
1401 | } | |
1402 | ||
1403 | BUFFER_TRACE(bitmap_bh, "journal_release_buffer"); | |
617ba13b | 1404 | ext4_journal_release_buffer(handle, bitmap_bh); |
ac27a0ec DK |
1405 | return ret; |
1406 | } | |
1407 | ||
1408 | /** | |
617ba13b | 1409 | * ext4_has_free_blocks() |
ac27a0ec DK |
1410 | * @sbi: in-core super block structure. |
1411 | * | |
1412 | * Check if filesystem has at least 1 free block available for allocation. | |
1413 | */ | |
617ba13b | 1414 | static int ext4_has_free_blocks(struct ext4_sb_info *sbi) |
ac27a0ec | 1415 | { |
617ba13b | 1416 | ext4_fsblk_t free_blocks, root_blocks; |
ac27a0ec DK |
1417 | |
1418 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
bd81d8ee | 1419 | root_blocks = ext4_r_blocks_count(sbi->s_es); |
ac27a0ec DK |
1420 | if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) && |
1421 | sbi->s_resuid != current->fsuid && | |
1422 | (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) { | |
1423 | return 0; | |
1424 | } | |
1425 | return 1; | |
1426 | } | |
1427 | ||
1428 | /** | |
617ba13b | 1429 | * ext4_should_retry_alloc() |
ac27a0ec DK |
1430 | * @sb: super block |
1431 | * @retries number of attemps has been made | |
1432 | * | |
617ba13b | 1433 | * ext4_should_retry_alloc() is called when ENOSPC is returned, and if |
ac27a0ec DK |
1434 | * it is profitable to retry the operation, this function will wait |
1435 | * for the current or commiting transaction to complete, and then | |
1436 | * return TRUE. | |
1437 | * | |
1438 | * if the total number of retries exceed three times, return FALSE. | |
1439 | */ | |
617ba13b | 1440 | int ext4_should_retry_alloc(struct super_block *sb, int *retries) |
ac27a0ec | 1441 | { |
617ba13b | 1442 | if (!ext4_has_free_blocks(EXT4_SB(sb)) || (*retries)++ > 3) |
ac27a0ec DK |
1443 | return 0; |
1444 | ||
1445 | jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id); | |
1446 | ||
dab291af | 1447 | return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal); |
ac27a0ec DK |
1448 | } |
1449 | ||
1450 | /** | |
617ba13b | 1451 | * ext4_new_blocks() -- core block(s) allocation function |
ac27a0ec DK |
1452 | * @handle: handle to this transaction |
1453 | * @inode: file inode | |
1454 | * @goal: given target block(filesystem wide) | |
1455 | * @count: target number of blocks to allocate | |
1456 | * @errp: error code | |
1457 | * | |
617ba13b | 1458 | * ext4_new_blocks uses a goal block to assist allocation. It tries to |
ac27a0ec DK |
1459 | * allocate block(s) from the block group contains the goal block first. If that |
1460 | * fails, it will try to allocate block(s) from other block groups without | |
1461 | * any specific goal block. | |
1462 | * | |
1463 | */ | |
617ba13b MC |
1464 | ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode, |
1465 | ext4_fsblk_t goal, unsigned long *count, int *errp) | |
ac27a0ec DK |
1466 | { |
1467 | struct buffer_head *bitmap_bh = NULL; | |
1468 | struct buffer_head *gdp_bh; | |
3a5b2ecd | 1469 | unsigned long group_no; |
ac27a0ec | 1470 | int goal_group; |
617ba13b MC |
1471 | ext4_grpblk_t grp_target_blk; /* blockgroup relative goal block */ |
1472 | ext4_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/ | |
1473 | ext4_fsblk_t ret_block; /* filesyetem-wide allocated block */ | |
ac27a0ec DK |
1474 | int bgi; /* blockgroup iteration index */ |
1475 | int fatal = 0, err; | |
1476 | int performed_allocation = 0; | |
617ba13b | 1477 | ext4_grpblk_t free_blocks; /* number of free blocks in a group */ |
ac27a0ec | 1478 | struct super_block *sb; |
617ba13b MC |
1479 | struct ext4_group_desc *gdp; |
1480 | struct ext4_super_block *es; | |
1481 | struct ext4_sb_info *sbi; | |
1482 | struct ext4_reserve_window_node *my_rsv = NULL; | |
1483 | struct ext4_block_alloc_info *block_i; | |
ac27a0ec | 1484 | unsigned short windowsz = 0; |
617ba13b | 1485 | #ifdef EXT4FS_DEBUG |
ac27a0ec DK |
1486 | static int goal_hits, goal_attempts; |
1487 | #endif | |
1488 | unsigned long ngroups; | |
1489 | unsigned long num = *count; | |
1490 | ||
1491 | *errp = -ENOSPC; | |
1492 | sb = inode->i_sb; | |
1493 | if (!sb) { | |
617ba13b | 1494 | printk("ext4_new_block: nonexistent device"); |
ac27a0ec DK |
1495 | return 0; |
1496 | } | |
1497 | ||
1498 | /* | |
1499 | * Check quota for allocation of this block. | |
1500 | */ | |
1501 | if (DQUOT_ALLOC_BLOCK(inode, num)) { | |
1502 | *errp = -EDQUOT; | |
1503 | return 0; | |
1504 | } | |
1505 | ||
617ba13b MC |
1506 | sbi = EXT4_SB(sb); |
1507 | es = EXT4_SB(sb)->s_es; | |
1508 | ext4_debug("goal=%lu.\n", goal); | |
ac27a0ec DK |
1509 | /* |
1510 | * Allocate a block from reservation only when | |
1511 | * filesystem is mounted with reservation(default,-o reservation), and | |
1512 | * it's a regular file, and | |
1513 | * the desired window size is greater than 0 (One could use ioctl | |
617ba13b | 1514 | * command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off |
ac27a0ec DK |
1515 | * reservation on that particular file) |
1516 | */ | |
617ba13b | 1517 | block_i = EXT4_I(inode)->i_block_alloc_info; |
ac27a0ec DK |
1518 | if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0)) |
1519 | my_rsv = &block_i->rsv_window_node; | |
1520 | ||
617ba13b | 1521 | if (!ext4_has_free_blocks(sbi)) { |
ac27a0ec DK |
1522 | *errp = -ENOSPC; |
1523 | goto out; | |
1524 | } | |
1525 | ||
1526 | /* | |
1527 | * First, test whether the goal block is free. | |
1528 | */ | |
1529 | if (goal < le32_to_cpu(es->s_first_data_block) || | |
bd81d8ee | 1530 | goal >= ext4_blocks_count(es)) |
ac27a0ec | 1531 | goal = le32_to_cpu(es->s_first_data_block); |
3a5b2ecd | 1532 | ext4_get_group_no_and_offset(sb, goal, &group_no, &grp_target_blk); |
ac27a0ec DK |
1533 | goal_group = group_no; |
1534 | retry_alloc: | |
617ba13b | 1535 | gdp = ext4_get_group_desc(sb, group_no, &gdp_bh); |
ac27a0ec DK |
1536 | if (!gdp) |
1537 | goto io_error; | |
1538 | ||
1539 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); | |
1540 | /* | |
1541 | * if there is not enough free blocks to make a new resevation | |
1542 | * turn off reservation for this allocation | |
1543 | */ | |
1544 | if (my_rsv && (free_blocks < windowsz) | |
1545 | && (rsv_is_empty(&my_rsv->rsv_window))) | |
1546 | my_rsv = NULL; | |
1547 | ||
1548 | if (free_blocks > 0) { | |
ac27a0ec DK |
1549 | bitmap_bh = read_block_bitmap(sb, group_no); |
1550 | if (!bitmap_bh) | |
1551 | goto io_error; | |
617ba13b | 1552 | grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, |
ac27a0ec DK |
1553 | group_no, bitmap_bh, grp_target_blk, |
1554 | my_rsv, &num, &fatal); | |
1555 | if (fatal) | |
1556 | goto out; | |
1557 | if (grp_alloc_blk >= 0) | |
1558 | goto allocated; | |
1559 | } | |
1560 | ||
617ba13b | 1561 | ngroups = EXT4_SB(sb)->s_groups_count; |
ac27a0ec DK |
1562 | smp_rmb(); |
1563 | ||
1564 | /* | |
1565 | * Now search the rest of the groups. We assume that | |
1566 | * i and gdp correctly point to the last group visited. | |
1567 | */ | |
1568 | for (bgi = 0; bgi < ngroups; bgi++) { | |
1569 | group_no++; | |
1570 | if (group_no >= ngroups) | |
1571 | group_no = 0; | |
617ba13b | 1572 | gdp = ext4_get_group_desc(sb, group_no, &gdp_bh); |
341cee43 HD |
1573 | if (!gdp) |
1574 | goto io_error; | |
ac27a0ec DK |
1575 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); |
1576 | /* | |
1577 | * skip this group if the number of | |
1578 | * free blocks is less than half of the reservation | |
1579 | * window size. | |
1580 | */ | |
1581 | if (free_blocks <= (windowsz/2)) | |
1582 | continue; | |
1583 | ||
1584 | brelse(bitmap_bh); | |
1585 | bitmap_bh = read_block_bitmap(sb, group_no); | |
1586 | if (!bitmap_bh) | |
1587 | goto io_error; | |
1588 | /* | |
1589 | * try to allocate block(s) from this group, without a goal(-1). | |
1590 | */ | |
617ba13b | 1591 | grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, |
ac27a0ec DK |
1592 | group_no, bitmap_bh, -1, my_rsv, |
1593 | &num, &fatal); | |
1594 | if (fatal) | |
1595 | goto out; | |
1596 | if (grp_alloc_blk >= 0) | |
1597 | goto allocated; | |
1598 | } | |
1599 | /* | |
1600 | * We may end up a bogus ealier ENOSPC error due to | |
1601 | * filesystem is "full" of reservations, but | |
1602 | * there maybe indeed free blocks avaliable on disk | |
1603 | * In this case, we just forget about the reservations | |
1604 | * just do block allocation as without reservations. | |
1605 | */ | |
1606 | if (my_rsv) { | |
1607 | my_rsv = NULL; | |
cd16c8f7 | 1608 | windowsz = 0; |
ac27a0ec DK |
1609 | group_no = goal_group; |
1610 | goto retry_alloc; | |
1611 | } | |
1612 | /* No space left on the device */ | |
1613 | *errp = -ENOSPC; | |
1614 | goto out; | |
1615 | ||
1616 | allocated: | |
1617 | ||
617ba13b | 1618 | ext4_debug("using block group %d(%d)\n", |
ac27a0ec DK |
1619 | group_no, gdp->bg_free_blocks_count); |
1620 | ||
1621 | BUFFER_TRACE(gdp_bh, "get_write_access"); | |
617ba13b | 1622 | fatal = ext4_journal_get_write_access(handle, gdp_bh); |
ac27a0ec DK |
1623 | if (fatal) |
1624 | goto out; | |
1625 | ||
617ba13b | 1626 | ret_block = grp_alloc_blk + ext4_group_first_block_no(sb, group_no); |
ac27a0ec | 1627 | |
8fadc143 | 1628 | if (in_range(ext4_block_bitmap(sb, gdp), ret_block, num) || |
29bc5b4f | 1629 | in_range(ext4_inode_bitmap(sb, gdp), ret_block, num) || |
8fadc143 | 1630 | in_range(ret_block, ext4_inode_table(sb, gdp), |
bd81d8ee | 1631 | EXT4_SB(sb)->s_itb_per_group) || |
8fadc143 | 1632 | in_range(ret_block + num - 1, ext4_inode_table(sb, gdp), |
bd81d8ee | 1633 | EXT4_SB(sb)->s_itb_per_group)) |
617ba13b | 1634 | ext4_error(sb, "ext4_new_block", |
ac27a0ec | 1635 | "Allocating block in system zone - " |
2ae02107 | 1636 | "blocks from %llu, length %lu", |
ac27a0ec DK |
1637 | ret_block, num); |
1638 | ||
1639 | performed_allocation = 1; | |
1640 | ||
e23291b9 | 1641 | #ifdef CONFIG_JBD2_DEBUG |
ac27a0ec DK |
1642 | { |
1643 | struct buffer_head *debug_bh; | |
1644 | ||
1645 | /* Record bitmap buffer state in the newly allocated block */ | |
1646 | debug_bh = sb_find_get_block(sb, ret_block); | |
1647 | if (debug_bh) { | |
1648 | BUFFER_TRACE(debug_bh, "state when allocated"); | |
1649 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state"); | |
1650 | brelse(debug_bh); | |
1651 | } | |
1652 | } | |
1653 | jbd_lock_bh_state(bitmap_bh); | |
1654 | spin_lock(sb_bgl_lock(sbi, group_no)); | |
1655 | if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) { | |
1656 | int i; | |
1657 | ||
1658 | for (i = 0; i < num; i++) { | |
617ba13b | 1659 | if (ext4_test_bit(grp_alloc_blk+i, |
ac27a0ec DK |
1660 | bh2jh(bitmap_bh)->b_committed_data)) { |
1661 | printk("%s: block was unexpectedly set in " | |
1662 | "b_committed_data\n", __FUNCTION__); | |
1663 | } | |
1664 | } | |
1665 | } | |
617ba13b | 1666 | ext4_debug("found bit %d\n", grp_alloc_blk); |
ac27a0ec DK |
1667 | spin_unlock(sb_bgl_lock(sbi, group_no)); |
1668 | jbd_unlock_bh_state(bitmap_bh); | |
1669 | #endif | |
1670 | ||
bd81d8ee | 1671 | if (ret_block + num - 1 >= ext4_blocks_count(es)) { |
617ba13b | 1672 | ext4_error(sb, "ext4_new_block", |
2ae02107 | 1673 | "block(%llu) >= blocks count(%llu) - " |
3a5b2ecd | 1674 | "block_group = %lu, es == %p ", ret_block, |
bd81d8ee | 1675 | ext4_blocks_count(es), group_no, es); |
ac27a0ec DK |
1676 | goto out; |
1677 | } | |
1678 | ||
1679 | /* | |
1680 | * It is up to the caller to add the new buffer to a journal | |
1681 | * list of some description. We don't know in advance whether | |
1682 | * the caller wants to use it as metadata or data. | |
1683 | */ | |
617ba13b | 1684 | ext4_debug("allocating block %lu. Goal hits %d of %d.\n", |
ac27a0ec DK |
1685 | ret_block, goal_hits, goal_attempts); |
1686 | ||
1687 | spin_lock(sb_bgl_lock(sbi, group_no)); | |
1688 | gdp->bg_free_blocks_count = | |
1689 | cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count)-num); | |
1690 | spin_unlock(sb_bgl_lock(sbi, group_no)); | |
3cb4f9fa | 1691 | percpu_counter_sub(&sbi->s_freeblocks_counter, num); |
ac27a0ec DK |
1692 | |
1693 | BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor"); | |
617ba13b | 1694 | err = ext4_journal_dirty_metadata(handle, gdp_bh); |
ac27a0ec DK |
1695 | if (!fatal) |
1696 | fatal = err; | |
1697 | ||
1698 | sb->s_dirt = 1; | |
1699 | if (fatal) | |
1700 | goto out; | |
1701 | ||
1702 | *errp = 0; | |
1703 | brelse(bitmap_bh); | |
1704 | DQUOT_FREE_BLOCK(inode, *count-num); | |
1705 | *count = num; | |
1706 | return ret_block; | |
1707 | ||
1708 | io_error: | |
1709 | *errp = -EIO; | |
1710 | out: | |
1711 | if (fatal) { | |
1712 | *errp = fatal; | |
617ba13b | 1713 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
1714 | } |
1715 | /* | |
1716 | * Undo the block allocation | |
1717 | */ | |
1718 | if (!performed_allocation) | |
1719 | DQUOT_FREE_BLOCK(inode, *count); | |
1720 | brelse(bitmap_bh); | |
1721 | return 0; | |
1722 | } | |
1723 | ||
617ba13b MC |
1724 | ext4_fsblk_t ext4_new_block(handle_t *handle, struct inode *inode, |
1725 | ext4_fsblk_t goal, int *errp) | |
ac27a0ec DK |
1726 | { |
1727 | unsigned long count = 1; | |
1728 | ||
617ba13b | 1729 | return ext4_new_blocks(handle, inode, goal, &count, errp); |
ac27a0ec DK |
1730 | } |
1731 | ||
1732 | /** | |
617ba13b | 1733 | * ext4_count_free_blocks() -- count filesystem free blocks |
ac27a0ec DK |
1734 | * @sb: superblock |
1735 | * | |
1736 | * Adds up the number of free blocks from each block group. | |
1737 | */ | |
617ba13b | 1738 | ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb) |
ac27a0ec | 1739 | { |
617ba13b MC |
1740 | ext4_fsblk_t desc_count; |
1741 | struct ext4_group_desc *gdp; | |
ac27a0ec | 1742 | int i; |
617ba13b MC |
1743 | unsigned long ngroups = EXT4_SB(sb)->s_groups_count; |
1744 | #ifdef EXT4FS_DEBUG | |
1745 | struct ext4_super_block *es; | |
1746 | ext4_fsblk_t bitmap_count; | |
ac27a0ec DK |
1747 | unsigned long x; |
1748 | struct buffer_head *bitmap_bh = NULL; | |
1749 | ||
617ba13b | 1750 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1751 | desc_count = 0; |
1752 | bitmap_count = 0; | |
1753 | gdp = NULL; | |
1754 | ||
1755 | smp_rmb(); | |
1756 | for (i = 0; i < ngroups; i++) { | |
617ba13b | 1757 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1758 | if (!gdp) |
1759 | continue; | |
1760 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); | |
1761 | brelse(bitmap_bh); | |
1762 | bitmap_bh = read_block_bitmap(sb, i); | |
1763 | if (bitmap_bh == NULL) | |
1764 | continue; | |
1765 | ||
617ba13b | 1766 | x = ext4_count_free(bitmap_bh, sb->s_blocksize); |
ac27a0ec DK |
1767 | printk("group %d: stored = %d, counted = %lu\n", |
1768 | i, le16_to_cpu(gdp->bg_free_blocks_count), x); | |
1769 | bitmap_count += x; | |
1770 | } | |
1771 | brelse(bitmap_bh); | |
2ae02107 MC |
1772 | printk("ext4_count_free_blocks: stored = %llu" |
1773 | ", computed = %llu, %llu\n", | |
bd81d8ee | 1774 | EXT4_FREE_BLOCKS_COUNT(es), |
ac27a0ec DK |
1775 | desc_count, bitmap_count); |
1776 | return bitmap_count; | |
1777 | #else | |
1778 | desc_count = 0; | |
1779 | smp_rmb(); | |
1780 | for (i = 0; i < ngroups; i++) { | |
617ba13b | 1781 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1782 | if (!gdp) |
1783 | continue; | |
1784 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); | |
1785 | } | |
1786 | ||
1787 | return desc_count; | |
1788 | #endif | |
1789 | } | |
1790 | ||
ac27a0ec DK |
1791 | static inline int test_root(int a, int b) |
1792 | { | |
1793 | int num = b; | |
1794 | ||
1795 | while (a > num) | |
1796 | num *= b; | |
1797 | return num == a; | |
1798 | } | |
1799 | ||
617ba13b | 1800 | static int ext4_group_sparse(int group) |
ac27a0ec DK |
1801 | { |
1802 | if (group <= 1) | |
1803 | return 1; | |
1804 | if (!(group & 1)) | |
1805 | return 0; | |
1806 | return (test_root(group, 7) || test_root(group, 5) || | |
1807 | test_root(group, 3)); | |
1808 | } | |
1809 | ||
1810 | /** | |
617ba13b | 1811 | * ext4_bg_has_super - number of blocks used by the superblock in group |
ac27a0ec DK |
1812 | * @sb: superblock for filesystem |
1813 | * @group: group number to check | |
1814 | * | |
1815 | * Return the number of blocks used by the superblock (primary or backup) | |
1816 | * in this group. Currently this will be only 0 or 1. | |
1817 | */ | |
617ba13b | 1818 | int ext4_bg_has_super(struct super_block *sb, int group) |
ac27a0ec | 1819 | { |
617ba13b MC |
1820 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
1821 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) && | |
1822 | !ext4_group_sparse(group)) | |
ac27a0ec DK |
1823 | return 0; |
1824 | return 1; | |
1825 | } | |
1826 | ||
617ba13b | 1827 | static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, int group) |
ac27a0ec | 1828 | { |
617ba13b MC |
1829 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); |
1830 | unsigned long first = metagroup * EXT4_DESC_PER_BLOCK(sb); | |
1831 | unsigned long last = first + EXT4_DESC_PER_BLOCK(sb) - 1; | |
ac27a0ec DK |
1832 | |
1833 | if (group == first || group == first + 1 || group == last) | |
1834 | return 1; | |
1835 | return 0; | |
1836 | } | |
1837 | ||
617ba13b | 1838 | static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, int group) |
ac27a0ec | 1839 | { |
617ba13b MC |
1840 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
1841 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) && | |
1842 | !ext4_group_sparse(group)) | |
ac27a0ec | 1843 | return 0; |
617ba13b | 1844 | return EXT4_SB(sb)->s_gdb_count; |
ac27a0ec DK |
1845 | } |
1846 | ||
1847 | /** | |
617ba13b | 1848 | * ext4_bg_num_gdb - number of blocks used by the group table in group |
ac27a0ec DK |
1849 | * @sb: superblock for filesystem |
1850 | * @group: group number to check | |
1851 | * | |
1852 | * Return the number of blocks used by the group descriptor table | |
1853 | * (primary or backup) in this group. In the future there may be a | |
1854 | * different number of descriptor blocks in each group. | |
1855 | */ | |
617ba13b | 1856 | unsigned long ext4_bg_num_gdb(struct super_block *sb, int group) |
ac27a0ec DK |
1857 | { |
1858 | unsigned long first_meta_bg = | |
617ba13b MC |
1859 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); |
1860 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); | |
ac27a0ec | 1861 | |
617ba13b | 1862 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) || |
ac27a0ec | 1863 | metagroup < first_meta_bg) |
617ba13b | 1864 | return ext4_bg_num_gdb_nometa(sb,group); |
ac27a0ec | 1865 | |
617ba13b | 1866 | return ext4_bg_num_gdb_meta(sb,group); |
ac27a0ec DK |
1867 | |
1868 | } |