]>
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 | ||
27 | /* | |
28 | * The free blocks are managed by bitmaps. A file system contains several | |
29 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
30 | * block for inodes, N blocks for the inode table and data blocks. | |
31 | * | |
32 | * The file system contains group descriptors which are located after the | |
33 | * super block. Each descriptor contains the number of the bitmap block and | |
34 | * the free blocks count in the block. The descriptors are loaded in memory | |
617ba13b | 35 | * when a file system is mounted (see ext4_read_super). |
ac27a0ec DK |
36 | */ |
37 | ||
38 | ||
39 | #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1) | |
40 | ||
41 | /** | |
617ba13b | 42 | * ext4_get_group_desc() -- load group descriptor from disk |
ac27a0ec DK |
43 | * @sb: super block |
44 | * @block_group: given block group | |
45 | * @bh: pointer to the buffer head to store the block | |
46 | * group descriptor | |
47 | */ | |
617ba13b | 48 | struct ext4_group_desc * ext4_get_group_desc(struct super_block * sb, |
ac27a0ec DK |
49 | unsigned int block_group, |
50 | struct buffer_head ** bh) | |
51 | { | |
52 | unsigned long group_desc; | |
53 | unsigned long offset; | |
617ba13b MC |
54 | struct ext4_group_desc * desc; |
55 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
ac27a0ec DK |
56 | |
57 | if (block_group >= sbi->s_groups_count) { | |
617ba13b | 58 | ext4_error (sb, "ext4_get_group_desc", |
ac27a0ec DK |
59 | "block_group >= groups_count - " |
60 | "block_group = %d, groups_count = %lu", | |
61 | block_group, sbi->s_groups_count); | |
62 | ||
63 | return NULL; | |
64 | } | |
65 | smp_rmb(); | |
66 | ||
617ba13b MC |
67 | group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); |
68 | offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); | |
ac27a0ec | 69 | if (!sbi->s_group_desc[group_desc]) { |
617ba13b | 70 | ext4_error (sb, "ext4_get_group_desc", |
ac27a0ec DK |
71 | "Group descriptor not loaded - " |
72 | "block_group = %d, group_desc = %lu, desc = %lu", | |
73 | block_group, group_desc, offset); | |
74 | return NULL; | |
75 | } | |
76 | ||
0d1ee42f AR |
77 | desc = (struct ext4_group_desc *)( |
78 | (__u8 *)sbi->s_group_desc[group_desc]->b_data + | |
79 | offset * EXT4_DESC_SIZE(sb)); | |
ac27a0ec DK |
80 | if (bh) |
81 | *bh = sbi->s_group_desc[group_desc]; | |
0d1ee42f | 82 | return desc; |
ac27a0ec DK |
83 | } |
84 | ||
85 | /** | |
86 | * read_block_bitmap() | |
87 | * @sb: super block | |
88 | * @block_group: given block group | |
89 | * | |
90 | * Read the bitmap for a given block_group, reading into the specified | |
91 | * slot in the superblock's bitmap cache. | |
92 | * | |
93 | * Return buffer_head on success or NULL in case of failure. | |
94 | */ | |
95 | static struct buffer_head * | |
96 | read_block_bitmap(struct super_block *sb, unsigned int block_group) | |
97 | { | |
617ba13b | 98 | struct ext4_group_desc * desc; |
ac27a0ec DK |
99 | struct buffer_head * bh = NULL; |
100 | ||
617ba13b | 101 | desc = ext4_get_group_desc (sb, block_group, NULL); |
ac27a0ec DK |
102 | if (!desc) |
103 | goto error_out; | |
8fadc143 | 104 | bh = sb_bread(sb, ext4_block_bitmap(sb, desc)); |
ac27a0ec | 105 | if (!bh) |
617ba13b | 106 | ext4_error (sb, "read_block_bitmap", |
ac27a0ec | 107 | "Cannot read block bitmap - " |
2ae02107 | 108 | "block_group = %d, block_bitmap = %llu", |
bd81d8ee | 109 | block_group, |
8fadc143 | 110 | ext4_block_bitmap(sb, desc)); |
ac27a0ec DK |
111 | error_out: |
112 | return bh; | |
113 | } | |
114 | /* | |
115 | * The reservation window structure operations | |
116 | * -------------------------------------------- | |
117 | * Operations include: | |
118 | * dump, find, add, remove, is_empty, find_next_reservable_window, etc. | |
119 | * | |
120 | * We use a red-black tree to represent per-filesystem reservation | |
121 | * windows. | |
122 | * | |
123 | */ | |
124 | ||
125 | /** | |
126 | * __rsv_window_dump() -- Dump the filesystem block allocation reservation map | |
127 | * @rb_root: root of per-filesystem reservation rb tree | |
128 | * @verbose: verbose mode | |
129 | * @fn: function which wishes to dump the reservation map | |
130 | * | |
131 | * If verbose is turned on, it will print the whole block reservation | |
132 | * windows(start, end). Otherwise, it will only print out the "bad" windows, | |
133 | * those windows that overlap with their immediate neighbors. | |
134 | */ | |
135 | #if 1 | |
136 | static void __rsv_window_dump(struct rb_root *root, int verbose, | |
137 | const char *fn) | |
138 | { | |
139 | struct rb_node *n; | |
617ba13b | 140 | struct ext4_reserve_window_node *rsv, *prev; |
ac27a0ec DK |
141 | int bad; |
142 | ||
143 | restart: | |
144 | n = rb_first(root); | |
145 | bad = 0; | |
146 | prev = NULL; | |
147 | ||
148 | printk("Block Allocation Reservation Windows Map (%s):\n", fn); | |
149 | while (n) { | |
617ba13b | 150 | rsv = list_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
151 | if (verbose) |
152 | printk("reservation window 0x%p " | |
2ae02107 | 153 | "start: %llu, end: %llu\n", |
ac27a0ec DK |
154 | rsv, rsv->rsv_start, rsv->rsv_end); |
155 | if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) { | |
156 | printk("Bad reservation %p (start >= end)\n", | |
157 | rsv); | |
158 | bad = 1; | |
159 | } | |
160 | if (prev && prev->rsv_end >= rsv->rsv_start) { | |
161 | printk("Bad reservation %p (prev->end >= start)\n", | |
162 | rsv); | |
163 | bad = 1; | |
164 | } | |
165 | if (bad) { | |
166 | if (!verbose) { | |
167 | printk("Restarting reservation walk in verbose mode\n"); | |
168 | verbose = 1; | |
169 | goto restart; | |
170 | } | |
171 | } | |
172 | n = rb_next(n); | |
173 | prev = rsv; | |
174 | } | |
175 | printk("Window map complete.\n"); | |
176 | if (bad) | |
177 | BUG(); | |
178 | } | |
179 | #define rsv_window_dump(root, verbose) \ | |
180 | __rsv_window_dump((root), (verbose), __FUNCTION__) | |
181 | #else | |
182 | #define rsv_window_dump(root, verbose) do {} while (0) | |
183 | #endif | |
184 | ||
185 | /** | |
186 | * goal_in_my_reservation() | |
187 | * @rsv: inode's reservation window | |
188 | * @grp_goal: given goal block relative to the allocation block group | |
189 | * @group: the current allocation block group | |
190 | * @sb: filesystem super block | |
191 | * | |
192 | * Test if the given goal block (group relative) is within the file's | |
193 | * own block reservation window range. | |
194 | * | |
195 | * If the reservation window is outside the goal allocation group, return 0; | |
196 | * grp_goal (given goal block) could be -1, which means no specific | |
197 | * goal block. In this case, always return 1. | |
198 | * If the goal block is within the reservation window, return 1; | |
199 | * otherwise, return 0; | |
200 | */ | |
201 | static int | |
617ba13b | 202 | goal_in_my_reservation(struct ext4_reserve_window *rsv, ext4_grpblk_t grp_goal, |
ac27a0ec DK |
203 | unsigned int group, struct super_block * sb) |
204 | { | |
617ba13b | 205 | ext4_fsblk_t group_first_block, group_last_block; |
ac27a0ec | 206 | |
617ba13b MC |
207 | group_first_block = ext4_group_first_block_no(sb, group); |
208 | group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
209 | |
210 | if ((rsv->_rsv_start > group_last_block) || | |
211 | (rsv->_rsv_end < group_first_block)) | |
212 | return 0; | |
213 | if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start) | |
214 | || (grp_goal + group_first_block > rsv->_rsv_end))) | |
215 | return 0; | |
216 | return 1; | |
217 | } | |
218 | ||
219 | /** | |
220 | * search_reserve_window() | |
221 | * @rb_root: root of reservation tree | |
222 | * @goal: target allocation block | |
223 | * | |
224 | * Find the reserved window which includes the goal, or the previous one | |
225 | * if the goal is not in any window. | |
226 | * Returns NULL if there are no windows or if all windows start after the goal. | |
227 | */ | |
617ba13b MC |
228 | static struct ext4_reserve_window_node * |
229 | search_reserve_window(struct rb_root *root, ext4_fsblk_t goal) | |
ac27a0ec DK |
230 | { |
231 | struct rb_node *n = root->rb_node; | |
617ba13b | 232 | struct ext4_reserve_window_node *rsv; |
ac27a0ec DK |
233 | |
234 | if (!n) | |
235 | return NULL; | |
236 | ||
237 | do { | |
617ba13b | 238 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
239 | |
240 | if (goal < rsv->rsv_start) | |
241 | n = n->rb_left; | |
242 | else if (goal > rsv->rsv_end) | |
243 | n = n->rb_right; | |
244 | else | |
245 | return rsv; | |
246 | } while (n); | |
247 | /* | |
248 | * We've fallen off the end of the tree: the goal wasn't inside | |
249 | * any particular node. OK, the previous node must be to one | |
250 | * side of the interval containing the goal. If it's the RHS, | |
251 | * we need to back up one. | |
252 | */ | |
253 | if (rsv->rsv_start > goal) { | |
254 | n = rb_prev(&rsv->rsv_node); | |
617ba13b | 255 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
256 | } |
257 | return rsv; | |
258 | } | |
259 | ||
260 | /** | |
617ba13b | 261 | * ext4_rsv_window_add() -- Insert a window to the block reservation rb tree. |
ac27a0ec DK |
262 | * @sb: super block |
263 | * @rsv: reservation window to add | |
264 | * | |
265 | * Must be called with rsv_lock hold. | |
266 | */ | |
617ba13b MC |
267 | void ext4_rsv_window_add(struct super_block *sb, |
268 | struct ext4_reserve_window_node *rsv) | |
ac27a0ec | 269 | { |
617ba13b | 270 | struct rb_root *root = &EXT4_SB(sb)->s_rsv_window_root; |
ac27a0ec | 271 | struct rb_node *node = &rsv->rsv_node; |
617ba13b | 272 | ext4_fsblk_t start = rsv->rsv_start; |
ac27a0ec DK |
273 | |
274 | struct rb_node ** p = &root->rb_node; | |
275 | struct rb_node * parent = NULL; | |
617ba13b | 276 | struct ext4_reserve_window_node *this; |
ac27a0ec DK |
277 | |
278 | while (*p) | |
279 | { | |
280 | parent = *p; | |
617ba13b | 281 | this = rb_entry(parent, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
282 | |
283 | if (start < this->rsv_start) | |
284 | p = &(*p)->rb_left; | |
285 | else if (start > this->rsv_end) | |
286 | p = &(*p)->rb_right; | |
287 | else { | |
288 | rsv_window_dump(root, 1); | |
289 | BUG(); | |
290 | } | |
291 | } | |
292 | ||
293 | rb_link_node(node, parent, p); | |
294 | rb_insert_color(node, root); | |
295 | } | |
296 | ||
297 | /** | |
617ba13b | 298 | * ext4_rsv_window_remove() -- unlink a window from the reservation rb tree |
ac27a0ec DK |
299 | * @sb: super block |
300 | * @rsv: reservation window to remove | |
301 | * | |
302 | * Mark the block reservation window as not allocated, and unlink it | |
303 | * from the filesystem reservation window rb tree. Must be called with | |
304 | * rsv_lock hold. | |
305 | */ | |
306 | static void rsv_window_remove(struct super_block *sb, | |
617ba13b | 307 | struct ext4_reserve_window_node *rsv) |
ac27a0ec | 308 | { |
617ba13b MC |
309 | rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
310 | rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; | |
ac27a0ec | 311 | rsv->rsv_alloc_hit = 0; |
617ba13b | 312 | rb_erase(&rsv->rsv_node, &EXT4_SB(sb)->s_rsv_window_root); |
ac27a0ec DK |
313 | } |
314 | ||
315 | /* | |
316 | * rsv_is_empty() -- Check if the reservation window is allocated. | |
317 | * @rsv: given reservation window to check | |
318 | * | |
617ba13b | 319 | * returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED. |
ac27a0ec | 320 | */ |
617ba13b | 321 | static inline int rsv_is_empty(struct ext4_reserve_window *rsv) |
ac27a0ec DK |
322 | { |
323 | /* a valid reservation end block could not be 0 */ | |
617ba13b | 324 | return rsv->_rsv_end == EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
ac27a0ec DK |
325 | } |
326 | ||
327 | /** | |
617ba13b | 328 | * ext4_init_block_alloc_info() |
ac27a0ec DK |
329 | * @inode: file inode structure |
330 | * | |
331 | * Allocate and initialize the reservation window structure, and | |
617ba13b | 332 | * link the window to the ext4 inode structure at last |
ac27a0ec DK |
333 | * |
334 | * The reservation window structure is only dynamically allocated | |
617ba13b MC |
335 | * and linked to ext4 inode the first time the open file |
336 | * needs a new block. So, before every ext4_new_block(s) call, for | |
ac27a0ec DK |
337 | * regular files, we should check whether the reservation window |
338 | * structure exists or not. In the latter case, this function is called. | |
339 | * Fail to do so will result in block reservation being turned off for that | |
340 | * open file. | |
341 | * | |
617ba13b | 342 | * This function is called from ext4_get_blocks_handle(), also called |
ac27a0ec DK |
343 | * when setting the reservation window size through ioctl before the file |
344 | * is open for write (needs block allocation). | |
345 | * | |
346 | * Needs truncate_mutex protection prior to call this function. | |
347 | */ | |
617ba13b | 348 | void ext4_init_block_alloc_info(struct inode *inode) |
ac27a0ec | 349 | { |
617ba13b MC |
350 | struct ext4_inode_info *ei = EXT4_I(inode); |
351 | struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info; | |
ac27a0ec DK |
352 | struct super_block *sb = inode->i_sb; |
353 | ||
354 | block_i = kmalloc(sizeof(*block_i), GFP_NOFS); | |
355 | if (block_i) { | |
617ba13b | 356 | struct ext4_reserve_window_node *rsv = &block_i->rsv_window_node; |
ac27a0ec | 357 | |
617ba13b MC |
358 | rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
359 | rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; | |
ac27a0ec DK |
360 | |
361 | /* | |
362 | * if filesystem is mounted with NORESERVATION, the goal | |
363 | * reservation window size is set to zero to indicate | |
364 | * block reservation is off | |
365 | */ | |
366 | if (!test_opt(sb, RESERVATION)) | |
367 | rsv->rsv_goal_size = 0; | |
368 | else | |
617ba13b | 369 | rsv->rsv_goal_size = EXT4_DEFAULT_RESERVE_BLOCKS; |
ac27a0ec DK |
370 | rsv->rsv_alloc_hit = 0; |
371 | block_i->last_alloc_logical_block = 0; | |
372 | block_i->last_alloc_physical_block = 0; | |
373 | } | |
374 | ei->i_block_alloc_info = block_i; | |
375 | } | |
376 | ||
377 | /** | |
617ba13b | 378 | * ext4_discard_reservation() |
ac27a0ec DK |
379 | * @inode: inode |
380 | * | |
381 | * Discard(free) block reservation window on last file close, or truncate | |
382 | * or at last iput(). | |
383 | * | |
384 | * It is being called in three cases: | |
617ba13b MC |
385 | * ext4_release_file(): last writer close the file |
386 | * ext4_clear_inode(): last iput(), when nobody link to this file. | |
387 | * ext4_truncate(): when the block indirect map is about to change. | |
ac27a0ec DK |
388 | * |
389 | */ | |
617ba13b | 390 | void ext4_discard_reservation(struct inode *inode) |
ac27a0ec | 391 | { |
617ba13b MC |
392 | struct ext4_inode_info *ei = EXT4_I(inode); |
393 | struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info; | |
394 | struct ext4_reserve_window_node *rsv; | |
395 | spinlock_t *rsv_lock = &EXT4_SB(inode->i_sb)->s_rsv_window_lock; | |
ac27a0ec DK |
396 | |
397 | if (!block_i) | |
398 | return; | |
399 | ||
400 | rsv = &block_i->rsv_window_node; | |
401 | if (!rsv_is_empty(&rsv->rsv_window)) { | |
402 | spin_lock(rsv_lock); | |
403 | if (!rsv_is_empty(&rsv->rsv_window)) | |
404 | rsv_window_remove(inode->i_sb, rsv); | |
405 | spin_unlock(rsv_lock); | |
406 | } | |
407 | } | |
408 | ||
409 | /** | |
617ba13b | 410 | * ext4_free_blocks_sb() -- Free given blocks and update quota |
ac27a0ec DK |
411 | * @handle: handle to this transaction |
412 | * @sb: super block | |
413 | * @block: start physcial block to free | |
414 | * @count: number of blocks to free | |
415 | * @pdquot_freed_blocks: pointer to quota | |
416 | */ | |
617ba13b MC |
417 | void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb, |
418 | ext4_fsblk_t block, unsigned long count, | |
ac27a0ec DK |
419 | unsigned long *pdquot_freed_blocks) |
420 | { | |
421 | struct buffer_head *bitmap_bh = NULL; | |
422 | struct buffer_head *gd_bh; | |
423 | unsigned long block_group; | |
617ba13b | 424 | ext4_grpblk_t bit; |
ac27a0ec DK |
425 | unsigned long i; |
426 | unsigned long overflow; | |
617ba13b MC |
427 | struct ext4_group_desc * desc; |
428 | struct ext4_super_block * es; | |
429 | struct ext4_sb_info *sbi; | |
ac27a0ec | 430 | int err = 0, ret; |
617ba13b | 431 | ext4_grpblk_t group_freed; |
ac27a0ec DK |
432 | |
433 | *pdquot_freed_blocks = 0; | |
617ba13b | 434 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
435 | es = sbi->s_es; |
436 | if (block < le32_to_cpu(es->s_first_data_block) || | |
437 | block + count < block || | |
bd81d8ee | 438 | block + count > ext4_blocks_count(es)) { |
617ba13b | 439 | ext4_error (sb, "ext4_free_blocks", |
ac27a0ec | 440 | "Freeing blocks not in datazone - " |
2ae02107 | 441 | "block = %llu, count = %lu", block, count); |
ac27a0ec DK |
442 | goto error_return; |
443 | } | |
444 | ||
bd81d8ee | 445 | ext4_debug ("freeing block(s) %llu-%llu\n", block, block + count - 1); |
ac27a0ec DK |
446 | |
447 | do_more: | |
448 | overflow = 0; | |
3a5b2ecd | 449 | ext4_get_group_no_and_offset(sb, block, &block_group, &bit); |
ac27a0ec DK |
450 | /* |
451 | * Check to see if we are freeing blocks across a group | |
452 | * boundary. | |
453 | */ | |
617ba13b MC |
454 | if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) { |
455 | overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb); | |
ac27a0ec DK |
456 | count -= overflow; |
457 | } | |
458 | brelse(bitmap_bh); | |
459 | bitmap_bh = read_block_bitmap(sb, block_group); | |
460 | if (!bitmap_bh) | |
461 | goto error_return; | |
617ba13b | 462 | desc = ext4_get_group_desc (sb, block_group, &gd_bh); |
ac27a0ec DK |
463 | if (!desc) |
464 | goto error_return; | |
465 | ||
8fadc143 AR |
466 | if (in_range(ext4_block_bitmap(sb, desc), block, count) || |
467 | in_range(ext4_inode_bitmap(sb, desc), block, count) || | |
468 | in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) || | |
469 | in_range(block + count - 1, ext4_inode_table(sb, desc), | |
bd81d8ee | 470 | sbi->s_itb_per_group)) |
617ba13b | 471 | ext4_error (sb, "ext4_free_blocks", |
ac27a0ec | 472 | "Freeing blocks in system zones - " |
2ae02107 | 473 | "Block = %llu, count = %lu", |
ac27a0ec DK |
474 | block, count); |
475 | ||
476 | /* | |
477 | * We are about to start releasing blocks in the bitmap, | |
478 | * so we need undo access. | |
479 | */ | |
480 | /* @@@ check errors */ | |
481 | BUFFER_TRACE(bitmap_bh, "getting undo access"); | |
617ba13b | 482 | err = ext4_journal_get_undo_access(handle, bitmap_bh); |
ac27a0ec DK |
483 | if (err) |
484 | goto error_return; | |
485 | ||
486 | /* | |
487 | * We are about to modify some metadata. Call the journal APIs | |
488 | * to unshare ->b_data if a currently-committing transaction is | |
489 | * using it | |
490 | */ | |
491 | BUFFER_TRACE(gd_bh, "get_write_access"); | |
617ba13b | 492 | err = ext4_journal_get_write_access(handle, gd_bh); |
ac27a0ec DK |
493 | if (err) |
494 | goto error_return; | |
495 | ||
496 | jbd_lock_bh_state(bitmap_bh); | |
497 | ||
498 | for (i = 0, group_freed = 0; i < count; i++) { | |
499 | /* | |
500 | * An HJ special. This is expensive... | |
501 | */ | |
502 | #ifdef CONFIG_JBD_DEBUG | |
503 | jbd_unlock_bh_state(bitmap_bh); | |
504 | { | |
505 | struct buffer_head *debug_bh; | |
506 | debug_bh = sb_find_get_block(sb, block + i); | |
507 | if (debug_bh) { | |
508 | BUFFER_TRACE(debug_bh, "Deleted!"); | |
509 | if (!bh2jh(bitmap_bh)->b_committed_data) | |
510 | BUFFER_TRACE(debug_bh, | |
511 | "No commited data in bitmap"); | |
512 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap"); | |
513 | __brelse(debug_bh); | |
514 | } | |
515 | } | |
516 | jbd_lock_bh_state(bitmap_bh); | |
517 | #endif | |
518 | if (need_resched()) { | |
519 | jbd_unlock_bh_state(bitmap_bh); | |
520 | cond_resched(); | |
521 | jbd_lock_bh_state(bitmap_bh); | |
522 | } | |
523 | /* @@@ This prevents newly-allocated data from being | |
524 | * freed and then reallocated within the same | |
525 | * transaction. | |
526 | * | |
527 | * Ideally we would want to allow that to happen, but to | |
dab291af | 528 | * do so requires making jbd2_journal_forget() capable of |
ac27a0ec DK |
529 | * revoking the queued write of a data block, which |
530 | * implies blocking on the journal lock. *forget() | |
531 | * cannot block due to truncate races. | |
532 | * | |
dab291af | 533 | * Eventually we can fix this by making jbd2_journal_forget() |
ac27a0ec DK |
534 | * return a status indicating whether or not it was able |
535 | * to revoke the buffer. On successful revoke, it is | |
536 | * safe not to set the allocation bit in the committed | |
537 | * bitmap, because we know that there is no outstanding | |
538 | * activity on the buffer any more and so it is safe to | |
539 | * reallocate it. | |
540 | */ | |
541 | BUFFER_TRACE(bitmap_bh, "set in b_committed_data"); | |
542 | J_ASSERT_BH(bitmap_bh, | |
543 | bh2jh(bitmap_bh)->b_committed_data != NULL); | |
617ba13b | 544 | ext4_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i, |
ac27a0ec DK |
545 | bh2jh(bitmap_bh)->b_committed_data); |
546 | ||
547 | /* | |
548 | * We clear the bit in the bitmap after setting the committed | |
549 | * data bit, because this is the reverse order to that which | |
550 | * the allocator uses. | |
551 | */ | |
552 | BUFFER_TRACE(bitmap_bh, "clear bit"); | |
617ba13b | 553 | if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group), |
ac27a0ec DK |
554 | bit + i, bitmap_bh->b_data)) { |
555 | jbd_unlock_bh_state(bitmap_bh); | |
617ba13b | 556 | ext4_error(sb, __FUNCTION__, |
2ae02107 | 557 | "bit already cleared for block %llu", |
bd81d8ee | 558 | (ext4_fsblk_t)(block + i)); |
ac27a0ec DK |
559 | jbd_lock_bh_state(bitmap_bh); |
560 | BUFFER_TRACE(bitmap_bh, "bit already cleared"); | |
561 | } else { | |
562 | group_freed++; | |
563 | } | |
564 | } | |
565 | jbd_unlock_bh_state(bitmap_bh); | |
566 | ||
567 | spin_lock(sb_bgl_lock(sbi, block_group)); | |
568 | desc->bg_free_blocks_count = | |
569 | cpu_to_le16(le16_to_cpu(desc->bg_free_blocks_count) + | |
570 | group_freed); | |
571 | spin_unlock(sb_bgl_lock(sbi, block_group)); | |
572 | percpu_counter_mod(&sbi->s_freeblocks_counter, count); | |
573 | ||
574 | /* We dirtied the bitmap block */ | |
575 | BUFFER_TRACE(bitmap_bh, "dirtied bitmap block"); | |
617ba13b | 576 | err = ext4_journal_dirty_metadata(handle, bitmap_bh); |
ac27a0ec DK |
577 | |
578 | /* And the group descriptor block */ | |
579 | BUFFER_TRACE(gd_bh, "dirtied group descriptor block"); | |
617ba13b | 580 | ret = ext4_journal_dirty_metadata(handle, gd_bh); |
ac27a0ec DK |
581 | if (!err) err = ret; |
582 | *pdquot_freed_blocks += group_freed; | |
583 | ||
584 | if (overflow && !err) { | |
585 | block += count; | |
586 | count = overflow; | |
587 | goto do_more; | |
588 | } | |
589 | sb->s_dirt = 1; | |
590 | error_return: | |
591 | brelse(bitmap_bh); | |
617ba13b | 592 | ext4_std_error(sb, err); |
ac27a0ec DK |
593 | return; |
594 | } | |
595 | ||
596 | /** | |
617ba13b | 597 | * ext4_free_blocks() -- Free given blocks and update quota |
ac27a0ec DK |
598 | * @handle: handle for this transaction |
599 | * @inode: inode | |
600 | * @block: start physical block to free | |
601 | * @count: number of blocks to count | |
602 | */ | |
617ba13b MC |
603 | void ext4_free_blocks(handle_t *handle, struct inode *inode, |
604 | ext4_fsblk_t block, unsigned long count) | |
ac27a0ec DK |
605 | { |
606 | struct super_block * sb; | |
607 | unsigned long dquot_freed_blocks; | |
608 | ||
609 | sb = inode->i_sb; | |
610 | if (!sb) { | |
617ba13b | 611 | printk ("ext4_free_blocks: nonexistent device"); |
ac27a0ec DK |
612 | return; |
613 | } | |
617ba13b | 614 | ext4_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks); |
ac27a0ec DK |
615 | if (dquot_freed_blocks) |
616 | DQUOT_FREE_BLOCK(inode, dquot_freed_blocks); | |
617 | return; | |
618 | } | |
619 | ||
620 | /** | |
617ba13b | 621 | * ext4_test_allocatable() |
ac27a0ec DK |
622 | * @nr: given allocation block group |
623 | * @bh: bufferhead contains the bitmap of the given block group | |
624 | * | |
617ba13b | 625 | * For ext4 allocations, we must not reuse any blocks which are |
ac27a0ec DK |
626 | * allocated in the bitmap buffer's "last committed data" copy. This |
627 | * prevents deletes from freeing up the page for reuse until we have | |
628 | * committed the delete transaction. | |
629 | * | |
630 | * If we didn't do this, then deleting something and reallocating it as | |
631 | * data would allow the old block to be overwritten before the | |
632 | * transaction committed (because we force data to disk before commit). | |
633 | * This would lead to corruption if we crashed between overwriting the | |
634 | * data and committing the delete. | |
635 | * | |
636 | * @@@ We may want to make this allocation behaviour conditional on | |
637 | * data-writes at some point, and disable it for metadata allocations or | |
638 | * sync-data inodes. | |
639 | */ | |
617ba13b | 640 | static int ext4_test_allocatable(ext4_grpblk_t nr, struct buffer_head *bh) |
ac27a0ec DK |
641 | { |
642 | int ret; | |
643 | struct journal_head *jh = bh2jh(bh); | |
644 | ||
617ba13b | 645 | if (ext4_test_bit(nr, bh->b_data)) |
ac27a0ec DK |
646 | return 0; |
647 | ||
648 | jbd_lock_bh_state(bh); | |
649 | if (!jh->b_committed_data) | |
650 | ret = 1; | |
651 | else | |
617ba13b | 652 | ret = !ext4_test_bit(nr, jh->b_committed_data); |
ac27a0ec DK |
653 | jbd_unlock_bh_state(bh); |
654 | return ret; | |
655 | } | |
656 | ||
657 | /** | |
658 | * bitmap_search_next_usable_block() | |
659 | * @start: the starting block (group relative) of the search | |
660 | * @bh: bufferhead contains the block group bitmap | |
661 | * @maxblocks: the ending block (group relative) of the reservation | |
662 | * | |
663 | * The bitmap search --- search forward alternately through the actual | |
664 | * bitmap on disk and the last-committed copy in journal, until we find a | |
665 | * bit free in both bitmaps. | |
666 | */ | |
617ba13b MC |
667 | static ext4_grpblk_t |
668 | bitmap_search_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh, | |
669 | ext4_grpblk_t maxblocks) | |
ac27a0ec | 670 | { |
617ba13b | 671 | ext4_grpblk_t next; |
ac27a0ec DK |
672 | struct journal_head *jh = bh2jh(bh); |
673 | ||
674 | while (start < maxblocks) { | |
617ba13b | 675 | next = ext4_find_next_zero_bit(bh->b_data, maxblocks, start); |
ac27a0ec DK |
676 | if (next >= maxblocks) |
677 | return -1; | |
617ba13b | 678 | if (ext4_test_allocatable(next, bh)) |
ac27a0ec DK |
679 | return next; |
680 | jbd_lock_bh_state(bh); | |
681 | if (jh->b_committed_data) | |
617ba13b | 682 | start = ext4_find_next_zero_bit(jh->b_committed_data, |
ac27a0ec DK |
683 | maxblocks, next); |
684 | jbd_unlock_bh_state(bh); | |
685 | } | |
686 | return -1; | |
687 | } | |
688 | ||
689 | /** | |
690 | * find_next_usable_block() | |
691 | * @start: the starting block (group relative) to find next | |
692 | * allocatable block in bitmap. | |
693 | * @bh: bufferhead contains the block group bitmap | |
694 | * @maxblocks: the ending block (group relative) for the search | |
695 | * | |
696 | * Find an allocatable block in a bitmap. We honor both the bitmap and | |
697 | * its last-committed copy (if that exists), and perform the "most | |
698 | * appropriate allocation" algorithm of looking for a free block near | |
699 | * the initial goal; then for a free byte somewhere in the bitmap; then | |
700 | * for any free bit in the bitmap. | |
701 | */ | |
617ba13b MC |
702 | static ext4_grpblk_t |
703 | find_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh, | |
704 | ext4_grpblk_t maxblocks) | |
ac27a0ec | 705 | { |
617ba13b | 706 | ext4_grpblk_t here, next; |
ac27a0ec DK |
707 | char *p, *r; |
708 | ||
709 | if (start > 0) { | |
710 | /* | |
711 | * The goal was occupied; search forward for a free | |
712 | * block within the next XX blocks. | |
713 | * | |
714 | * end_goal is more or less random, but it has to be | |
617ba13b | 715 | * less than EXT4_BLOCKS_PER_GROUP. Aligning up to the |
ac27a0ec DK |
716 | * next 64-bit boundary is simple.. |
717 | */ | |
617ba13b | 718 | ext4_grpblk_t end_goal = (start + 63) & ~63; |
ac27a0ec DK |
719 | if (end_goal > maxblocks) |
720 | end_goal = maxblocks; | |
617ba13b MC |
721 | here = ext4_find_next_zero_bit(bh->b_data, end_goal, start); |
722 | if (here < end_goal && ext4_test_allocatable(here, bh)) | |
ac27a0ec | 723 | return here; |
617ba13b | 724 | ext4_debug("Bit not found near goal\n"); |
ac27a0ec DK |
725 | } |
726 | ||
727 | here = start; | |
728 | if (here < 0) | |
729 | here = 0; | |
730 | ||
731 | p = ((char *)bh->b_data) + (here >> 3); | |
732 | r = memscan(p, 0, (maxblocks - here + 7) >> 3); | |
733 | next = (r - ((char *)bh->b_data)) << 3; | |
734 | ||
617ba13b | 735 | if (next < maxblocks && next >= start && ext4_test_allocatable(next, bh)) |
ac27a0ec DK |
736 | return next; |
737 | ||
738 | /* | |
739 | * The bitmap search --- search forward alternately through the actual | |
740 | * bitmap and the last-committed copy until we find a bit free in | |
741 | * both | |
742 | */ | |
743 | here = bitmap_search_next_usable_block(here, bh, maxblocks); | |
744 | return here; | |
745 | } | |
746 | ||
747 | /** | |
748 | * claim_block() | |
749 | * @block: the free block (group relative) to allocate | |
750 | * @bh: the bufferhead containts the block group bitmap | |
751 | * | |
752 | * We think we can allocate this block in this bitmap. Try to set the bit. | |
753 | * If that succeeds then check that nobody has allocated and then freed the | |
754 | * block since we saw that is was not marked in b_committed_data. If it _was_ | |
755 | * allocated and freed then clear the bit in the bitmap again and return | |
756 | * zero (failure). | |
757 | */ | |
758 | static inline int | |
617ba13b | 759 | claim_block(spinlock_t *lock, ext4_grpblk_t block, struct buffer_head *bh) |
ac27a0ec DK |
760 | { |
761 | struct journal_head *jh = bh2jh(bh); | |
762 | int ret; | |
763 | ||
617ba13b | 764 | if (ext4_set_bit_atomic(lock, block, bh->b_data)) |
ac27a0ec DK |
765 | return 0; |
766 | jbd_lock_bh_state(bh); | |
617ba13b MC |
767 | if (jh->b_committed_data && ext4_test_bit(block,jh->b_committed_data)) { |
768 | ext4_clear_bit_atomic(lock, block, bh->b_data); | |
ac27a0ec DK |
769 | ret = 0; |
770 | } else { | |
771 | ret = 1; | |
772 | } | |
773 | jbd_unlock_bh_state(bh); | |
774 | return ret; | |
775 | } | |
776 | ||
777 | /** | |
617ba13b | 778 | * ext4_try_to_allocate() |
ac27a0ec DK |
779 | * @sb: superblock |
780 | * @handle: handle to this transaction | |
781 | * @group: given allocation block group | |
782 | * @bitmap_bh: bufferhead holds the block bitmap | |
783 | * @grp_goal: given target block within the group | |
784 | * @count: target number of blocks to allocate | |
785 | * @my_rsv: reservation window | |
786 | * | |
787 | * Attempt to allocate blocks within a give range. Set the range of allocation | |
788 | * first, then find the first free bit(s) from the bitmap (within the range), | |
789 | * and at last, allocate the blocks by claiming the found free bit as allocated. | |
790 | * | |
791 | * To set the range of this allocation: | |
792 | * if there is a reservation window, only try to allocate block(s) from the | |
793 | * file's own reservation window; | |
794 | * Otherwise, the allocation range starts from the give goal block, ends at | |
795 | * the block group's last block. | |
796 | * | |
797 | * If we failed to allocate the desired block then we may end up crossing to a | |
798 | * new bitmap. In that case we must release write access to the old one via | |
617ba13b | 799 | * ext4_journal_release_buffer(), else we'll run out of credits. |
ac27a0ec | 800 | */ |
617ba13b MC |
801 | static ext4_grpblk_t |
802 | ext4_try_to_allocate(struct super_block *sb, handle_t *handle, int group, | |
803 | struct buffer_head *bitmap_bh, ext4_grpblk_t grp_goal, | |
804 | unsigned long *count, struct ext4_reserve_window *my_rsv) | |
ac27a0ec | 805 | { |
617ba13b MC |
806 | ext4_fsblk_t group_first_block; |
807 | ext4_grpblk_t start, end; | |
ac27a0ec DK |
808 | unsigned long num = 0; |
809 | ||
810 | /* we do allocation within the reservation window if we have a window */ | |
811 | if (my_rsv) { | |
617ba13b | 812 | group_first_block = ext4_group_first_block_no(sb, group); |
ac27a0ec DK |
813 | if (my_rsv->_rsv_start >= group_first_block) |
814 | start = my_rsv->_rsv_start - group_first_block; | |
815 | else | |
816 | /* reservation window cross group boundary */ | |
817 | start = 0; | |
818 | end = my_rsv->_rsv_end - group_first_block + 1; | |
617ba13b | 819 | if (end > EXT4_BLOCKS_PER_GROUP(sb)) |
ac27a0ec | 820 | /* reservation window crosses group boundary */ |
617ba13b | 821 | end = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
822 | if ((start <= grp_goal) && (grp_goal < end)) |
823 | start = grp_goal; | |
824 | else | |
825 | grp_goal = -1; | |
826 | } else { | |
827 | if (grp_goal > 0) | |
828 | start = grp_goal; | |
829 | else | |
830 | start = 0; | |
617ba13b | 831 | end = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
832 | } |
833 | ||
617ba13b | 834 | BUG_ON(start > EXT4_BLOCKS_PER_GROUP(sb)); |
ac27a0ec DK |
835 | |
836 | repeat: | |
617ba13b | 837 | if (grp_goal < 0 || !ext4_test_allocatable(grp_goal, bitmap_bh)) { |
ac27a0ec DK |
838 | grp_goal = find_next_usable_block(start, bitmap_bh, end); |
839 | if (grp_goal < 0) | |
840 | goto fail_access; | |
841 | if (!my_rsv) { | |
842 | int i; | |
843 | ||
844 | for (i = 0; i < 7 && grp_goal > start && | |
617ba13b | 845 | ext4_test_allocatable(grp_goal - 1, |
ac27a0ec DK |
846 | bitmap_bh); |
847 | i++, grp_goal--) | |
848 | ; | |
849 | } | |
850 | } | |
851 | start = grp_goal; | |
852 | ||
617ba13b | 853 | if (!claim_block(sb_bgl_lock(EXT4_SB(sb), group), |
ac27a0ec DK |
854 | grp_goal, bitmap_bh)) { |
855 | /* | |
856 | * The block was allocated by another thread, or it was | |
857 | * allocated and then freed by another thread | |
858 | */ | |
859 | start++; | |
860 | grp_goal++; | |
861 | if (start >= end) | |
862 | goto fail_access; | |
863 | goto repeat; | |
864 | } | |
865 | num++; | |
866 | grp_goal++; | |
867 | while (num < *count && grp_goal < end | |
617ba13b MC |
868 | && ext4_test_allocatable(grp_goal, bitmap_bh) |
869 | && claim_block(sb_bgl_lock(EXT4_SB(sb), group), | |
ac27a0ec DK |
870 | grp_goal, bitmap_bh)) { |
871 | num++; | |
872 | grp_goal++; | |
873 | } | |
874 | *count = num; | |
875 | return grp_goal - num; | |
876 | fail_access: | |
877 | *count = num; | |
878 | return -1; | |
879 | } | |
880 | ||
881 | /** | |
882 | * find_next_reservable_window(): | |
883 | * find a reservable space within the given range. | |
884 | * It does not allocate the reservation window for now: | |
885 | * alloc_new_reservation() will do the work later. | |
886 | * | |
887 | * @search_head: the head of the searching list; | |
888 | * This is not necessarily the list head of the whole filesystem | |
889 | * | |
890 | * We have both head and start_block to assist the search | |
891 | * for the reservable space. The list starts from head, | |
892 | * but we will shift to the place where start_block is, | |
893 | * then start from there, when looking for a reservable space. | |
894 | * | |
895 | * @size: the target new reservation window size | |
896 | * | |
897 | * @group_first_block: the first block we consider to start | |
898 | * the real search from | |
899 | * | |
900 | * @last_block: | |
901 | * the maximum block number that our goal reservable space | |
902 | * could start from. This is normally the last block in this | |
903 | * group. The search will end when we found the start of next | |
904 | * possible reservable space is out of this boundary. | |
905 | * This could handle the cross boundary reservation window | |
906 | * request. | |
907 | * | |
908 | * basically we search from the given range, rather than the whole | |
909 | * reservation double linked list, (start_block, last_block) | |
910 | * to find a free region that is of my size and has not | |
911 | * been reserved. | |
912 | * | |
913 | */ | |
914 | static int find_next_reservable_window( | |
617ba13b MC |
915 | struct ext4_reserve_window_node *search_head, |
916 | struct ext4_reserve_window_node *my_rsv, | |
ac27a0ec | 917 | struct super_block * sb, |
617ba13b MC |
918 | ext4_fsblk_t start_block, |
919 | ext4_fsblk_t last_block) | |
ac27a0ec DK |
920 | { |
921 | struct rb_node *next; | |
617ba13b MC |
922 | struct ext4_reserve_window_node *rsv, *prev; |
923 | ext4_fsblk_t cur; | |
ac27a0ec DK |
924 | int size = my_rsv->rsv_goal_size; |
925 | ||
926 | /* TODO: make the start of the reservation window byte-aligned */ | |
927 | /* cur = *start_block & ~7;*/ | |
928 | cur = start_block; | |
929 | rsv = search_head; | |
930 | if (!rsv) | |
931 | return -1; | |
932 | ||
933 | while (1) { | |
934 | if (cur <= rsv->rsv_end) | |
935 | cur = rsv->rsv_end + 1; | |
936 | ||
937 | /* TODO? | |
938 | * in the case we could not find a reservable space | |
939 | * that is what is expected, during the re-search, we could | |
940 | * remember what's the largest reservable space we could have | |
941 | * and return that one. | |
942 | * | |
943 | * For now it will fail if we could not find the reservable | |
944 | * space with expected-size (or more)... | |
945 | */ | |
946 | if (cur > last_block) | |
947 | return -1; /* fail */ | |
948 | ||
949 | prev = rsv; | |
950 | next = rb_next(&rsv->rsv_node); | |
617ba13b | 951 | rsv = list_entry(next,struct ext4_reserve_window_node,rsv_node); |
ac27a0ec DK |
952 | |
953 | /* | |
954 | * Reached the last reservation, we can just append to the | |
955 | * previous one. | |
956 | */ | |
957 | if (!next) | |
958 | break; | |
959 | ||
960 | if (cur + size <= rsv->rsv_start) { | |
961 | /* | |
962 | * Found a reserveable space big enough. We could | |
963 | * have a reservation across the group boundary here | |
964 | */ | |
965 | break; | |
966 | } | |
967 | } | |
968 | /* | |
969 | * we come here either : | |
970 | * when we reach the end of the whole list, | |
971 | * and there is empty reservable space after last entry in the list. | |
972 | * append it to the end of the list. | |
973 | * | |
974 | * or we found one reservable space in the middle of the list, | |
975 | * return the reservation window that we could append to. | |
976 | * succeed. | |
977 | */ | |
978 | ||
979 | if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window))) | |
980 | rsv_window_remove(sb, my_rsv); | |
981 | ||
982 | /* | |
983 | * Let's book the whole avaliable window for now. We will check the | |
984 | * disk bitmap later and then, if there are free blocks then we adjust | |
985 | * the window size if it's larger than requested. | |
986 | * Otherwise, we will remove this node from the tree next time | |
987 | * call find_next_reservable_window. | |
988 | */ | |
989 | my_rsv->rsv_start = cur; | |
990 | my_rsv->rsv_end = cur + size - 1; | |
991 | my_rsv->rsv_alloc_hit = 0; | |
992 | ||
993 | if (prev != my_rsv) | |
617ba13b | 994 | ext4_rsv_window_add(sb, my_rsv); |
ac27a0ec DK |
995 | |
996 | return 0; | |
997 | } | |
998 | ||
999 | /** | |
1000 | * alloc_new_reservation()--allocate a new reservation window | |
1001 | * | |
1002 | * To make a new reservation, we search part of the filesystem | |
1003 | * reservation list (the list that inside the group). We try to | |
1004 | * allocate a new reservation window near the allocation goal, | |
1005 | * or the beginning of the group, if there is no goal. | |
1006 | * | |
1007 | * We first find a reservable space after the goal, then from | |
1008 | * there, we check the bitmap for the first free block after | |
1009 | * it. If there is no free block until the end of group, then the | |
1010 | * whole group is full, we failed. Otherwise, check if the free | |
1011 | * block is inside the expected reservable space, if so, we | |
1012 | * succeed. | |
1013 | * If the first free block is outside the reservable space, then | |
1014 | * start from the first free block, we search for next available | |
1015 | * space, and go on. | |
1016 | * | |
1017 | * on succeed, a new reservation will be found and inserted into the list | |
1018 | * It contains at least one free block, and it does not overlap with other | |
1019 | * reservation windows. | |
1020 | * | |
1021 | * failed: we failed to find a reservation window in this group | |
1022 | * | |
1023 | * @rsv: the reservation | |
1024 | * | |
1025 | * @grp_goal: The goal (group-relative). It is where the search for a | |
1026 | * free reservable space should start from. | |
1027 | * if we have a grp_goal(grp_goal >0 ), then start from there, | |
1028 | * no grp_goal(grp_goal = -1), we start from the first block | |
1029 | * of the group. | |
1030 | * | |
1031 | * @sb: the super block | |
1032 | * @group: the group we are trying to allocate in | |
1033 | * @bitmap_bh: the block group block bitmap | |
1034 | * | |
1035 | */ | |
617ba13b MC |
1036 | static int alloc_new_reservation(struct ext4_reserve_window_node *my_rsv, |
1037 | ext4_grpblk_t grp_goal, struct super_block *sb, | |
ac27a0ec DK |
1038 | unsigned int group, struct buffer_head *bitmap_bh) |
1039 | { | |
617ba13b MC |
1040 | struct ext4_reserve_window_node *search_head; |
1041 | ext4_fsblk_t group_first_block, group_end_block, start_block; | |
1042 | ext4_grpblk_t first_free_block; | |
1043 | struct rb_root *fs_rsv_root = &EXT4_SB(sb)->s_rsv_window_root; | |
ac27a0ec DK |
1044 | unsigned long size; |
1045 | int ret; | |
617ba13b | 1046 | spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock; |
ac27a0ec | 1047 | |
617ba13b MC |
1048 | group_first_block = ext4_group_first_block_no(sb, group); |
1049 | group_end_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
1050 | |
1051 | if (grp_goal < 0) | |
1052 | start_block = group_first_block; | |
1053 | else | |
1054 | start_block = grp_goal + group_first_block; | |
1055 | ||
1056 | size = my_rsv->rsv_goal_size; | |
1057 | ||
1058 | if (!rsv_is_empty(&my_rsv->rsv_window)) { | |
1059 | /* | |
1060 | * if the old reservation is cross group boundary | |
1061 | * and if the goal is inside the old reservation window, | |
1062 | * we will come here when we just failed to allocate from | |
1063 | * the first part of the window. We still have another part | |
1064 | * that belongs to the next group. In this case, there is no | |
1065 | * point to discard our window and try to allocate a new one | |
1066 | * in this group(which will fail). we should | |
1067 | * keep the reservation window, just simply move on. | |
1068 | * | |
1069 | * Maybe we could shift the start block of the reservation | |
1070 | * window to the first block of next group. | |
1071 | */ | |
1072 | ||
1073 | if ((my_rsv->rsv_start <= group_end_block) && | |
1074 | (my_rsv->rsv_end > group_end_block) && | |
1075 | (start_block >= my_rsv->rsv_start)) | |
1076 | return -1; | |
1077 | ||
1078 | if ((my_rsv->rsv_alloc_hit > | |
1079 | (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) { | |
1080 | /* | |
1081 | * if the previously allocation hit ratio is | |
1082 | * greater than 1/2, then we double the size of | |
1083 | * the reservation window the next time, | |
1084 | * otherwise we keep the same size window | |
1085 | */ | |
1086 | size = size * 2; | |
617ba13b MC |
1087 | if (size > EXT4_MAX_RESERVE_BLOCKS) |
1088 | size = EXT4_MAX_RESERVE_BLOCKS; | |
ac27a0ec DK |
1089 | my_rsv->rsv_goal_size= size; |
1090 | } | |
1091 | } | |
1092 | ||
1093 | spin_lock(rsv_lock); | |
1094 | /* | |
1095 | * shift the search start to the window near the goal block | |
1096 | */ | |
1097 | search_head = search_reserve_window(fs_rsv_root, start_block); | |
1098 | ||
1099 | /* | |
1100 | * find_next_reservable_window() simply finds a reservable window | |
1101 | * inside the given range(start_block, group_end_block). | |
1102 | * | |
1103 | * To make sure the reservation window has a free bit inside it, we | |
1104 | * need to check the bitmap after we found a reservable window. | |
1105 | */ | |
1106 | retry: | |
1107 | ret = find_next_reservable_window(search_head, my_rsv, sb, | |
1108 | start_block, group_end_block); | |
1109 | ||
1110 | if (ret == -1) { | |
1111 | if (!rsv_is_empty(&my_rsv->rsv_window)) | |
1112 | rsv_window_remove(sb, my_rsv); | |
1113 | spin_unlock(rsv_lock); | |
1114 | return -1; | |
1115 | } | |
1116 | ||
1117 | /* | |
1118 | * On success, find_next_reservable_window() returns the | |
1119 | * reservation window where there is a reservable space after it. | |
1120 | * Before we reserve this reservable space, we need | |
1121 | * to make sure there is at least a free block inside this region. | |
1122 | * | |
1123 | * searching the first free bit on the block bitmap and copy of | |
1124 | * last committed bitmap alternatively, until we found a allocatable | |
1125 | * block. Search start from the start block of the reservable space | |
1126 | * we just found. | |
1127 | */ | |
1128 | spin_unlock(rsv_lock); | |
1129 | first_free_block = bitmap_search_next_usable_block( | |
1130 | my_rsv->rsv_start - group_first_block, | |
1131 | bitmap_bh, group_end_block - group_first_block + 1); | |
1132 | ||
1133 | if (first_free_block < 0) { | |
1134 | /* | |
1135 | * no free block left on the bitmap, no point | |
1136 | * to reserve the space. return failed. | |
1137 | */ | |
1138 | spin_lock(rsv_lock); | |
1139 | if (!rsv_is_empty(&my_rsv->rsv_window)) | |
1140 | rsv_window_remove(sb, my_rsv); | |
1141 | spin_unlock(rsv_lock); | |
1142 | return -1; /* failed */ | |
1143 | } | |
1144 | ||
1145 | start_block = first_free_block + group_first_block; | |
1146 | /* | |
1147 | * check if the first free block is within the | |
1148 | * free space we just reserved | |
1149 | */ | |
1150 | if (start_block >= my_rsv->rsv_start && start_block < my_rsv->rsv_end) | |
1151 | return 0; /* success */ | |
1152 | /* | |
1153 | * if the first free bit we found is out of the reservable space | |
1154 | * continue search for next reservable space, | |
1155 | * start from where the free block is, | |
1156 | * we also shift the list head to where we stopped last time | |
1157 | */ | |
1158 | search_head = my_rsv; | |
1159 | spin_lock(rsv_lock); | |
1160 | goto retry; | |
1161 | } | |
1162 | ||
1163 | /** | |
1164 | * try_to_extend_reservation() | |
1165 | * @my_rsv: given reservation window | |
1166 | * @sb: super block | |
1167 | * @size: the delta to extend | |
1168 | * | |
1169 | * Attempt to expand the reservation window large enough to have | |
1170 | * required number of free blocks | |
1171 | * | |
617ba13b | 1172 | * Since ext4_try_to_allocate() will always allocate blocks within |
ac27a0ec DK |
1173 | * the reservation window range, if the window size is too small, |
1174 | * multiple blocks allocation has to stop at the end of the reservation | |
1175 | * window. To make this more efficient, given the total number of | |
1176 | * blocks needed and the current size of the window, we try to | |
1177 | * expand the reservation window size if necessary on a best-effort | |
617ba13b | 1178 | * basis before ext4_new_blocks() tries to allocate blocks, |
ac27a0ec | 1179 | */ |
617ba13b | 1180 | static void try_to_extend_reservation(struct ext4_reserve_window_node *my_rsv, |
ac27a0ec DK |
1181 | struct super_block *sb, int size) |
1182 | { | |
617ba13b | 1183 | struct ext4_reserve_window_node *next_rsv; |
ac27a0ec | 1184 | struct rb_node *next; |
617ba13b | 1185 | spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock; |
ac27a0ec DK |
1186 | |
1187 | if (!spin_trylock(rsv_lock)) | |
1188 | return; | |
1189 | ||
1190 | next = rb_next(&my_rsv->rsv_node); | |
1191 | ||
1192 | if (!next) | |
1193 | my_rsv->rsv_end += size; | |
1194 | else { | |
617ba13b | 1195 | next_rsv = list_entry(next, struct ext4_reserve_window_node, rsv_node); |
ac27a0ec DK |
1196 | |
1197 | if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size) | |
1198 | my_rsv->rsv_end += size; | |
1199 | else | |
1200 | my_rsv->rsv_end = next_rsv->rsv_start - 1; | |
1201 | } | |
1202 | spin_unlock(rsv_lock); | |
1203 | } | |
1204 | ||
1205 | /** | |
617ba13b | 1206 | * ext4_try_to_allocate_with_rsv() |
ac27a0ec DK |
1207 | * @sb: superblock |
1208 | * @handle: handle to this transaction | |
1209 | * @group: given allocation block group | |
1210 | * @bitmap_bh: bufferhead holds the block bitmap | |
1211 | * @grp_goal: given target block within the group | |
1212 | * @count: target number of blocks to allocate | |
1213 | * @my_rsv: reservation window | |
1214 | * @errp: pointer to store the error code | |
1215 | * | |
1216 | * This is the main function used to allocate a new block and its reservation | |
1217 | * window. | |
1218 | * | |
1219 | * Each time when a new block allocation is need, first try to allocate from | |
1220 | * its own reservation. If it does not have a reservation window, instead of | |
1221 | * looking for a free bit on bitmap first, then look up the reservation list to | |
1222 | * see if it is inside somebody else's reservation window, we try to allocate a | |
1223 | * reservation window for it starting from the goal first. Then do the block | |
1224 | * allocation within the reservation window. | |
1225 | * | |
1226 | * This will avoid keeping on searching the reservation list again and | |
1227 | * again when somebody is looking for a free block (without | |
1228 | * reservation), and there are lots of free blocks, but they are all | |
1229 | * being reserved. | |
1230 | * | |
1231 | * We use a red-black tree for the per-filesystem reservation list. | |
1232 | * | |
1233 | */ | |
617ba13b MC |
1234 | static ext4_grpblk_t |
1235 | ext4_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle, | |
ac27a0ec | 1236 | unsigned int group, struct buffer_head *bitmap_bh, |
617ba13b MC |
1237 | ext4_grpblk_t grp_goal, |
1238 | struct ext4_reserve_window_node * my_rsv, | |
ac27a0ec DK |
1239 | unsigned long *count, int *errp) |
1240 | { | |
617ba13b MC |
1241 | ext4_fsblk_t group_first_block, group_last_block; |
1242 | ext4_grpblk_t ret = 0; | |
ac27a0ec DK |
1243 | int fatal; |
1244 | unsigned long num = *count; | |
1245 | ||
1246 | *errp = 0; | |
1247 | ||
1248 | /* | |
1249 | * Make sure we use undo access for the bitmap, because it is critical | |
1250 | * that we do the frozen_data COW on bitmap buffers in all cases even | |
1251 | * if the buffer is in BJ_Forget state in the committing transaction. | |
1252 | */ | |
1253 | BUFFER_TRACE(bitmap_bh, "get undo access for new block"); | |
617ba13b | 1254 | fatal = ext4_journal_get_undo_access(handle, bitmap_bh); |
ac27a0ec DK |
1255 | if (fatal) { |
1256 | *errp = fatal; | |
1257 | return -1; | |
1258 | } | |
1259 | ||
1260 | /* | |
1261 | * we don't deal with reservation when | |
1262 | * filesystem is mounted without reservation | |
1263 | * or the file is not a regular file | |
1264 | * or last attempt to allocate a block with reservation turned on failed | |
1265 | */ | |
1266 | if (my_rsv == NULL ) { | |
617ba13b | 1267 | ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh, |
ac27a0ec DK |
1268 | grp_goal, count, NULL); |
1269 | goto out; | |
1270 | } | |
1271 | /* | |
1272 | * grp_goal is a group relative block number (if there is a goal) | |
617ba13b | 1273 | * 0 < grp_goal < EXT4_BLOCKS_PER_GROUP(sb) |
ac27a0ec DK |
1274 | * first block is a filesystem wide block number |
1275 | * first block is the block number of the first block in this group | |
1276 | */ | |
617ba13b MC |
1277 | group_first_block = ext4_group_first_block_no(sb, group); |
1278 | group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); | |
ac27a0ec DK |
1279 | |
1280 | /* | |
1281 | * Basically we will allocate a new block from inode's reservation | |
1282 | * window. | |
1283 | * | |
1284 | * We need to allocate a new reservation window, if: | |
1285 | * a) inode does not have a reservation window; or | |
1286 | * b) last attempt to allocate a block from existing reservation | |
1287 | * failed; or | |
1288 | * c) we come here with a goal and with a reservation window | |
1289 | * | |
1290 | * We do not need to allocate a new reservation window if we come here | |
1291 | * at the beginning with a goal and the goal is inside the window, or | |
1292 | * we don't have a goal but already have a reservation window. | |
1293 | * then we could go to allocate from the reservation window directly. | |
1294 | */ | |
1295 | while (1) { | |
1296 | if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) || | |
1297 | !goal_in_my_reservation(&my_rsv->rsv_window, | |
1298 | grp_goal, group, sb)) { | |
1299 | if (my_rsv->rsv_goal_size < *count) | |
1300 | my_rsv->rsv_goal_size = *count; | |
1301 | ret = alloc_new_reservation(my_rsv, grp_goal, sb, | |
1302 | group, bitmap_bh); | |
1303 | if (ret < 0) | |
1304 | break; /* failed */ | |
1305 | ||
1306 | if (!goal_in_my_reservation(&my_rsv->rsv_window, | |
1307 | grp_goal, group, sb)) | |
1308 | grp_goal = -1; | |
1309 | } else if (grp_goal > 0 && | |
1310 | (my_rsv->rsv_end-grp_goal+1) < *count) | |
1311 | try_to_extend_reservation(my_rsv, sb, | |
1312 | *count-my_rsv->rsv_end + grp_goal - 1); | |
1313 | ||
1314 | if ((my_rsv->rsv_start > group_last_block) || | |
1315 | (my_rsv->rsv_end < group_first_block)) { | |
617ba13b | 1316 | rsv_window_dump(&EXT4_SB(sb)->s_rsv_window_root, 1); |
ac27a0ec DK |
1317 | BUG(); |
1318 | } | |
617ba13b | 1319 | ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh, |
ac27a0ec DK |
1320 | grp_goal, &num, &my_rsv->rsv_window); |
1321 | if (ret >= 0) { | |
1322 | my_rsv->rsv_alloc_hit += num; | |
1323 | *count = num; | |
1324 | break; /* succeed */ | |
1325 | } | |
1326 | num = *count; | |
1327 | } | |
1328 | out: | |
1329 | if (ret >= 0) { | |
1330 | BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for " | |
1331 | "bitmap block"); | |
617ba13b | 1332 | fatal = ext4_journal_dirty_metadata(handle, bitmap_bh); |
ac27a0ec DK |
1333 | if (fatal) { |
1334 | *errp = fatal; | |
1335 | return -1; | |
1336 | } | |
1337 | return ret; | |
1338 | } | |
1339 | ||
1340 | BUFFER_TRACE(bitmap_bh, "journal_release_buffer"); | |
617ba13b | 1341 | ext4_journal_release_buffer(handle, bitmap_bh); |
ac27a0ec DK |
1342 | return ret; |
1343 | } | |
1344 | ||
1345 | /** | |
617ba13b | 1346 | * ext4_has_free_blocks() |
ac27a0ec DK |
1347 | * @sbi: in-core super block structure. |
1348 | * | |
1349 | * Check if filesystem has at least 1 free block available for allocation. | |
1350 | */ | |
617ba13b | 1351 | static int ext4_has_free_blocks(struct ext4_sb_info *sbi) |
ac27a0ec | 1352 | { |
617ba13b | 1353 | ext4_fsblk_t free_blocks, root_blocks; |
ac27a0ec DK |
1354 | |
1355 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
bd81d8ee | 1356 | root_blocks = ext4_r_blocks_count(sbi->s_es); |
ac27a0ec DK |
1357 | if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) && |
1358 | sbi->s_resuid != current->fsuid && | |
1359 | (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) { | |
1360 | return 0; | |
1361 | } | |
1362 | return 1; | |
1363 | } | |
1364 | ||
1365 | /** | |
617ba13b | 1366 | * ext4_should_retry_alloc() |
ac27a0ec DK |
1367 | * @sb: super block |
1368 | * @retries number of attemps has been made | |
1369 | * | |
617ba13b | 1370 | * ext4_should_retry_alloc() is called when ENOSPC is returned, and if |
ac27a0ec DK |
1371 | * it is profitable to retry the operation, this function will wait |
1372 | * for the current or commiting transaction to complete, and then | |
1373 | * return TRUE. | |
1374 | * | |
1375 | * if the total number of retries exceed three times, return FALSE. | |
1376 | */ | |
617ba13b | 1377 | int ext4_should_retry_alloc(struct super_block *sb, int *retries) |
ac27a0ec | 1378 | { |
617ba13b | 1379 | if (!ext4_has_free_blocks(EXT4_SB(sb)) || (*retries)++ > 3) |
ac27a0ec DK |
1380 | return 0; |
1381 | ||
1382 | jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id); | |
1383 | ||
dab291af | 1384 | return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal); |
ac27a0ec DK |
1385 | } |
1386 | ||
1387 | /** | |
617ba13b | 1388 | * ext4_new_blocks() -- core block(s) allocation function |
ac27a0ec DK |
1389 | * @handle: handle to this transaction |
1390 | * @inode: file inode | |
1391 | * @goal: given target block(filesystem wide) | |
1392 | * @count: target number of blocks to allocate | |
1393 | * @errp: error code | |
1394 | * | |
617ba13b | 1395 | * ext4_new_blocks uses a goal block to assist allocation. It tries to |
ac27a0ec DK |
1396 | * allocate block(s) from the block group contains the goal block first. If that |
1397 | * fails, it will try to allocate block(s) from other block groups without | |
1398 | * any specific goal block. | |
1399 | * | |
1400 | */ | |
617ba13b MC |
1401 | ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode, |
1402 | ext4_fsblk_t goal, unsigned long *count, int *errp) | |
ac27a0ec DK |
1403 | { |
1404 | struct buffer_head *bitmap_bh = NULL; | |
1405 | struct buffer_head *gdp_bh; | |
3a5b2ecd | 1406 | unsigned long group_no; |
ac27a0ec | 1407 | int goal_group; |
617ba13b MC |
1408 | ext4_grpblk_t grp_target_blk; /* blockgroup relative goal block */ |
1409 | ext4_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/ | |
1410 | ext4_fsblk_t ret_block; /* filesyetem-wide allocated block */ | |
ac27a0ec DK |
1411 | int bgi; /* blockgroup iteration index */ |
1412 | int fatal = 0, err; | |
1413 | int performed_allocation = 0; | |
617ba13b | 1414 | ext4_grpblk_t free_blocks; /* number of free blocks in a group */ |
ac27a0ec | 1415 | struct super_block *sb; |
617ba13b MC |
1416 | struct ext4_group_desc *gdp; |
1417 | struct ext4_super_block *es; | |
1418 | struct ext4_sb_info *sbi; | |
1419 | struct ext4_reserve_window_node *my_rsv = NULL; | |
1420 | struct ext4_block_alloc_info *block_i; | |
ac27a0ec | 1421 | unsigned short windowsz = 0; |
617ba13b | 1422 | #ifdef EXT4FS_DEBUG |
ac27a0ec DK |
1423 | static int goal_hits, goal_attempts; |
1424 | #endif | |
1425 | unsigned long ngroups; | |
1426 | unsigned long num = *count; | |
1427 | ||
1428 | *errp = -ENOSPC; | |
1429 | sb = inode->i_sb; | |
1430 | if (!sb) { | |
617ba13b | 1431 | printk("ext4_new_block: nonexistent device"); |
ac27a0ec DK |
1432 | return 0; |
1433 | } | |
1434 | ||
1435 | /* | |
1436 | * Check quota for allocation of this block. | |
1437 | */ | |
1438 | if (DQUOT_ALLOC_BLOCK(inode, num)) { | |
1439 | *errp = -EDQUOT; | |
1440 | return 0; | |
1441 | } | |
1442 | ||
617ba13b MC |
1443 | sbi = EXT4_SB(sb); |
1444 | es = EXT4_SB(sb)->s_es; | |
1445 | ext4_debug("goal=%lu.\n", goal); | |
ac27a0ec DK |
1446 | /* |
1447 | * Allocate a block from reservation only when | |
1448 | * filesystem is mounted with reservation(default,-o reservation), and | |
1449 | * it's a regular file, and | |
1450 | * the desired window size is greater than 0 (One could use ioctl | |
617ba13b | 1451 | * command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off |
ac27a0ec DK |
1452 | * reservation on that particular file) |
1453 | */ | |
617ba13b | 1454 | block_i = EXT4_I(inode)->i_block_alloc_info; |
ac27a0ec DK |
1455 | if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0)) |
1456 | my_rsv = &block_i->rsv_window_node; | |
1457 | ||
617ba13b | 1458 | if (!ext4_has_free_blocks(sbi)) { |
ac27a0ec DK |
1459 | *errp = -ENOSPC; |
1460 | goto out; | |
1461 | } | |
1462 | ||
1463 | /* | |
1464 | * First, test whether the goal block is free. | |
1465 | */ | |
1466 | if (goal < le32_to_cpu(es->s_first_data_block) || | |
bd81d8ee | 1467 | goal >= ext4_blocks_count(es)) |
ac27a0ec | 1468 | goal = le32_to_cpu(es->s_first_data_block); |
3a5b2ecd | 1469 | ext4_get_group_no_and_offset(sb, goal, &group_no, &grp_target_blk); |
ac27a0ec DK |
1470 | goal_group = group_no; |
1471 | retry_alloc: | |
617ba13b | 1472 | gdp = ext4_get_group_desc(sb, group_no, &gdp_bh); |
ac27a0ec DK |
1473 | if (!gdp) |
1474 | goto io_error; | |
1475 | ||
1476 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); | |
1477 | /* | |
1478 | * if there is not enough free blocks to make a new resevation | |
1479 | * turn off reservation for this allocation | |
1480 | */ | |
1481 | if (my_rsv && (free_blocks < windowsz) | |
1482 | && (rsv_is_empty(&my_rsv->rsv_window))) | |
1483 | my_rsv = NULL; | |
1484 | ||
1485 | if (free_blocks > 0) { | |
ac27a0ec DK |
1486 | bitmap_bh = read_block_bitmap(sb, group_no); |
1487 | if (!bitmap_bh) | |
1488 | goto io_error; | |
617ba13b | 1489 | grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, |
ac27a0ec DK |
1490 | group_no, bitmap_bh, grp_target_blk, |
1491 | my_rsv, &num, &fatal); | |
1492 | if (fatal) | |
1493 | goto out; | |
1494 | if (grp_alloc_blk >= 0) | |
1495 | goto allocated; | |
1496 | } | |
1497 | ||
617ba13b | 1498 | ngroups = EXT4_SB(sb)->s_groups_count; |
ac27a0ec DK |
1499 | smp_rmb(); |
1500 | ||
1501 | /* | |
1502 | * Now search the rest of the groups. We assume that | |
1503 | * i and gdp correctly point to the last group visited. | |
1504 | */ | |
1505 | for (bgi = 0; bgi < ngroups; bgi++) { | |
1506 | group_no++; | |
1507 | if (group_no >= ngroups) | |
1508 | group_no = 0; | |
617ba13b | 1509 | gdp = ext4_get_group_desc(sb, group_no, &gdp_bh); |
ac27a0ec DK |
1510 | if (!gdp) { |
1511 | *errp = -EIO; | |
1512 | goto out; | |
1513 | } | |
1514 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); | |
1515 | /* | |
1516 | * skip this group if the number of | |
1517 | * free blocks is less than half of the reservation | |
1518 | * window size. | |
1519 | */ | |
1520 | if (free_blocks <= (windowsz/2)) | |
1521 | continue; | |
1522 | ||
1523 | brelse(bitmap_bh); | |
1524 | bitmap_bh = read_block_bitmap(sb, group_no); | |
1525 | if (!bitmap_bh) | |
1526 | goto io_error; | |
1527 | /* | |
1528 | * try to allocate block(s) from this group, without a goal(-1). | |
1529 | */ | |
617ba13b | 1530 | grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, |
ac27a0ec DK |
1531 | group_no, bitmap_bh, -1, my_rsv, |
1532 | &num, &fatal); | |
1533 | if (fatal) | |
1534 | goto out; | |
1535 | if (grp_alloc_blk >= 0) | |
1536 | goto allocated; | |
1537 | } | |
1538 | /* | |
1539 | * We may end up a bogus ealier ENOSPC error due to | |
1540 | * filesystem is "full" of reservations, but | |
1541 | * there maybe indeed free blocks avaliable on disk | |
1542 | * In this case, we just forget about the reservations | |
1543 | * just do block allocation as without reservations. | |
1544 | */ | |
1545 | if (my_rsv) { | |
1546 | my_rsv = NULL; | |
1547 | group_no = goal_group; | |
1548 | goto retry_alloc; | |
1549 | } | |
1550 | /* No space left on the device */ | |
1551 | *errp = -ENOSPC; | |
1552 | goto out; | |
1553 | ||
1554 | allocated: | |
1555 | ||
617ba13b | 1556 | ext4_debug("using block group %d(%d)\n", |
ac27a0ec DK |
1557 | group_no, gdp->bg_free_blocks_count); |
1558 | ||
1559 | BUFFER_TRACE(gdp_bh, "get_write_access"); | |
617ba13b | 1560 | fatal = ext4_journal_get_write_access(handle, gdp_bh); |
ac27a0ec DK |
1561 | if (fatal) |
1562 | goto out; | |
1563 | ||
617ba13b | 1564 | ret_block = grp_alloc_blk + ext4_group_first_block_no(sb, group_no); |
ac27a0ec | 1565 | |
8fadc143 AR |
1566 | if (in_range(ext4_block_bitmap(sb, gdp), ret_block, num) || |
1567 | in_range(ext4_block_bitmap(sb, gdp), ret_block, num) || | |
1568 | in_range(ret_block, ext4_inode_table(sb, gdp), | |
bd81d8ee | 1569 | EXT4_SB(sb)->s_itb_per_group) || |
8fadc143 | 1570 | in_range(ret_block + num - 1, ext4_inode_table(sb, gdp), |
bd81d8ee | 1571 | EXT4_SB(sb)->s_itb_per_group)) |
617ba13b | 1572 | ext4_error(sb, "ext4_new_block", |
ac27a0ec | 1573 | "Allocating block in system zone - " |
2ae02107 | 1574 | "blocks from %llu, length %lu", |
ac27a0ec DK |
1575 | ret_block, num); |
1576 | ||
1577 | performed_allocation = 1; | |
1578 | ||
1579 | #ifdef CONFIG_JBD_DEBUG | |
1580 | { | |
1581 | struct buffer_head *debug_bh; | |
1582 | ||
1583 | /* Record bitmap buffer state in the newly allocated block */ | |
1584 | debug_bh = sb_find_get_block(sb, ret_block); | |
1585 | if (debug_bh) { | |
1586 | BUFFER_TRACE(debug_bh, "state when allocated"); | |
1587 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state"); | |
1588 | brelse(debug_bh); | |
1589 | } | |
1590 | } | |
1591 | jbd_lock_bh_state(bitmap_bh); | |
1592 | spin_lock(sb_bgl_lock(sbi, group_no)); | |
1593 | if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) { | |
1594 | int i; | |
1595 | ||
1596 | for (i = 0; i < num; i++) { | |
617ba13b | 1597 | if (ext4_test_bit(grp_alloc_blk+i, |
ac27a0ec DK |
1598 | bh2jh(bitmap_bh)->b_committed_data)) { |
1599 | printk("%s: block was unexpectedly set in " | |
1600 | "b_committed_data\n", __FUNCTION__); | |
1601 | } | |
1602 | } | |
1603 | } | |
617ba13b | 1604 | ext4_debug("found bit %d\n", grp_alloc_blk); |
ac27a0ec DK |
1605 | spin_unlock(sb_bgl_lock(sbi, group_no)); |
1606 | jbd_unlock_bh_state(bitmap_bh); | |
1607 | #endif | |
1608 | ||
bd81d8ee | 1609 | if (ret_block + num - 1 >= ext4_blocks_count(es)) { |
617ba13b | 1610 | ext4_error(sb, "ext4_new_block", |
2ae02107 | 1611 | "block(%llu) >= blocks count(%llu) - " |
3a5b2ecd | 1612 | "block_group = %lu, es == %p ", ret_block, |
bd81d8ee | 1613 | ext4_blocks_count(es), group_no, es); |
ac27a0ec DK |
1614 | goto out; |
1615 | } | |
1616 | ||
1617 | /* | |
1618 | * It is up to the caller to add the new buffer to a journal | |
1619 | * list of some description. We don't know in advance whether | |
1620 | * the caller wants to use it as metadata or data. | |
1621 | */ | |
617ba13b | 1622 | ext4_debug("allocating block %lu. Goal hits %d of %d.\n", |
ac27a0ec DK |
1623 | ret_block, goal_hits, goal_attempts); |
1624 | ||
1625 | spin_lock(sb_bgl_lock(sbi, group_no)); | |
1626 | gdp->bg_free_blocks_count = | |
1627 | cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count)-num); | |
1628 | spin_unlock(sb_bgl_lock(sbi, group_no)); | |
1629 | percpu_counter_mod(&sbi->s_freeblocks_counter, -num); | |
1630 | ||
1631 | BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor"); | |
617ba13b | 1632 | err = ext4_journal_dirty_metadata(handle, gdp_bh); |
ac27a0ec DK |
1633 | if (!fatal) |
1634 | fatal = err; | |
1635 | ||
1636 | sb->s_dirt = 1; | |
1637 | if (fatal) | |
1638 | goto out; | |
1639 | ||
1640 | *errp = 0; | |
1641 | brelse(bitmap_bh); | |
1642 | DQUOT_FREE_BLOCK(inode, *count-num); | |
1643 | *count = num; | |
1644 | return ret_block; | |
1645 | ||
1646 | io_error: | |
1647 | *errp = -EIO; | |
1648 | out: | |
1649 | if (fatal) { | |
1650 | *errp = fatal; | |
617ba13b | 1651 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
1652 | } |
1653 | /* | |
1654 | * Undo the block allocation | |
1655 | */ | |
1656 | if (!performed_allocation) | |
1657 | DQUOT_FREE_BLOCK(inode, *count); | |
1658 | brelse(bitmap_bh); | |
1659 | return 0; | |
1660 | } | |
1661 | ||
617ba13b MC |
1662 | ext4_fsblk_t ext4_new_block(handle_t *handle, struct inode *inode, |
1663 | ext4_fsblk_t goal, int *errp) | |
ac27a0ec DK |
1664 | { |
1665 | unsigned long count = 1; | |
1666 | ||
617ba13b | 1667 | return ext4_new_blocks(handle, inode, goal, &count, errp); |
ac27a0ec DK |
1668 | } |
1669 | ||
1670 | /** | |
617ba13b | 1671 | * ext4_count_free_blocks() -- count filesystem free blocks |
ac27a0ec DK |
1672 | * @sb: superblock |
1673 | * | |
1674 | * Adds up the number of free blocks from each block group. | |
1675 | */ | |
617ba13b | 1676 | ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb) |
ac27a0ec | 1677 | { |
617ba13b MC |
1678 | ext4_fsblk_t desc_count; |
1679 | struct ext4_group_desc *gdp; | |
ac27a0ec | 1680 | int i; |
617ba13b MC |
1681 | unsigned long ngroups = EXT4_SB(sb)->s_groups_count; |
1682 | #ifdef EXT4FS_DEBUG | |
1683 | struct ext4_super_block *es; | |
1684 | ext4_fsblk_t bitmap_count; | |
ac27a0ec DK |
1685 | unsigned long x; |
1686 | struct buffer_head *bitmap_bh = NULL; | |
1687 | ||
617ba13b | 1688 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1689 | desc_count = 0; |
1690 | bitmap_count = 0; | |
1691 | gdp = NULL; | |
1692 | ||
1693 | smp_rmb(); | |
1694 | for (i = 0; i < ngroups; i++) { | |
617ba13b | 1695 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1696 | if (!gdp) |
1697 | continue; | |
1698 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); | |
1699 | brelse(bitmap_bh); | |
1700 | bitmap_bh = read_block_bitmap(sb, i); | |
1701 | if (bitmap_bh == NULL) | |
1702 | continue; | |
1703 | ||
617ba13b | 1704 | x = ext4_count_free(bitmap_bh, sb->s_blocksize); |
ac27a0ec DK |
1705 | printk("group %d: stored = %d, counted = %lu\n", |
1706 | i, le16_to_cpu(gdp->bg_free_blocks_count), x); | |
1707 | bitmap_count += x; | |
1708 | } | |
1709 | brelse(bitmap_bh); | |
2ae02107 MC |
1710 | printk("ext4_count_free_blocks: stored = %llu" |
1711 | ", computed = %llu, %llu\n", | |
bd81d8ee | 1712 | EXT4_FREE_BLOCKS_COUNT(es), |
ac27a0ec DK |
1713 | desc_count, bitmap_count); |
1714 | return bitmap_count; | |
1715 | #else | |
1716 | desc_count = 0; | |
1717 | smp_rmb(); | |
1718 | for (i = 0; i < ngroups; i++) { | |
617ba13b | 1719 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1720 | if (!gdp) |
1721 | continue; | |
1722 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); | |
1723 | } | |
1724 | ||
1725 | return desc_count; | |
1726 | #endif | |
1727 | } | |
1728 | ||
1729 | static inline int | |
617ba13b | 1730 | block_in_use(ext4_fsblk_t block, struct super_block *sb, unsigned char *map) |
ac27a0ec | 1731 | { |
3a5b2ecd MC |
1732 | ext4_grpblk_t offset; |
1733 | ||
1734 | ext4_get_group_no_and_offset(sb, block, NULL, &offset); | |
1735 | return ext4_test_bit (offset, map); | |
ac27a0ec DK |
1736 | } |
1737 | ||
1738 | static inline int test_root(int a, int b) | |
1739 | { | |
1740 | int num = b; | |
1741 | ||
1742 | while (a > num) | |
1743 | num *= b; | |
1744 | return num == a; | |
1745 | } | |
1746 | ||
617ba13b | 1747 | static int ext4_group_sparse(int group) |
ac27a0ec DK |
1748 | { |
1749 | if (group <= 1) | |
1750 | return 1; | |
1751 | if (!(group & 1)) | |
1752 | return 0; | |
1753 | return (test_root(group, 7) || test_root(group, 5) || | |
1754 | test_root(group, 3)); | |
1755 | } | |
1756 | ||
1757 | /** | |
617ba13b | 1758 | * ext4_bg_has_super - number of blocks used by the superblock in group |
ac27a0ec DK |
1759 | * @sb: superblock for filesystem |
1760 | * @group: group number to check | |
1761 | * | |
1762 | * Return the number of blocks used by the superblock (primary or backup) | |
1763 | * in this group. Currently this will be only 0 or 1. | |
1764 | */ | |
617ba13b | 1765 | int ext4_bg_has_super(struct super_block *sb, int group) |
ac27a0ec | 1766 | { |
617ba13b MC |
1767 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
1768 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) && | |
1769 | !ext4_group_sparse(group)) | |
ac27a0ec DK |
1770 | return 0; |
1771 | return 1; | |
1772 | } | |
1773 | ||
617ba13b | 1774 | static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, int group) |
ac27a0ec | 1775 | { |
617ba13b MC |
1776 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); |
1777 | unsigned long first = metagroup * EXT4_DESC_PER_BLOCK(sb); | |
1778 | unsigned long last = first + EXT4_DESC_PER_BLOCK(sb) - 1; | |
ac27a0ec DK |
1779 | |
1780 | if (group == first || group == first + 1 || group == last) | |
1781 | return 1; | |
1782 | return 0; | |
1783 | } | |
1784 | ||
617ba13b | 1785 | static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, int group) |
ac27a0ec | 1786 | { |
617ba13b MC |
1787 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
1788 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) && | |
1789 | !ext4_group_sparse(group)) | |
ac27a0ec | 1790 | return 0; |
617ba13b | 1791 | return EXT4_SB(sb)->s_gdb_count; |
ac27a0ec DK |
1792 | } |
1793 | ||
1794 | /** | |
617ba13b | 1795 | * ext4_bg_num_gdb - number of blocks used by the group table in group |
ac27a0ec DK |
1796 | * @sb: superblock for filesystem |
1797 | * @group: group number to check | |
1798 | * | |
1799 | * Return the number of blocks used by the group descriptor table | |
1800 | * (primary or backup) in this group. In the future there may be a | |
1801 | * different number of descriptor blocks in each group. | |
1802 | */ | |
617ba13b | 1803 | unsigned long ext4_bg_num_gdb(struct super_block *sb, int group) |
ac27a0ec DK |
1804 | { |
1805 | unsigned long first_meta_bg = | |
617ba13b MC |
1806 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); |
1807 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); | |
ac27a0ec | 1808 | |
617ba13b | 1809 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) || |
ac27a0ec | 1810 | metagroup < first_meta_bg) |
617ba13b | 1811 | return ext4_bg_num_gdb_nometa(sb,group); |
ac27a0ec | 1812 | |
617ba13b | 1813 | return ext4_bg_num_gdb_meta(sb,group); |
ac27a0ec DK |
1814 | |
1815 | } |