1 |
62 |
marcus.erl |
/*
|
2 |
|
|
* linux/fs/ext4/balloc.c
|
3 |
|
|
*
|
4 |
|
|
* Copyright (C) 1992, 1993, 1994, 1995
|
5 |
|
|
* Remy Card (card@masi.ibp.fr)
|
6 |
|
|
* Laboratoire MASI - Institut Blaise Pascal
|
7 |
|
|
* Universite Pierre et Marie Curie (Paris VI)
|
8 |
|
|
*
|
9 |
|
|
* Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
|
10 |
|
|
* Big-endian to little-endian byte-swapping/bitmaps by
|
11 |
|
|
* David S. Miller (davem@caip.rutgers.edu), 1995
|
12 |
|
|
*/
|
13 |
|
|
|
14 |
|
|
#include <linux/time.h>
|
15 |
|
|
#include <linux/capability.h>
|
16 |
|
|
#include <linux/fs.h>
|
17 |
|
|
#include <linux/jbd2.h>
|
18 |
|
|
#include <linux/ext4_fs.h>
|
19 |
|
|
#include <linux/ext4_jbd2.h>
|
20 |
|
|
#include <linux/quotaops.h>
|
21 |
|
|
#include <linux/buffer_head.h>
|
22 |
|
|
|
23 |
|
|
#include "group.h"
|
24 |
|
|
/*
|
25 |
|
|
* balloc.c contains the blocks allocation and deallocation routines
|
26 |
|
|
*/
|
27 |
|
|
|
28 |
|
|
/*
|
29 |
|
|
* Calculate the block group number and offset, given a block number
|
30 |
|
|
*/
|
31 |
|
|
void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
|
32 |
|
|
unsigned long *blockgrpp, ext4_grpblk_t *offsetp)
|
33 |
|
|
{
|
34 |
|
|
struct ext4_super_block *es = EXT4_SB(sb)->s_es;
|
35 |
|
|
ext4_grpblk_t offset;
|
36 |
|
|
|
37 |
|
|
blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
|
38 |
|
|
offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
|
39 |
|
|
if (offsetp)
|
40 |
|
|
*offsetp = offset;
|
41 |
|
|
if (blockgrpp)
|
42 |
|
|
*blockgrpp = blocknr;
|
43 |
|
|
|
44 |
|
|
}
|
45 |
|
|
|
46 |
|
|
/* Initializes an uninitialized block bitmap if given, and returns the
|
47 |
|
|
* number of blocks free in the group. */
|
48 |
|
|
unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
|
49 |
|
|
int block_group, struct ext4_group_desc *gdp)
|
50 |
|
|
{
|
51 |
|
|
unsigned long start;
|
52 |
|
|
int bit, bit_max;
|
53 |
|
|
unsigned free_blocks, group_blocks;
|
54 |
|
|
struct ext4_sb_info *sbi = EXT4_SB(sb);
|
55 |
|
|
|
56 |
|
|
if (bh) {
|
57 |
|
|
J_ASSERT_BH(bh, buffer_locked(bh));
|
58 |
|
|
|
59 |
|
|
/* If checksum is bad mark all blocks used to prevent allocation
|
60 |
|
|
* essentially implementing a per-group read-only flag. */
|
61 |
|
|
if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
|
62 |
|
|
ext4_error(sb, __FUNCTION__,
|
63 |
|
|
"Checksum bad for group %u\n", block_group);
|
64 |
|
|
gdp->bg_free_blocks_count = 0;
|
65 |
|
|
gdp->bg_free_inodes_count = 0;
|
66 |
|
|
gdp->bg_itable_unused = 0;
|
67 |
|
|
memset(bh->b_data, 0xff, sb->s_blocksize);
|
68 |
|
|
return 0;
|
69 |
|
|
}
|
70 |
|
|
memset(bh->b_data, 0, sb->s_blocksize);
|
71 |
|
|
}
|
72 |
|
|
|
73 |
|
|
/* Check for superblock and gdt backups in this group */
|
74 |
|
|
bit_max = ext4_bg_has_super(sb, block_group);
|
75 |
|
|
|
76 |
|
|
if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
|
77 |
|
|
block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
|
78 |
|
|
sbi->s_desc_per_block) {
|
79 |
|
|
if (bit_max) {
|
80 |
|
|
bit_max += ext4_bg_num_gdb(sb, block_group);
|
81 |
|
|
bit_max +=
|
82 |
|
|
le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
|
83 |
|
|
}
|
84 |
|
|
} else { /* For META_BG_BLOCK_GROUPS */
|
85 |
|
|
int group_rel = (block_group -
|
86 |
|
|
le32_to_cpu(sbi->s_es->s_first_meta_bg)) %
|
87 |
|
|
EXT4_DESC_PER_BLOCK(sb);
|
88 |
|
|
if (group_rel == 0 || group_rel == 1 ||
|
89 |
|
|
(group_rel == EXT4_DESC_PER_BLOCK(sb) - 1))
|
90 |
|
|
bit_max += 1;
|
91 |
|
|
}
|
92 |
|
|
|
93 |
|
|
if (block_group == sbi->s_groups_count - 1) {
|
94 |
|
|
/*
|
95 |
|
|
* Even though mke2fs always initialize first and last group
|
96 |
|
|
* if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
|
97 |
|
|
* to make sure we calculate the right free blocks
|
98 |
|
|
*/
|
99 |
|
|
group_blocks = ext4_blocks_count(sbi->s_es) -
|
100 |
|
|
le32_to_cpu(sbi->s_es->s_first_data_block) -
|
101 |
|
|
(EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count -1));
|
102 |
|
|
} else {
|
103 |
|
|
group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
|
104 |
|
|
}
|
105 |
|
|
|
106 |
|
|
free_blocks = group_blocks - bit_max;
|
107 |
|
|
|
108 |
|
|
if (bh) {
|
109 |
|
|
for (bit = 0; bit < bit_max; bit++)
|
110 |
|
|
ext4_set_bit(bit, bh->b_data);
|
111 |
|
|
|
112 |
|
|
start = block_group * EXT4_BLOCKS_PER_GROUP(sb) +
|
113 |
|
|
le32_to_cpu(sbi->s_es->s_first_data_block);
|
114 |
|
|
|
115 |
|
|
/* Set bits for block and inode bitmaps, and inode table */
|
116 |
|
|
ext4_set_bit(ext4_block_bitmap(sb, gdp) - start, bh->b_data);
|
117 |
|
|
ext4_set_bit(ext4_inode_bitmap(sb, gdp) - start, bh->b_data);
|
118 |
|
|
for (bit = (ext4_inode_table(sb, gdp) - start),
|
119 |
|
|
bit_max = bit + sbi->s_itb_per_group; bit < bit_max; bit++)
|
120 |
|
|
ext4_set_bit(bit, bh->b_data);
|
121 |
|
|
|
122 |
|
|
/*
|
123 |
|
|
* Also if the number of blocks within the group is
|
124 |
|
|
* less than the blocksize * 8 ( which is the size
|
125 |
|
|
* of bitmap ), set rest of the block bitmap to 1
|
126 |
|
|
*/
|
127 |
|
|
mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
|
128 |
|
|
}
|
129 |
|
|
|
130 |
|
|
return free_blocks - sbi->s_itb_per_group - 2;
|
131 |
|
|
}
|
132 |
|
|
|
133 |
|
|
|
134 |
|
|
/*
|
135 |
|
|
* The free blocks are managed by bitmaps. A file system contains several
|
136 |
|
|
* blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
|
137 |
|
|
* block for inodes, N blocks for the inode table and data blocks.
|
138 |
|
|
*
|
139 |
|
|
* The file system contains group descriptors which are located after the
|
140 |
|
|
* super block. Each descriptor contains the number of the bitmap block and
|
141 |
|
|
* the free blocks count in the block. The descriptors are loaded in memory
|
142 |
|
|
* when a file system is mounted (see ext4_fill_super).
|
143 |
|
|
*/
|
144 |
|
|
|
145 |
|
|
|
146 |
|
|
#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
|
147 |
|
|
|
148 |
|
|
/**
|
149 |
|
|
* ext4_get_group_desc() -- load group descriptor from disk
|
150 |
|
|
* @sb: super block
|
151 |
|
|
* @block_group: given block group
|
152 |
|
|
* @bh: pointer to the buffer head to store the block
|
153 |
|
|
* group descriptor
|
154 |
|
|
*/
|
155 |
|
|
struct ext4_group_desc * ext4_get_group_desc(struct super_block * sb,
|
156 |
|
|
unsigned int block_group,
|
157 |
|
|
struct buffer_head ** bh)
|
158 |
|
|
{
|
159 |
|
|
unsigned long group_desc;
|
160 |
|
|
unsigned long offset;
|
161 |
|
|
struct ext4_group_desc * desc;
|
162 |
|
|
struct ext4_sb_info *sbi = EXT4_SB(sb);
|
163 |
|
|
|
164 |
|
|
if (block_group >= sbi->s_groups_count) {
|
165 |
|
|
ext4_error (sb, "ext4_get_group_desc",
|
166 |
|
|
"block_group >= groups_count - "
|
167 |
|
|
"block_group = %d, groups_count = %lu",
|
168 |
|
|
block_group, sbi->s_groups_count);
|
169 |
|
|
|
170 |
|
|
return NULL;
|
171 |
|
|
}
|
172 |
|
|
smp_rmb();
|
173 |
|
|
|
174 |
|
|
group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
|
175 |
|
|
offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
|
176 |
|
|
if (!sbi->s_group_desc[group_desc]) {
|
177 |
|
|
ext4_error (sb, "ext4_get_group_desc",
|
178 |
|
|
"Group descriptor not loaded - "
|
179 |
|
|
"block_group = %d, group_desc = %lu, desc = %lu",
|
180 |
|
|
block_group, group_desc, offset);
|
181 |
|
|
return NULL;
|
182 |
|
|
}
|
183 |
|
|
|
184 |
|
|
desc = (struct ext4_group_desc *)(
|
185 |
|
|
(__u8 *)sbi->s_group_desc[group_desc]->b_data +
|
186 |
|
|
offset * EXT4_DESC_SIZE(sb));
|
187 |
|
|
if (bh)
|
188 |
|
|
*bh = sbi->s_group_desc[group_desc];
|
189 |
|
|
return desc;
|
190 |
|
|
}
|
191 |
|
|
|
192 |
|
|
/**
|
193 |
|
|
* read_block_bitmap()
|
194 |
|
|
* @sb: super block
|
195 |
|
|
* @block_group: given block group
|
196 |
|
|
*
|
197 |
|
|
* Read the bitmap for a given block_group, reading into the specified
|
198 |
|
|
* slot in the superblock's bitmap cache.
|
199 |
|
|
*
|
200 |
|
|
* Return buffer_head on success or NULL in case of failure.
|
201 |
|
|
*/
|
202 |
|
|
struct buffer_head *
|
203 |
|
|
read_block_bitmap(struct super_block *sb, unsigned int block_group)
|
204 |
|
|
{
|
205 |
|
|
struct ext4_group_desc * desc;
|
206 |
|
|
struct buffer_head * bh = NULL;
|
207 |
|
|
ext4_fsblk_t bitmap_blk;
|
208 |
|
|
|
209 |
|
|
desc = ext4_get_group_desc(sb, block_group, NULL);
|
210 |
|
|
if (!desc)
|
211 |
|
|
return NULL;
|
212 |
|
|
bitmap_blk = ext4_block_bitmap(sb, desc);
|
213 |
|
|
if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
|
214 |
|
|
bh = sb_getblk(sb, bitmap_blk);
|
215 |
|
|
if (!buffer_uptodate(bh)) {
|
216 |
|
|
lock_buffer(bh);
|
217 |
|
|
if (!buffer_uptodate(bh)) {
|
218 |
|
|
ext4_init_block_bitmap(sb, bh, block_group,
|
219 |
|
|
desc);
|
220 |
|
|
set_buffer_uptodate(bh);
|
221 |
|
|
}
|
222 |
|
|
unlock_buffer(bh);
|
223 |
|
|
}
|
224 |
|
|
} else {
|
225 |
|
|
bh = sb_bread(sb, bitmap_blk);
|
226 |
|
|
}
|
227 |
|
|
if (!bh)
|
228 |
|
|
ext4_error (sb, __FUNCTION__,
|
229 |
|
|
"Cannot read block bitmap - "
|
230 |
|
|
"block_group = %d, block_bitmap = %llu",
|
231 |
|
|
block_group, bitmap_blk);
|
232 |
|
|
return bh;
|
233 |
|
|
}
|
234 |
|
|
/*
|
235 |
|
|
* The reservation window structure operations
|
236 |
|
|
* --------------------------------------------
|
237 |
|
|
* Operations include:
|
238 |
|
|
* dump, find, add, remove, is_empty, find_next_reservable_window, etc.
|
239 |
|
|
*
|
240 |
|
|
* We use a red-black tree to represent per-filesystem reservation
|
241 |
|
|
* windows.
|
242 |
|
|
*
|
243 |
|
|
*/
|
244 |
|
|
|
245 |
|
|
/**
|
246 |
|
|
* __rsv_window_dump() -- Dump the filesystem block allocation reservation map
|
247 |
|
|
* @rb_root: root of per-filesystem reservation rb tree
|
248 |
|
|
* @verbose: verbose mode
|
249 |
|
|
* @fn: function which wishes to dump the reservation map
|
250 |
|
|
*
|
251 |
|
|
* If verbose is turned on, it will print the whole block reservation
|
252 |
|
|
* windows(start, end). Otherwise, it will only print out the "bad" windows,
|
253 |
|
|
* those windows that overlap with their immediate neighbors.
|
254 |
|
|
*/
|
255 |
|
|
#if 1
|
256 |
|
|
static void __rsv_window_dump(struct rb_root *root, int verbose,
|
257 |
|
|
const char *fn)
|
258 |
|
|
{
|
259 |
|
|
struct rb_node *n;
|
260 |
|
|
struct ext4_reserve_window_node *rsv, *prev;
|
261 |
|
|
int bad;
|
262 |
|
|
|
263 |
|
|
restart:
|
264 |
|
|
n = rb_first(root);
|
265 |
|
|
bad = 0;
|
266 |
|
|
prev = NULL;
|
267 |
|
|
|
268 |
|
|
printk("Block Allocation Reservation Windows Map (%s):\n", fn);
|
269 |
|
|
while (n) {
|
270 |
|
|
rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node);
|
271 |
|
|
if (verbose)
|
272 |
|
|
printk("reservation window 0x%p "
|
273 |
|
|
"start: %llu, end: %llu\n",
|
274 |
|
|
rsv, rsv->rsv_start, rsv->rsv_end);
|
275 |
|
|
if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
|
276 |
|
|
printk("Bad reservation %p (start >= end)\n",
|
277 |
|
|
rsv);
|
278 |
|
|
bad = 1;
|
279 |
|
|
}
|
280 |
|
|
if (prev && prev->rsv_end >= rsv->rsv_start) {
|
281 |
|
|
printk("Bad reservation %p (prev->end >= start)\n",
|
282 |
|
|
rsv);
|
283 |
|
|
bad = 1;
|
284 |
|
|
}
|
285 |
|
|
if (bad) {
|
286 |
|
|
if (!verbose) {
|
287 |
|
|
printk("Restarting reservation walk in verbose mode\n");
|
288 |
|
|
verbose = 1;
|
289 |
|
|
goto restart;
|
290 |
|
|
}
|
291 |
|
|
}
|
292 |
|
|
n = rb_next(n);
|
293 |
|
|
prev = rsv;
|
294 |
|
|
}
|
295 |
|
|
printk("Window map complete.\n");
|
296 |
|
|
if (bad)
|
297 |
|
|
BUG();
|
298 |
|
|
}
|
299 |
|
|
#define rsv_window_dump(root, verbose) \
|
300 |
|
|
__rsv_window_dump((root), (verbose), __FUNCTION__)
|
301 |
|
|
#else
|
302 |
|
|
#define rsv_window_dump(root, verbose) do {} while (0)
|
303 |
|
|
#endif
|
304 |
|
|
|
305 |
|
|
/**
|
306 |
|
|
* goal_in_my_reservation()
|
307 |
|
|
* @rsv: inode's reservation window
|
308 |
|
|
* @grp_goal: given goal block relative to the allocation block group
|
309 |
|
|
* @group: the current allocation block group
|
310 |
|
|
* @sb: filesystem super block
|
311 |
|
|
*
|
312 |
|
|
* Test if the given goal block (group relative) is within the file's
|
313 |
|
|
* own block reservation window range.
|
314 |
|
|
*
|
315 |
|
|
* If the reservation window is outside the goal allocation group, return 0;
|
316 |
|
|
* grp_goal (given goal block) could be -1, which means no specific
|
317 |
|
|
* goal block. In this case, always return 1.
|
318 |
|
|
* If the goal block is within the reservation window, return 1;
|
319 |
|
|
* otherwise, return 0;
|
320 |
|
|
*/
|
321 |
|
|
static int
|
322 |
|
|
goal_in_my_reservation(struct ext4_reserve_window *rsv, ext4_grpblk_t grp_goal,
|
323 |
|
|
unsigned int group, struct super_block * sb)
|
324 |
|
|
{
|
325 |
|
|
ext4_fsblk_t group_first_block, group_last_block;
|
326 |
|
|
|
327 |
|
|
group_first_block = ext4_group_first_block_no(sb, group);
|
328 |
|
|
group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1);
|
329 |
|
|
|
330 |
|
|
if ((rsv->_rsv_start > group_last_block) ||
|
331 |
|
|
(rsv->_rsv_end < group_first_block))
|
332 |
|
|
return 0;
|
333 |
|
|
if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
|
334 |
|
|
|| (grp_goal + group_first_block > rsv->_rsv_end)))
|
335 |
|
|
return 0;
|
336 |
|
|
return 1;
|
337 |
|
|
}
|
338 |
|
|
|
339 |
|
|
/**
|
340 |
|
|
* search_reserve_window()
|
341 |
|
|
* @rb_root: root of reservation tree
|
342 |
|
|
* @goal: target allocation block
|
343 |
|
|
*
|
344 |
|
|
* Find the reserved window which includes the goal, or the previous one
|
345 |
|
|
* if the goal is not in any window.
|
346 |
|
|
* Returns NULL if there are no windows or if all windows start after the goal.
|
347 |
|
|
*/
|
348 |
|
|
static struct ext4_reserve_window_node *
|
349 |
|
|
search_reserve_window(struct rb_root *root, ext4_fsblk_t goal)
|
350 |
|
|
{
|
351 |
|
|
struct rb_node *n = root->rb_node;
|
352 |
|
|
struct ext4_reserve_window_node *rsv;
|
353 |
|
|
|
354 |
|
|
if (!n)
|
355 |
|
|
return NULL;
|
356 |
|
|
|
357 |
|
|
do {
|
358 |
|
|
rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node);
|
359 |
|
|
|
360 |
|
|
if (goal < rsv->rsv_start)
|
361 |
|
|
n = n->rb_left;
|
362 |
|
|
else if (goal > rsv->rsv_end)
|
363 |
|
|
n = n->rb_right;
|
364 |
|
|
else
|
365 |
|
|
return rsv;
|
366 |
|
|
} while (n);
|
367 |
|
|
/*
|
368 |
|
|
* We've fallen off the end of the tree: the goal wasn't inside
|
369 |
|
|
* any particular node. OK, the previous node must be to one
|
370 |
|
|
* side of the interval containing the goal. If it's the RHS,
|
371 |
|
|
* we need to back up one.
|
372 |
|
|
*/
|
373 |
|
|
if (rsv->rsv_start > goal) {
|
374 |
|
|
n = rb_prev(&rsv->rsv_node);
|
375 |
|
|
rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node);
|
376 |
|
|
}
|
377 |
|
|
return rsv;
|
378 |
|
|
}
|
379 |
|
|
|
380 |
|
|
/**
|
381 |
|
|
* ext4_rsv_window_add() -- Insert a window to the block reservation rb tree.
|
382 |
|
|
* @sb: super block
|
383 |
|
|
* @rsv: reservation window to add
|
384 |
|
|
*
|
385 |
|
|
* Must be called with rsv_lock hold.
|
386 |
|
|
*/
|
387 |
|
|
void ext4_rsv_window_add(struct super_block *sb,
|
388 |
|
|
struct ext4_reserve_window_node *rsv)
|
389 |
|
|
{
|
390 |
|
|
struct rb_root *root = &EXT4_SB(sb)->s_rsv_window_root;
|
391 |
|
|
struct rb_node *node = &rsv->rsv_node;
|
392 |
|
|
ext4_fsblk_t start = rsv->rsv_start;
|
393 |
|
|
|
394 |
|
|
struct rb_node ** p = &root->rb_node;
|
395 |
|
|
struct rb_node * parent = NULL;
|
396 |
|
|
struct ext4_reserve_window_node *this;
|
397 |
|
|
|
398 |
|
|
while (*p)
|
399 |
|
|
{
|
400 |
|
|
parent = *p;
|
401 |
|
|
this = rb_entry(parent, struct ext4_reserve_window_node, rsv_node);
|
402 |
|
|
|
403 |
|
|
if (start < this->rsv_start)
|
404 |
|
|
p = &(*p)->rb_left;
|
405 |
|
|
else if (start > this->rsv_end)
|
406 |
|
|
p = &(*p)->rb_right;
|
407 |
|
|
else {
|
408 |
|
|
rsv_window_dump(root, 1);
|
409 |
|
|
BUG();
|
410 |
|
|
}
|
411 |
|
|
}
|
412 |
|
|
|
413 |
|
|
rb_link_node(node, parent, p);
|
414 |
|
|
rb_insert_color(node, root);
|
415 |
|
|
}
|
416 |
|
|
|
417 |
|
|
/**
|
418 |
|
|
* ext4_rsv_window_remove() -- unlink a window from the reservation rb tree
|
419 |
|
|
* @sb: super block
|
420 |
|
|
* @rsv: reservation window to remove
|
421 |
|
|
*
|
422 |
|
|
* Mark the block reservation window as not allocated, and unlink it
|
423 |
|
|
* from the filesystem reservation window rb tree. Must be called with
|
424 |
|
|
* rsv_lock hold.
|
425 |
|
|
*/
|
426 |
|
|
static void rsv_window_remove(struct super_block *sb,
|
427 |
|
|
struct ext4_reserve_window_node *rsv)
|
428 |
|
|
{
|
429 |
|
|
rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
|
430 |
|
|
rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
|
431 |
|
|
rsv->rsv_alloc_hit = 0;
|
432 |
|
|
rb_erase(&rsv->rsv_node, &EXT4_SB(sb)->s_rsv_window_root);
|
433 |
|
|
}
|
434 |
|
|
|
435 |
|
|
/*
|
436 |
|
|
* rsv_is_empty() -- Check if the reservation window is allocated.
|
437 |
|
|
* @rsv: given reservation window to check
|
438 |
|
|
*
|
439 |
|
|
* returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED.
|
440 |
|
|
*/
|
441 |
|
|
static inline int rsv_is_empty(struct ext4_reserve_window *rsv)
|
442 |
|
|
{
|
443 |
|
|
/* a valid reservation end block could not be 0 */
|
444 |
|
|
return rsv->_rsv_end == EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
|
445 |
|
|
}
|
446 |
|
|
|
447 |
|
|
/**
|
448 |
|
|
* ext4_init_block_alloc_info()
|
449 |
|
|
* @inode: file inode structure
|
450 |
|
|
*
|
451 |
|
|
* Allocate and initialize the reservation window structure, and
|
452 |
|
|
* link the window to the ext4 inode structure at last
|
453 |
|
|
*
|
454 |
|
|
* The reservation window structure is only dynamically allocated
|
455 |
|
|
* and linked to ext4 inode the first time the open file
|
456 |
|
|
* needs a new block. So, before every ext4_new_block(s) call, for
|
457 |
|
|
* regular files, we should check whether the reservation window
|
458 |
|
|
* structure exists or not. In the latter case, this function is called.
|
459 |
|
|
* Fail to do so will result in block reservation being turned off for that
|
460 |
|
|
* open file.
|
461 |
|
|
*
|
462 |
|
|
* This function is called from ext4_get_blocks_handle(), also called
|
463 |
|
|
* when setting the reservation window size through ioctl before the file
|
464 |
|
|
* is open for write (needs block allocation).
|
465 |
|
|
*
|
466 |
|
|
* Needs truncate_mutex protection prior to call this function.
|
467 |
|
|
*/
|
468 |
|
|
void ext4_init_block_alloc_info(struct inode *inode)
|
469 |
|
|
{
|
470 |
|
|
struct ext4_inode_info *ei = EXT4_I(inode);
|
471 |
|
|
struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info;
|
472 |
|
|
struct super_block *sb = inode->i_sb;
|
473 |
|
|
|
474 |
|
|
block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
|
475 |
|
|
if (block_i) {
|
476 |
|
|
struct ext4_reserve_window_node *rsv = &block_i->rsv_window_node;
|
477 |
|
|
|
478 |
|
|
rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
|
479 |
|
|
rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
|
480 |
|
|
|
481 |
|
|
/*
|
482 |
|
|
* if filesystem is mounted with NORESERVATION, the goal
|
483 |
|
|
* reservation window size is set to zero to indicate
|
484 |
|
|
* block reservation is off
|
485 |
|
|
*/
|
486 |
|
|
if (!test_opt(sb, RESERVATION))
|
487 |
|
|
rsv->rsv_goal_size = 0;
|
488 |
|
|
else
|
489 |
|
|
rsv->rsv_goal_size = EXT4_DEFAULT_RESERVE_BLOCKS;
|
490 |
|
|
rsv->rsv_alloc_hit = 0;
|
491 |
|
|
block_i->last_alloc_logical_block = 0;
|
492 |
|
|
block_i->last_alloc_physical_block = 0;
|
493 |
|
|
}
|
494 |
|
|
ei->i_block_alloc_info = block_i;
|
495 |
|
|
}
|
496 |
|
|
|
497 |
|
|
/**
|
498 |
|
|
* ext4_discard_reservation()
|
499 |
|
|
* @inode: inode
|
500 |
|
|
*
|
501 |
|
|
* Discard(free) block reservation window on last file close, or truncate
|
502 |
|
|
* or at last iput().
|
503 |
|
|
*
|
504 |
|
|
* It is being called in three cases:
|
505 |
|
|
* ext4_release_file(): last writer close the file
|
506 |
|
|
* ext4_clear_inode(): last iput(), when nobody link to this file.
|
507 |
|
|
* ext4_truncate(): when the block indirect map is about to change.
|
508 |
|
|
*
|
509 |
|
|
*/
|
510 |
|
|
void ext4_discard_reservation(struct inode *inode)
|
511 |
|
|
{
|
512 |
|
|
struct ext4_inode_info *ei = EXT4_I(inode);
|
513 |
|
|
struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info;
|
514 |
|
|
struct ext4_reserve_window_node *rsv;
|
515 |
|
|
spinlock_t *rsv_lock = &EXT4_SB(inode->i_sb)->s_rsv_window_lock;
|
516 |
|
|
|
517 |
|
|
if (!block_i)
|
518 |
|
|
return;
|
519 |
|
|
|
520 |
|
|
rsv = &block_i->rsv_window_node;
|
521 |
|
|
if (!rsv_is_empty(&rsv->rsv_window)) {
|
522 |
|
|
spin_lock(rsv_lock);
|
523 |
|
|
if (!rsv_is_empty(&rsv->rsv_window))
|
524 |
|
|
rsv_window_remove(inode->i_sb, rsv);
|
525 |
|
|
spin_unlock(rsv_lock);
|
526 |
|
|
}
|
527 |
|
|
}
|
528 |
|
|
|
529 |
|
|
/**
|
530 |
|
|
* ext4_free_blocks_sb() -- Free given blocks and update quota
|
531 |
|
|
* @handle: handle to this transaction
|
532 |
|
|
* @sb: super block
|
533 |
|
|
* @block: start physcial block to free
|
534 |
|
|
* @count: number of blocks to free
|
535 |
|
|
* @pdquot_freed_blocks: pointer to quota
|
536 |
|
|
*/
|
537 |
|
|
void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb,
|
538 |
|
|
ext4_fsblk_t block, unsigned long count,
|
539 |
|
|
unsigned long *pdquot_freed_blocks)
|
540 |
|
|
{
|
541 |
|
|
struct buffer_head *bitmap_bh = NULL;
|
542 |
|
|
struct buffer_head *gd_bh;
|
543 |
|
|
unsigned long block_group;
|
544 |
|
|
ext4_grpblk_t bit;
|
545 |
|
|
unsigned long i;
|
546 |
|
|
unsigned long overflow;
|
547 |
|
|
struct ext4_group_desc * desc;
|
548 |
|
|
struct ext4_super_block * es;
|
549 |
|
|
struct ext4_sb_info *sbi;
|
550 |
|
|
int err = 0, ret;
|
551 |
|
|
ext4_grpblk_t group_freed;
|
552 |
|
|
|
553 |
|
|
*pdquot_freed_blocks = 0;
|
554 |
|
|
sbi = EXT4_SB(sb);
|
555 |
|
|
es = sbi->s_es;
|
556 |
|
|
if (block < le32_to_cpu(es->s_first_data_block) ||
|
557 |
|
|
block + count < block ||
|
558 |
|
|
block + count > ext4_blocks_count(es)) {
|
559 |
|
|
ext4_error (sb, "ext4_free_blocks",
|
560 |
|
|
"Freeing blocks not in datazone - "
|
561 |
|
|
"block = %llu, count = %lu", block, count);
|
562 |
|
|
goto error_return;
|
563 |
|
|
}
|
564 |
|
|
|
565 |
|
|
ext4_debug ("freeing block(s) %llu-%llu\n", block, block + count - 1);
|
566 |
|
|
|
567 |
|
|
do_more:
|
568 |
|
|
overflow = 0;
|
569 |
|
|
ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
|
570 |
|
|
/*
|
571 |
|
|
* Check to see if we are freeing blocks across a group
|
572 |
|
|
* boundary.
|
573 |
|
|
*/
|
574 |
|
|
if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
|
575 |
|
|
overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb);
|
576 |
|
|
count -= overflow;
|
577 |
|
|
}
|
578 |
|
|
brelse(bitmap_bh);
|
579 |
|
|
bitmap_bh = read_block_bitmap(sb, block_group);
|
580 |
|
|
if (!bitmap_bh)
|
581 |
|
|
goto error_return;
|
582 |
|
|
desc = ext4_get_group_desc (sb, block_group, &gd_bh);
|
583 |
|
|
if (!desc)
|
584 |
|
|
goto error_return;
|
585 |
|
|
|
586 |
|
|
if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
|
587 |
|
|
in_range(ext4_inode_bitmap(sb, desc), block, count) ||
|
588 |
|
|
in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
|
589 |
|
|
in_range(block + count - 1, ext4_inode_table(sb, desc),
|
590 |
|
|
sbi->s_itb_per_group))
|
591 |
|
|
ext4_error (sb, "ext4_free_blocks",
|
592 |
|
|
"Freeing blocks in system zones - "
|
593 |
|
|
"Block = %llu, count = %lu",
|
594 |
|
|
block, count);
|
595 |
|
|
|
596 |
|
|
/*
|
597 |
|
|
* We are about to start releasing blocks in the bitmap,
|
598 |
|
|
* so we need undo access.
|
599 |
|
|
*/
|
600 |
|
|
/* @@@ check errors */
|
601 |
|
|
BUFFER_TRACE(bitmap_bh, "getting undo access");
|
602 |
|
|
err = ext4_journal_get_undo_access(handle, bitmap_bh);
|
603 |
|
|
if (err)
|
604 |
|
|
goto error_return;
|
605 |
|
|
|
606 |
|
|
/*
|
607 |
|
|
* We are about to modify some metadata. Call the journal APIs
|
608 |
|
|
* to unshare ->b_data if a currently-committing transaction is
|
609 |
|
|
* using it
|
610 |
|
|
*/
|
611 |
|
|
BUFFER_TRACE(gd_bh, "get_write_access");
|
612 |
|
|
err = ext4_journal_get_write_access(handle, gd_bh);
|
613 |
|
|
if (err)
|
614 |
|
|
goto error_return;
|
615 |
|
|
|
616 |
|
|
jbd_lock_bh_state(bitmap_bh);
|
617 |
|
|
|
618 |
|
|
for (i = 0, group_freed = 0; i < count; i++) {
|
619 |
|
|
/*
|
620 |
|
|
* An HJ special. This is expensive...
|
621 |
|
|
*/
|
622 |
|
|
#ifdef CONFIG_JBD2_DEBUG
|
623 |
|
|
jbd_unlock_bh_state(bitmap_bh);
|
624 |
|
|
{
|
625 |
|
|
struct buffer_head *debug_bh;
|
626 |
|
|
debug_bh = sb_find_get_block(sb, block + i);
|
627 |
|
|
if (debug_bh) {
|
628 |
|
|
BUFFER_TRACE(debug_bh, "Deleted!");
|
629 |
|
|
if (!bh2jh(bitmap_bh)->b_committed_data)
|
630 |
|
|
BUFFER_TRACE(debug_bh,
|
631 |
|
|
"No commited data in bitmap");
|
632 |
|
|
BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
|
633 |
|
|
__brelse(debug_bh);
|
634 |
|
|
}
|
635 |
|
|
}
|
636 |
|
|
jbd_lock_bh_state(bitmap_bh);
|
637 |
|
|
#endif
|
638 |
|
|
if (need_resched()) {
|
639 |
|
|
jbd_unlock_bh_state(bitmap_bh);
|
640 |
|
|
cond_resched();
|
641 |
|
|
jbd_lock_bh_state(bitmap_bh);
|
642 |
|
|
}
|
643 |
|
|
/* @@@ This prevents newly-allocated data from being
|
644 |
|
|
* freed and then reallocated within the same
|
645 |
|
|
* transaction.
|
646 |
|
|
*
|
647 |
|
|
* Ideally we would want to allow that to happen, but to
|
648 |
|
|
* do so requires making jbd2_journal_forget() capable of
|
649 |
|
|
* revoking the queued write of a data block, which
|
650 |
|
|
* implies blocking on the journal lock. *forget()
|
651 |
|
|
* cannot block due to truncate races.
|
652 |
|
|
*
|
653 |
|
|
* Eventually we can fix this by making jbd2_journal_forget()
|
654 |
|
|
* return a status indicating whether or not it was able
|
655 |
|
|
* to revoke the buffer. On successful revoke, it is
|
656 |
|
|
* safe not to set the allocation bit in the committed
|
657 |
|
|
* bitmap, because we know that there is no outstanding
|
658 |
|
|
* activity on the buffer any more and so it is safe to
|
659 |
|
|
* reallocate it.
|
660 |
|
|
*/
|
661 |
|
|
BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
|
662 |
|
|
J_ASSERT_BH(bitmap_bh,
|
663 |
|
|
bh2jh(bitmap_bh)->b_committed_data != NULL);
|
664 |
|
|
ext4_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
|
665 |
|
|
bh2jh(bitmap_bh)->b_committed_data);
|
666 |
|
|
|
667 |
|
|
/*
|
668 |
|
|
* We clear the bit in the bitmap after setting the committed
|
669 |
|
|
* data bit, because this is the reverse order to that which
|
670 |
|
|
* the allocator uses.
|
671 |
|
|
*/
|
672 |
|
|
BUFFER_TRACE(bitmap_bh, "clear bit");
|
673 |
|
|
if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
|
674 |
|
|
bit + i, bitmap_bh->b_data)) {
|
675 |
|
|
jbd_unlock_bh_state(bitmap_bh);
|
676 |
|
|
ext4_error(sb, __FUNCTION__,
|
677 |
|
|
"bit already cleared for block %llu",
|
678 |
|
|
(ext4_fsblk_t)(block + i));
|
679 |
|
|
jbd_lock_bh_state(bitmap_bh);
|
680 |
|
|
BUFFER_TRACE(bitmap_bh, "bit already cleared");
|
681 |
|
|
} else {
|
682 |
|
|
group_freed++;
|
683 |
|
|
}
|
684 |
|
|
}
|
685 |
|
|
jbd_unlock_bh_state(bitmap_bh);
|
686 |
|
|
|
687 |
|
|
spin_lock(sb_bgl_lock(sbi, block_group));
|
688 |
|
|
desc->bg_free_blocks_count =
|
689 |
|
|
cpu_to_le16(le16_to_cpu(desc->bg_free_blocks_count) +
|
690 |
|
|
group_freed);
|
691 |
|
|
desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
|
692 |
|
|
spin_unlock(sb_bgl_lock(sbi, block_group));
|
693 |
|
|
percpu_counter_add(&sbi->s_freeblocks_counter, count);
|
694 |
|
|
|
695 |
|
|
/* We dirtied the bitmap block */
|
696 |
|
|
BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
|
697 |
|
|
err = ext4_journal_dirty_metadata(handle, bitmap_bh);
|
698 |
|
|
|
699 |
|
|
/* And the group descriptor block */
|
700 |
|
|
BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
|
701 |
|
|
ret = ext4_journal_dirty_metadata(handle, gd_bh);
|
702 |
|
|
if (!err) err = ret;
|
703 |
|
|
*pdquot_freed_blocks += group_freed;
|
704 |
|
|
|
705 |
|
|
if (overflow && !err) {
|
706 |
|
|
block += count;
|
707 |
|
|
count = overflow;
|
708 |
|
|
goto do_more;
|
709 |
|
|
}
|
710 |
|
|
sb->s_dirt = 1;
|
711 |
|
|
error_return:
|
712 |
|
|
brelse(bitmap_bh);
|
713 |
|
|
ext4_std_error(sb, err);
|
714 |
|
|
return;
|
715 |
|
|
}
|
716 |
|
|
|
717 |
|
|
/**
|
718 |
|
|
* ext4_free_blocks() -- Free given blocks and update quota
|
719 |
|
|
* @handle: handle for this transaction
|
720 |
|
|
* @inode: inode
|
721 |
|
|
* @block: start physical block to free
|
722 |
|
|
* @count: number of blocks to count
|
723 |
|
|
*/
|
724 |
|
|
void ext4_free_blocks(handle_t *handle, struct inode *inode,
|
725 |
|
|
ext4_fsblk_t block, unsigned long count)
|
726 |
|
|
{
|
727 |
|
|
struct super_block * sb;
|
728 |
|
|
unsigned long dquot_freed_blocks;
|
729 |
|
|
|
730 |
|
|
sb = inode->i_sb;
|
731 |
|
|
if (!sb) {
|
732 |
|
|
printk ("ext4_free_blocks: nonexistent device");
|
733 |
|
|
return;
|
734 |
|
|
}
|
735 |
|
|
ext4_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
|
736 |
|
|
if (dquot_freed_blocks)
|
737 |
|
|
DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
|
738 |
|
|
return;
|
739 |
|
|
}
|
740 |
|
|
|
741 |
|
|
/**
|
742 |
|
|
* ext4_test_allocatable()
|
743 |
|
|
* @nr: given allocation block group
|
744 |
|
|
* @bh: bufferhead contains the bitmap of the given block group
|
745 |
|
|
*
|
746 |
|
|
* For ext4 allocations, we must not reuse any blocks which are
|
747 |
|
|
* allocated in the bitmap buffer's "last committed data" copy. This
|
748 |
|
|
* prevents deletes from freeing up the page for reuse until we have
|
749 |
|
|
* committed the delete transaction.
|
750 |
|
|
*
|
751 |
|
|
* If we didn't do this, then deleting something and reallocating it as
|
752 |
|
|
* data would allow the old block to be overwritten before the
|
753 |
|
|
* transaction committed (because we force data to disk before commit).
|
754 |
|
|
* This would lead to corruption if we crashed between overwriting the
|
755 |
|
|
* data and committing the delete.
|
756 |
|
|
*
|
757 |
|
|
* @@@ We may want to make this allocation behaviour conditional on
|
758 |
|
|
* data-writes at some point, and disable it for metadata allocations or
|
759 |
|
|
* sync-data inodes.
|
760 |
|
|
*/
|
761 |
|
|
static int ext4_test_allocatable(ext4_grpblk_t nr, struct buffer_head *bh)
|
762 |
|
|
{
|
763 |
|
|
int ret;
|
764 |
|
|
struct journal_head *jh = bh2jh(bh);
|
765 |
|
|
|
766 |
|
|
if (ext4_test_bit(nr, bh->b_data))
|
767 |
|
|
return 0;
|
768 |
|
|
|
769 |
|
|
jbd_lock_bh_state(bh);
|
770 |
|
|
if (!jh->b_committed_data)
|
771 |
|
|
ret = 1;
|
772 |
|
|
else
|
773 |
|
|
ret = !ext4_test_bit(nr, jh->b_committed_data);
|
774 |
|
|
jbd_unlock_bh_state(bh);
|
775 |
|
|
return ret;
|
776 |
|
|
}
|
777 |
|
|
|
778 |
|
|
/**
|
779 |
|
|
* bitmap_search_next_usable_block()
|
780 |
|
|
* @start: the starting block (group relative) of the search
|
781 |
|
|
* @bh: bufferhead contains the block group bitmap
|
782 |
|
|
* @maxblocks: the ending block (group relative) of the reservation
|
783 |
|
|
*
|
784 |
|
|
* The bitmap search --- search forward alternately through the actual
|
785 |
|
|
* bitmap on disk and the last-committed copy in journal, until we find a
|
786 |
|
|
* bit free in both bitmaps.
|
787 |
|
|
*/
|
788 |
|
|
static ext4_grpblk_t
|
789 |
|
|
bitmap_search_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh,
|
790 |
|
|
ext4_grpblk_t maxblocks)
|
791 |
|
|
{
|
792 |
|
|
ext4_grpblk_t next;
|
793 |
|
|
struct journal_head *jh = bh2jh(bh);
|
794 |
|
|
|
795 |
|
|
while (start < maxblocks) {
|
796 |
|
|
next = ext4_find_next_zero_bit(bh->b_data, maxblocks, start);
|
797 |
|
|
if (next >= maxblocks)
|
798 |
|
|
return -1;
|
799 |
|
|
if (ext4_test_allocatable(next, bh))
|
800 |
|
|
return next;
|
801 |
|
|
jbd_lock_bh_state(bh);
|
802 |
|
|
if (jh->b_committed_data)
|
803 |
|
|
start = ext4_find_next_zero_bit(jh->b_committed_data,
|
804 |
|
|
maxblocks, next);
|
805 |
|
|
jbd_unlock_bh_state(bh);
|
806 |
|
|
}
|
807 |
|
|
return -1;
|
808 |
|
|
}
|
809 |
|
|
|
810 |
|
|
/**
|
811 |
|
|
* find_next_usable_block()
|
812 |
|
|
* @start: the starting block (group relative) to find next
|
813 |
|
|
* allocatable block in bitmap.
|
814 |
|
|
* @bh: bufferhead contains the block group bitmap
|
815 |
|
|
* @maxblocks: the ending block (group relative) for the search
|
816 |
|
|
*
|
817 |
|
|
* Find an allocatable block in a bitmap. We honor both the bitmap and
|
818 |
|
|
* its last-committed copy (if that exists), and perform the "most
|
819 |
|
|
* appropriate allocation" algorithm of looking for a free block near
|
820 |
|
|
* the initial goal; then for a free byte somewhere in the bitmap; then
|
821 |
|
|
* for any free bit in the bitmap.
|
822 |
|
|
*/
|
823 |
|
|
static ext4_grpblk_t
|
824 |
|
|
find_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh,
|
825 |
|
|
ext4_grpblk_t maxblocks)
|
826 |
|
|
{
|
827 |
|
|
ext4_grpblk_t here, next;
|
828 |
|
|
char *p, *r;
|
829 |
|
|
|
830 |
|
|
if (start > 0) {
|
831 |
|
|
/*
|
832 |
|
|
* The goal was occupied; search forward for a free
|
833 |
|
|
* block within the next XX blocks.
|
834 |
|
|
*
|
835 |
|
|
* end_goal is more or less random, but it has to be
|
836 |
|
|
* less than EXT4_BLOCKS_PER_GROUP. Aligning up to the
|
837 |
|
|
* next 64-bit boundary is simple..
|
838 |
|
|
*/
|
839 |
|
|
ext4_grpblk_t end_goal = (start + 63) & ~63;
|
840 |
|
|
if (end_goal > maxblocks)
|
841 |
|
|
end_goal = maxblocks;
|
842 |
|
|
here = ext4_find_next_zero_bit(bh->b_data, end_goal, start);
|
843 |
|
|
if (here < end_goal && ext4_test_allocatable(here, bh))
|
844 |
|
|
return here;
|
845 |
|
|
ext4_debug("Bit not found near goal\n");
|
846 |
|
|
}
|
847 |
|
|
|
848 |
|
|
here = start;
|
849 |
|
|
if (here < 0)
|
850 |
|
|
here = 0;
|
851 |
|
|
|
852 |
|
|
p = ((char *)bh->b_data) + (here >> 3);
|
853 |
|
|
r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
|
854 |
|
|
next = (r - ((char *)bh->b_data)) << 3;
|
855 |
|
|
|
856 |
|
|
if (next < maxblocks && next >= start && ext4_test_allocatable(next, bh))
|
857 |
|
|
return next;
|
858 |
|
|
|
859 |
|
|
/*
|
860 |
|
|
* The bitmap search --- search forward alternately through the actual
|
861 |
|
|
* bitmap and the last-committed copy until we find a bit free in
|
862 |
|
|
* both
|
863 |
|
|
*/
|
864 |
|
|
here = bitmap_search_next_usable_block(here, bh, maxblocks);
|
865 |
|
|
return here;
|
866 |
|
|
}
|
867 |
|
|
|
868 |
|
|
/**
|
869 |
|
|
* claim_block()
|
870 |
|
|
* @block: the free block (group relative) to allocate
|
871 |
|
|
* @bh: the bufferhead containts the block group bitmap
|
872 |
|
|
*
|
873 |
|
|
* We think we can allocate this block in this bitmap. Try to set the bit.
|
874 |
|
|
* If that succeeds then check that nobody has allocated and then freed the
|
875 |
|
|
* block since we saw that is was not marked in b_committed_data. If it _was_
|
876 |
|
|
* allocated and freed then clear the bit in the bitmap again and return
|
877 |
|
|
* zero (failure).
|
878 |
|
|
*/
|
879 |
|
|
static inline int
|
880 |
|
|
claim_block(spinlock_t *lock, ext4_grpblk_t block, struct buffer_head *bh)
|
881 |
|
|
{
|
882 |
|
|
struct journal_head *jh = bh2jh(bh);
|
883 |
|
|
int ret;
|
884 |
|
|
|
885 |
|
|
if (ext4_set_bit_atomic(lock, block, bh->b_data))
|
886 |
|
|
return 0;
|
887 |
|
|
jbd_lock_bh_state(bh);
|
888 |
|
|
if (jh->b_committed_data && ext4_test_bit(block,jh->b_committed_data)) {
|
889 |
|
|
ext4_clear_bit_atomic(lock, block, bh->b_data);
|
890 |
|
|
ret = 0;
|
891 |
|
|
} else {
|
892 |
|
|
ret = 1;
|
893 |
|
|
}
|
894 |
|
|
jbd_unlock_bh_state(bh);
|
895 |
|
|
return ret;
|
896 |
|
|
}
|
897 |
|
|
|
898 |
|
|
/**
|
899 |
|
|
* ext4_try_to_allocate()
|
900 |
|
|
* @sb: superblock
|
901 |
|
|
* @handle: handle to this transaction
|
902 |
|
|
* @group: given allocation block group
|
903 |
|
|
* @bitmap_bh: bufferhead holds the block bitmap
|
904 |
|
|
* @grp_goal: given target block within the group
|
905 |
|
|
* @count: target number of blocks to allocate
|
906 |
|
|
* @my_rsv: reservation window
|
907 |
|
|
*
|
908 |
|
|
* Attempt to allocate blocks within a give range. Set the range of allocation
|
909 |
|
|
* first, then find the first free bit(s) from the bitmap (within the range),
|
910 |
|
|
* and at last, allocate the blocks by claiming the found free bit as allocated.
|
911 |
|
|
*
|
912 |
|
|
* To set the range of this allocation:
|
913 |
|
|
* if there is a reservation window, only try to allocate block(s) from the
|
914 |
|
|
* file's own reservation window;
|
915 |
|
|
* Otherwise, the allocation range starts from the give goal block, ends at
|
916 |
|
|
* the block group's last block.
|
917 |
|
|
*
|
918 |
|
|
* If we failed to allocate the desired block then we may end up crossing to a
|
919 |
|
|
* new bitmap. In that case we must release write access to the old one via
|
920 |
|
|
* ext4_journal_release_buffer(), else we'll run out of credits.
|
921 |
|
|
*/
|
922 |
|
|
static ext4_grpblk_t
|
923 |
|
|
ext4_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
|
924 |
|
|
struct buffer_head *bitmap_bh, ext4_grpblk_t grp_goal,
|
925 |
|
|
unsigned long *count, struct ext4_reserve_window *my_rsv)
|
926 |
|
|
{
|
927 |
|
|
ext4_fsblk_t group_first_block;
|
928 |
|
|
ext4_grpblk_t start, end;
|
929 |
|
|
unsigned long num = 0;
|
930 |
|
|
|
931 |
|
|
/* we do allocation within the reservation window if we have a window */
|
932 |
|
|
if (my_rsv) {
|
933 |
|
|
group_first_block = ext4_group_first_block_no(sb, group);
|
934 |
|
|
if (my_rsv->_rsv_start >= group_first_block)
|
935 |
|
|
start = my_rsv->_rsv_start - group_first_block;
|
936 |
|
|
else
|
937 |
|
|
/* reservation window cross group boundary */
|
938 |
|
|
start = 0;
|
939 |
|
|
end = my_rsv->_rsv_end - group_first_block + 1;
|
940 |
|
|
if (end > EXT4_BLOCKS_PER_GROUP(sb))
|
941 |
|
|
/* reservation window crosses group boundary */
|
942 |
|
|
end = EXT4_BLOCKS_PER_GROUP(sb);
|
943 |
|
|
if ((start <= grp_goal) && (grp_goal < end))
|
944 |
|
|
start = grp_goal;
|
945 |
|
|
else
|
946 |
|
|
grp_goal = -1;
|
947 |
|
|
} else {
|
948 |
|
|
if (grp_goal > 0)
|
949 |
|
|
start = grp_goal;
|
950 |
|
|
else
|
951 |
|
|
start = 0;
|
952 |
|
|
end = EXT4_BLOCKS_PER_GROUP(sb);
|
953 |
|
|
}
|
954 |
|
|
|
955 |
|
|
BUG_ON(start > EXT4_BLOCKS_PER_GROUP(sb));
|
956 |
|
|
|
957 |
|
|
repeat:
|
958 |
|
|
if (grp_goal < 0 || !ext4_test_allocatable(grp_goal, bitmap_bh)) {
|
959 |
|
|
grp_goal = find_next_usable_block(start, bitmap_bh, end);
|
960 |
|
|
if (grp_goal < 0)
|
961 |
|
|
goto fail_access;
|
962 |
|
|
if (!my_rsv) {
|
963 |
|
|
int i;
|
964 |
|
|
|
965 |
|
|
for (i = 0; i < 7 && grp_goal > start &&
|
966 |
|
|
ext4_test_allocatable(grp_goal - 1,
|
967 |
|
|
bitmap_bh);
|
968 |
|
|
i++, grp_goal--)
|
969 |
|
|
;
|
970 |
|
|
}
|
971 |
|
|
}
|
972 |
|
|
start = grp_goal;
|
973 |
|
|
|
974 |
|
|
if (!claim_block(sb_bgl_lock(EXT4_SB(sb), group),
|
975 |
|
|
grp_goal, bitmap_bh)) {
|
976 |
|
|
/*
|
977 |
|
|
* The block was allocated by another thread, or it was
|
978 |
|
|
* allocated and then freed by another thread
|
979 |
|
|
*/
|
980 |
|
|
start++;
|
981 |
|
|
grp_goal++;
|
982 |
|
|
if (start >= end)
|
983 |
|
|
goto fail_access;
|
984 |
|
|
goto repeat;
|
985 |
|
|
}
|
986 |
|
|
num++;
|
987 |
|
|
grp_goal++;
|
988 |
|
|
while (num < *count && grp_goal < end
|
989 |
|
|
&& ext4_test_allocatable(grp_goal, bitmap_bh)
|
990 |
|
|
&& claim_block(sb_bgl_lock(EXT4_SB(sb), group),
|
991 |
|
|
grp_goal, bitmap_bh)) {
|
992 |
|
|
num++;
|
993 |
|
|
grp_goal++;
|
994 |
|
|
}
|
995 |
|
|
*count = num;
|
996 |
|
|
return grp_goal - num;
|
997 |
|
|
fail_access:
|
998 |
|
|
*count = num;
|
999 |
|
|
return -1;
|
1000 |
|
|
}
|
1001 |
|
|
|
1002 |
|
|
/**
|
1003 |
|
|
* find_next_reservable_window():
|
1004 |
|
|
* find a reservable space within the given range.
|
1005 |
|
|
* It does not allocate the reservation window for now:
|
1006 |
|
|
* alloc_new_reservation() will do the work later.
|
1007 |
|
|
*
|
1008 |
|
|
* @search_head: the head of the searching list;
|
1009 |
|
|
* This is not necessarily the list head of the whole filesystem
|
1010 |
|
|
*
|
1011 |
|
|
* We have both head and start_block to assist the search
|
1012 |
|
|
* for the reservable space. The list starts from head,
|
1013 |
|
|
* but we will shift to the place where start_block is,
|
1014 |
|
|
* then start from there, when looking for a reservable space.
|
1015 |
|
|
*
|
1016 |
|
|
* @size: the target new reservation window size
|
1017 |
|
|
*
|
1018 |
|
|
* @group_first_block: the first block we consider to start
|
1019 |
|
|
* the real search from
|
1020 |
|
|
*
|
1021 |
|
|
* @last_block:
|
1022 |
|
|
* the maximum block number that our goal reservable space
|
1023 |
|
|
* could start from. This is normally the last block in this
|
1024 |
|
|
* group. The search will end when we found the start of next
|
1025 |
|
|
* possible reservable space is out of this boundary.
|
1026 |
|
|
* This could handle the cross boundary reservation window
|
1027 |
|
|
* request.
|
1028 |
|
|
*
|
1029 |
|
|
* basically we search from the given range, rather than the whole
|
1030 |
|
|
* reservation double linked list, (start_block, last_block)
|
1031 |
|
|
* to find a free region that is of my size and has not
|
1032 |
|
|
* been reserved.
|
1033 |
|
|
*
|
1034 |
|
|
*/
|
1035 |
|
|
static int find_next_reservable_window(
|
1036 |
|
|
struct ext4_reserve_window_node *search_head,
|
1037 |
|
|
struct ext4_reserve_window_node *my_rsv,
|
1038 |
|
|
struct super_block * sb,
|
1039 |
|
|
ext4_fsblk_t start_block,
|
1040 |
|
|
ext4_fsblk_t last_block)
|
1041 |
|
|
{
|
1042 |
|
|
struct rb_node *next;
|
1043 |
|
|
struct ext4_reserve_window_node *rsv, *prev;
|
1044 |
|
|
ext4_fsblk_t cur;
|
1045 |
|
|
int size = my_rsv->rsv_goal_size;
|
1046 |
|
|
|
1047 |
|
|
/* TODO: make the start of the reservation window byte-aligned */
|
1048 |
|
|
/* cur = *start_block & ~7;*/
|
1049 |
|
|
cur = start_block;
|
1050 |
|
|
rsv = search_head;
|
1051 |
|
|
if (!rsv)
|
1052 |
|
|
return -1;
|
1053 |
|
|
|
1054 |
|
|
while (1) {
|
1055 |
|
|
if (cur <= rsv->rsv_end)
|
1056 |
|
|
cur = rsv->rsv_end + 1;
|
1057 |
|
|
|
1058 |
|
|
/* TODO?
|
1059 |
|
|
* in the case we could not find a reservable space
|
1060 |
|
|
* that is what is expected, during the re-search, we could
|
1061 |
|
|
* remember what's the largest reservable space we could have
|
1062 |
|
|
* and return that one.
|
1063 |
|
|
*
|
1064 |
|
|
* For now it will fail if we could not find the reservable
|
1065 |
|
|
* space with expected-size (or more)...
|
1066 |
|
|
*/
|
1067 |
|
|
if (cur > last_block)
|
1068 |
|
|
return -1; /* fail */
|
1069 |
|
|
|
1070 |
|
|
prev = rsv;
|
1071 |
|
|
next = rb_next(&rsv->rsv_node);
|
1072 |
|
|
rsv = rb_entry(next,struct ext4_reserve_window_node,rsv_node);
|
1073 |
|
|
|
1074 |
|
|
/*
|
1075 |
|
|
* Reached the last reservation, we can just append to the
|
1076 |
|
|
* previous one.
|
1077 |
|
|
*/
|
1078 |
|
|
if (!next)
|
1079 |
|
|
break;
|
1080 |
|
|
|
1081 |
|
|
if (cur + size <= rsv->rsv_start) {
|
1082 |
|
|
/*
|
1083 |
|
|
* Found a reserveable space big enough. We could
|
1084 |
|
|
* have a reservation across the group boundary here
|
1085 |
|
|
*/
|
1086 |
|
|
break;
|
1087 |
|
|
}
|
1088 |
|
|
}
|
1089 |
|
|
/*
|
1090 |
|
|
* we come here either :
|
1091 |
|
|
* when we reach the end of the whole list,
|
1092 |
|
|
* and there is empty reservable space after last entry in the list.
|
1093 |
|
|
* append it to the end of the list.
|
1094 |
|
|
*
|
1095 |
|
|
* or we found one reservable space in the middle of the list,
|
1096 |
|
|
* return the reservation window that we could append to.
|
1097 |
|
|
* succeed.
|
1098 |
|
|
*/
|
1099 |
|
|
|
1100 |
|
|
if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
|
1101 |
|
|
rsv_window_remove(sb, my_rsv);
|
1102 |
|
|
|
1103 |
|
|
/*
|
1104 |
|
|
* Let's book the whole avaliable window for now. We will check the
|
1105 |
|
|
* disk bitmap later and then, if there are free blocks then we adjust
|
1106 |
|
|
* the window size if it's larger than requested.
|
1107 |
|
|
* Otherwise, we will remove this node from the tree next time
|
1108 |
|
|
* call find_next_reservable_window.
|
1109 |
|
|
*/
|
1110 |
|
|
my_rsv->rsv_start = cur;
|
1111 |
|
|
my_rsv->rsv_end = cur + size - 1;
|
1112 |
|
|
my_rsv->rsv_alloc_hit = 0;
|
1113 |
|
|
|
1114 |
|
|
if (prev != my_rsv)
|
1115 |
|
|
ext4_rsv_window_add(sb, my_rsv);
|
1116 |
|
|
|
1117 |
|
|
return 0;
|
1118 |
|
|
}
|
1119 |
|
|
|
1120 |
|
|
/**
|
1121 |
|
|
* alloc_new_reservation()--allocate a new reservation window
|
1122 |
|
|
*
|
1123 |
|
|
* To make a new reservation, we search part of the filesystem
|
1124 |
|
|
* reservation list (the list that inside the group). We try to
|
1125 |
|
|
* allocate a new reservation window near the allocation goal,
|
1126 |
|
|
* or the beginning of the group, if there is no goal.
|
1127 |
|
|
*
|
1128 |
|
|
* We first find a reservable space after the goal, then from
|
1129 |
|
|
* there, we check the bitmap for the first free block after
|
1130 |
|
|
* it. If there is no free block until the end of group, then the
|
1131 |
|
|
* whole group is full, we failed. Otherwise, check if the free
|
1132 |
|
|
* block is inside the expected reservable space, if so, we
|
1133 |
|
|
* succeed.
|
1134 |
|
|
* If the first free block is outside the reservable space, then
|
1135 |
|
|
* start from the first free block, we search for next available
|
1136 |
|
|
* space, and go on.
|
1137 |
|
|
*
|
1138 |
|
|
* on succeed, a new reservation will be found and inserted into the list
|
1139 |
|
|
* It contains at least one free block, and it does not overlap with other
|
1140 |
|
|
* reservation windows.
|
1141 |
|
|
*
|
1142 |
|
|
* failed: we failed to find a reservation window in this group
|
1143 |
|
|
*
|
1144 |
|
|
* @rsv: the reservation
|
1145 |
|
|
*
|
1146 |
|
|
* @grp_goal: The goal (group-relative). It is where the search for a
|
1147 |
|
|
* free reservable space should start from.
|
1148 |
|
|
* if we have a grp_goal(grp_goal >0 ), then start from there,
|
1149 |
|
|
* no grp_goal(grp_goal = -1), we start from the first block
|
1150 |
|
|
* of the group.
|
1151 |
|
|
*
|
1152 |
|
|
* @sb: the super block
|
1153 |
|
|
* @group: the group we are trying to allocate in
|
1154 |
|
|
* @bitmap_bh: the block group block bitmap
|
1155 |
|
|
*
|
1156 |
|
|
*/
|
1157 |
|
|
static int alloc_new_reservation(struct ext4_reserve_window_node *my_rsv,
|
1158 |
|
|
ext4_grpblk_t grp_goal, struct super_block *sb,
|
1159 |
|
|
unsigned int group, struct buffer_head *bitmap_bh)
|
1160 |
|
|
{
|
1161 |
|
|
struct ext4_reserve_window_node *search_head;
|
1162 |
|
|
ext4_fsblk_t group_first_block, group_end_block, start_block;
|
1163 |
|
|
ext4_grpblk_t first_free_block;
|
1164 |
|
|
struct rb_root *fs_rsv_root = &EXT4_SB(sb)->s_rsv_window_root;
|
1165 |
|
|
unsigned long size;
|
1166 |
|
|
int ret;
|
1167 |
|
|
spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock;
|
1168 |
|
|
|
1169 |
|
|
group_first_block = ext4_group_first_block_no(sb, group);
|
1170 |
|
|
group_end_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1);
|
1171 |
|
|
|
1172 |
|
|
if (grp_goal < 0)
|
1173 |
|
|
start_block = group_first_block;
|
1174 |
|
|
else
|
1175 |
|
|
start_block = grp_goal + group_first_block;
|
1176 |
|
|
|
1177 |
|
|
size = my_rsv->rsv_goal_size;
|
1178 |
|
|
|
1179 |
|
|
if (!rsv_is_empty(&my_rsv->rsv_window)) {
|
1180 |
|
|
/*
|
1181 |
|
|
* if the old reservation is cross group boundary
|
1182 |
|
|
* and if the goal is inside the old reservation window,
|
1183 |
|
|
* we will come here when we just failed to allocate from
|
1184 |
|
|
* the first part of the window. We still have another part
|
1185 |
|
|
* that belongs to the next group. In this case, there is no
|
1186 |
|
|
* point to discard our window and try to allocate a new one
|
1187 |
|
|
* in this group(which will fail). we should
|
1188 |
|
|
* keep the reservation window, just simply move on.
|
1189 |
|
|
*
|
1190 |
|
|
* Maybe we could shift the start block of the reservation
|
1191 |
|
|
* window to the first block of next group.
|
1192 |
|
|
*/
|
1193 |
|
|
|
1194 |
|
|
if ((my_rsv->rsv_start <= group_end_block) &&
|
1195 |
|
|
(my_rsv->rsv_end > group_end_block) &&
|
1196 |
|
|
(start_block >= my_rsv->rsv_start))
|
1197 |
|
|
return -1;
|
1198 |
|
|
|
1199 |
|
|
if ((my_rsv->rsv_alloc_hit >
|
1200 |
|
|
(my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
|
1201 |
|
|
/*
|
1202 |
|
|
* if the previously allocation hit ratio is
|
1203 |
|
|
* greater than 1/2, then we double the size of
|
1204 |
|
|
* the reservation window the next time,
|
1205 |
|
|
* otherwise we keep the same size window
|
1206 |
|
|
*/
|
1207 |
|
|
size = size * 2;
|
1208 |
|
|
if (size > EXT4_MAX_RESERVE_BLOCKS)
|
1209 |
|
|
size = EXT4_MAX_RESERVE_BLOCKS;
|
1210 |
|
|
my_rsv->rsv_goal_size= size;
|
1211 |
|
|
}
|
1212 |
|
|
}
|
1213 |
|
|
|
1214 |
|
|
spin_lock(rsv_lock);
|
1215 |
|
|
/*
|
1216 |
|
|
* shift the search start to the window near the goal block
|
1217 |
|
|
*/
|
1218 |
|
|
search_head = search_reserve_window(fs_rsv_root, start_block);
|
1219 |
|
|
|
1220 |
|
|
/*
|
1221 |
|
|
* find_next_reservable_window() simply finds a reservable window
|
1222 |
|
|
* inside the given range(start_block, group_end_block).
|
1223 |
|
|
*
|
1224 |
|
|
* To make sure the reservation window has a free bit inside it, we
|
1225 |
|
|
* need to check the bitmap after we found a reservable window.
|
1226 |
|
|
*/
|
1227 |
|
|
retry:
|
1228 |
|
|
ret = find_next_reservable_window(search_head, my_rsv, sb,
|
1229 |
|
|
start_block, group_end_block);
|
1230 |
|
|
|
1231 |
|
|
if (ret == -1) {
|
1232 |
|
|
if (!rsv_is_empty(&my_rsv->rsv_window))
|
1233 |
|
|
rsv_window_remove(sb, my_rsv);
|
1234 |
|
|
spin_unlock(rsv_lock);
|
1235 |
|
|
return -1;
|
1236 |
|
|
}
|
1237 |
|
|
|
1238 |
|
|
/*
|
1239 |
|
|
* On success, find_next_reservable_window() returns the
|
1240 |
|
|
* reservation window where there is a reservable space after it.
|
1241 |
|
|
* Before we reserve this reservable space, we need
|
1242 |
|
|
* to make sure there is at least a free block inside this region.
|
1243 |
|
|
*
|
1244 |
|
|
* searching the first free bit on the block bitmap and copy of
|
1245 |
|
|
* last committed bitmap alternatively, until we found a allocatable
|
1246 |
|
|
* block. Search start from the start block of the reservable space
|
1247 |
|
|
* we just found.
|
1248 |
|
|
*/
|
1249 |
|
|
spin_unlock(rsv_lock);
|
1250 |
|
|
first_free_block = bitmap_search_next_usable_block(
|
1251 |
|
|
my_rsv->rsv_start - group_first_block,
|
1252 |
|
|
bitmap_bh, group_end_block - group_first_block + 1);
|
1253 |
|
|
|
1254 |
|
|
if (first_free_block < 0) {
|
1255 |
|
|
/*
|
1256 |
|
|
* no free block left on the bitmap, no point
|
1257 |
|
|
* to reserve the space. return failed.
|
1258 |
|
|
*/
|
1259 |
|
|
spin_lock(rsv_lock);
|
1260 |
|
|
if (!rsv_is_empty(&my_rsv->rsv_window))
|
1261 |
|
|
rsv_window_remove(sb, my_rsv);
|
1262 |
|
|
spin_unlock(rsv_lock);
|
1263 |
|
|
return -1; /* failed */
|
1264 |
|
|
}
|
1265 |
|
|
|
1266 |
|
|
start_block = first_free_block + group_first_block;
|
1267 |
|
|
/*
|
1268 |
|
|
* check if the first free block is within the
|
1269 |
|
|
* free space we just reserved
|
1270 |
|
|
*/
|
1271 |
|
|
if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
|
1272 |
|
|
return 0; /* success */
|
1273 |
|
|
/*
|
1274 |
|
|
* if the first free bit we found is out of the reservable space
|
1275 |
|
|
* continue search for next reservable space,
|
1276 |
|
|
* start from where the free block is,
|
1277 |
|
|
* we also shift the list head to where we stopped last time
|
1278 |
|
|
*/
|
1279 |
|
|
search_head = my_rsv;
|
1280 |
|
|
spin_lock(rsv_lock);
|
1281 |
|
|
goto retry;
|
1282 |
|
|
}
|
1283 |
|
|
|
1284 |
|
|
/**
|
1285 |
|
|
* try_to_extend_reservation()
|
1286 |
|
|
* @my_rsv: given reservation window
|
1287 |
|
|
* @sb: super block
|
1288 |
|
|
* @size: the delta to extend
|
1289 |
|
|
*
|
1290 |
|
|
* Attempt to expand the reservation window large enough to have
|
1291 |
|
|
* required number of free blocks
|
1292 |
|
|
*
|
1293 |
|
|
* Since ext4_try_to_allocate() will always allocate blocks within
|
1294 |
|
|
* the reservation window range, if the window size is too small,
|
1295 |
|
|
* multiple blocks allocation has to stop at the end of the reservation
|
1296 |
|
|
* window. To make this more efficient, given the total number of
|
1297 |
|
|
* blocks needed and the current size of the window, we try to
|
1298 |
|
|
* expand the reservation window size if necessary on a best-effort
|
1299 |
|
|
* basis before ext4_new_blocks() tries to allocate blocks,
|
1300 |
|
|
*/
|
1301 |
|
|
static void try_to_extend_reservation(struct ext4_reserve_window_node *my_rsv,
|
1302 |
|
|
struct super_block *sb, int size)
|
1303 |
|
|
{
|
1304 |
|
|
struct ext4_reserve_window_node *next_rsv;
|
1305 |
|
|
struct rb_node *next;
|
1306 |
|
|
spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock;
|
1307 |
|
|
|
1308 |
|
|
if (!spin_trylock(rsv_lock))
|
1309 |
|
|
return;
|
1310 |
|
|
|
1311 |
|
|
next = rb_next(&my_rsv->rsv_node);
|
1312 |
|
|
|
1313 |
|
|
if (!next)
|
1314 |
|
|
my_rsv->rsv_end += size;
|
1315 |
|
|
else {
|
1316 |
|
|
next_rsv = rb_entry(next, struct ext4_reserve_window_node, rsv_node);
|
1317 |
|
|
|
1318 |
|
|
if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
|
1319 |
|
|
my_rsv->rsv_end += size;
|
1320 |
|
|
else
|
1321 |
|
|
my_rsv->rsv_end = next_rsv->rsv_start - 1;
|
1322 |
|
|
}
|
1323 |
|
|
spin_unlock(rsv_lock);
|
1324 |
|
|
}
|
1325 |
|
|
|
1326 |
|
|
/**
|
1327 |
|
|
* ext4_try_to_allocate_with_rsv()
|
1328 |
|
|
* @sb: superblock
|
1329 |
|
|
* @handle: handle to this transaction
|
1330 |
|
|
* @group: given allocation block group
|
1331 |
|
|
* @bitmap_bh: bufferhead holds the block bitmap
|
1332 |
|
|
* @grp_goal: given target block within the group
|
1333 |
|
|
* @count: target number of blocks to allocate
|
1334 |
|
|
* @my_rsv: reservation window
|
1335 |
|
|
* @errp: pointer to store the error code
|
1336 |
|
|
*
|
1337 |
|
|
* This is the main function used to allocate a new block and its reservation
|
1338 |
|
|
* window.
|
1339 |
|
|
*
|
1340 |
|
|
* Each time when a new block allocation is need, first try to allocate from
|
1341 |
|
|
* its own reservation. If it does not have a reservation window, instead of
|
1342 |
|
|
* looking for a free bit on bitmap first, then look up the reservation list to
|
1343 |
|
|
* see if it is inside somebody else's reservation window, we try to allocate a
|
1344 |
|
|
* reservation window for it starting from the goal first. Then do the block
|
1345 |
|
|
* allocation within the reservation window.
|
1346 |
|
|
*
|
1347 |
|
|
* This will avoid keeping on searching the reservation list again and
|
1348 |
|
|
* again when somebody is looking for a free block (without
|
1349 |
|
|
* reservation), and there are lots of free blocks, but they are all
|
1350 |
|
|
* being reserved.
|
1351 |
|
|
*
|
1352 |
|
|
* We use a red-black tree for the per-filesystem reservation list.
|
1353 |
|
|
*
|
1354 |
|
|
*/
|
1355 |
|
|
static ext4_grpblk_t
|
1356 |
|
|
ext4_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
|
1357 |
|
|
unsigned int group, struct buffer_head *bitmap_bh,
|
1358 |
|
|
ext4_grpblk_t grp_goal,
|
1359 |
|
|
struct ext4_reserve_window_node * my_rsv,
|
1360 |
|
|
unsigned long *count, int *errp)
|
1361 |
|
|
{
|
1362 |
|
|
ext4_fsblk_t group_first_block, group_last_block;
|
1363 |
|
|
ext4_grpblk_t ret = 0;
|
1364 |
|
|
int fatal;
|
1365 |
|
|
unsigned long num = *count;
|
1366 |
|
|
|
1367 |
|
|
*errp = 0;
|
1368 |
|
|
|
1369 |
|
|
/*
|
1370 |
|
|
* Make sure we use undo access for the bitmap, because it is critical
|
1371 |
|
|
* that we do the frozen_data COW on bitmap buffers in all cases even
|
1372 |
|
|
* if the buffer is in BJ_Forget state in the committing transaction.
|
1373 |
|
|
*/
|
1374 |
|
|
BUFFER_TRACE(bitmap_bh, "get undo access for new block");
|
1375 |
|
|
fatal = ext4_journal_get_undo_access(handle, bitmap_bh);
|
1376 |
|
|
if (fatal) {
|
1377 |
|
|
*errp = fatal;
|
1378 |
|
|
return -1;
|
1379 |
|
|
}
|
1380 |
|
|
|
1381 |
|
|
/*
|
1382 |
|
|
* we don't deal with reservation when
|
1383 |
|
|
* filesystem is mounted without reservation
|
1384 |
|
|
* or the file is not a regular file
|
1385 |
|
|
* or last attempt to allocate a block with reservation turned on failed
|
1386 |
|
|
*/
|
1387 |
|
|
if (my_rsv == NULL ) {
|
1388 |
|
|
ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh,
|
1389 |
|
|
grp_goal, count, NULL);
|
1390 |
|
|
goto out;
|
1391 |
|
|
}
|
1392 |
|
|
/*
|
1393 |
|
|
* grp_goal is a group relative block number (if there is a goal)
|
1394 |
|
|
* 0 <= grp_goal < EXT4_BLOCKS_PER_GROUP(sb)
|
1395 |
|
|
* first block is a filesystem wide block number
|
1396 |
|
|
* first block is the block number of the first block in this group
|
1397 |
|
|
*/
|
1398 |
|
|
group_first_block = ext4_group_first_block_no(sb, group);
|
1399 |
|
|
group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1);
|
1400 |
|
|
|
1401 |
|
|
/*
|
1402 |
|
|
* Basically we will allocate a new block from inode's reservation
|
1403 |
|
|
* window.
|
1404 |
|
|
*
|
1405 |
|
|
* We need to allocate a new reservation window, if:
|
1406 |
|
|
* a) inode does not have a reservation window; or
|
1407 |
|
|
* b) last attempt to allocate a block from existing reservation
|
1408 |
|
|
* failed; or
|
1409 |
|
|
* c) we come here with a goal and with a reservation window
|
1410 |
|
|
*
|
1411 |
|
|
* We do not need to allocate a new reservation window if we come here
|
1412 |
|
|
* at the beginning with a goal and the goal is inside the window, or
|
1413 |
|
|
* we don't have a goal but already have a reservation window.
|
1414 |
|
|
* then we could go to allocate from the reservation window directly.
|
1415 |
|
|
*/
|
1416 |
|
|
while (1) {
|
1417 |
|
|
if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
|
1418 |
|
|
!goal_in_my_reservation(&my_rsv->rsv_window,
|
1419 |
|
|
grp_goal, group, sb)) {
|
1420 |
|
|
if (my_rsv->rsv_goal_size < *count)
|
1421 |
|
|
my_rsv->rsv_goal_size = *count;
|
1422 |
|
|
ret = alloc_new_reservation(my_rsv, grp_goal, sb,
|
1423 |
|
|
group, bitmap_bh);
|
1424 |
|
|
if (ret < 0)
|
1425 |
|
|
break; /* failed */
|
1426 |
|
|
|
1427 |
|
|
if (!goal_in_my_reservation(&my_rsv->rsv_window,
|
1428 |
|
|
grp_goal, group, sb))
|
1429 |
|
|
grp_goal = -1;
|
1430 |
|
|
} else if (grp_goal >= 0) {
|
1431 |
|
|
int curr = my_rsv->rsv_end -
|
1432 |
|
|
(grp_goal + group_first_block) + 1;
|
1433 |
|
|
|
1434 |
|
|
if (curr < *count)
|
1435 |
|
|
try_to_extend_reservation(my_rsv, sb,
|
1436 |
|
|
*count - curr);
|
1437 |
|
|
}
|
1438 |
|
|
|
1439 |
|
|
if ((my_rsv->rsv_start > group_last_block) ||
|
1440 |
|
|
(my_rsv->rsv_end < group_first_block)) {
|
1441 |
|
|
rsv_window_dump(&EXT4_SB(sb)->s_rsv_window_root, 1);
|
1442 |
|
|
BUG();
|
1443 |
|
|
}
|
1444 |
|
|
ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh,
|
1445 |
|
|
grp_goal, &num, &my_rsv->rsv_window);
|
1446 |
|
|
if (ret >= 0) {
|
1447 |
|
|
my_rsv->rsv_alloc_hit += num;
|
1448 |
|
|
*count = num;
|
1449 |
|
|
break; /* succeed */
|
1450 |
|
|
}
|
1451 |
|
|
num = *count;
|
1452 |
|
|
}
|
1453 |
|
|
out:
|
1454 |
|
|
if (ret >= 0) {
|
1455 |
|
|
BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
|
1456 |
|
|
"bitmap block");
|
1457 |
|
|
fatal = ext4_journal_dirty_metadata(handle, bitmap_bh);
|
1458 |
|
|
if (fatal) {
|
1459 |
|
|
*errp = fatal;
|
1460 |
|
|
return -1;
|
1461 |
|
|
}
|
1462 |
|
|
return ret;
|
1463 |
|
|
}
|
1464 |
|
|
|
1465 |
|
|
BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
|
1466 |
|
|
ext4_journal_release_buffer(handle, bitmap_bh);
|
1467 |
|
|
return ret;
|
1468 |
|
|
}
|
1469 |
|
|
|
1470 |
|
|
/**
|
1471 |
|
|
* ext4_has_free_blocks()
|
1472 |
|
|
* @sbi: in-core super block structure.
|
1473 |
|
|
*
|
1474 |
|
|
* Check if filesystem has at least 1 free block available for allocation.
|
1475 |
|
|
*/
|
1476 |
|
|
static int ext4_has_free_blocks(struct ext4_sb_info *sbi)
|
1477 |
|
|
{
|
1478 |
|
|
ext4_fsblk_t free_blocks, root_blocks;
|
1479 |
|
|
|
1480 |
|
|
free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
|
1481 |
|
|
root_blocks = ext4_r_blocks_count(sbi->s_es);
|
1482 |
|
|
if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
|
1483 |
|
|
sbi->s_resuid != current->fsuid &&
|
1484 |
|
|
(sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
|
1485 |
|
|
return 0;
|
1486 |
|
|
}
|
1487 |
|
|
return 1;
|
1488 |
|
|
}
|
1489 |
|
|
|
1490 |
|
|
/**
|
1491 |
|
|
* ext4_should_retry_alloc()
|
1492 |
|
|
* @sb: super block
|
1493 |
|
|
* @retries number of attemps has been made
|
1494 |
|
|
*
|
1495 |
|
|
* ext4_should_retry_alloc() is called when ENOSPC is returned, and if
|
1496 |
|
|
* it is profitable to retry the operation, this function will wait
|
1497 |
|
|
* for the current or commiting transaction to complete, and then
|
1498 |
|
|
* return TRUE.
|
1499 |
|
|
*
|
1500 |
|
|
* if the total number of retries exceed three times, return FALSE.
|
1501 |
|
|
*/
|
1502 |
|
|
int ext4_should_retry_alloc(struct super_block *sb, int *retries)
|
1503 |
|
|
{
|
1504 |
|
|
if (!ext4_has_free_blocks(EXT4_SB(sb)) || (*retries)++ > 3)
|
1505 |
|
|
return 0;
|
1506 |
|
|
|
1507 |
|
|
jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
|
1508 |
|
|
|
1509 |
|
|
return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
|
1510 |
|
|
}
|
1511 |
|
|
|
1512 |
|
|
/**
|
1513 |
|
|
* ext4_new_blocks() -- core block(s) allocation function
|
1514 |
|
|
* @handle: handle to this transaction
|
1515 |
|
|
* @inode: file inode
|
1516 |
|
|
* @goal: given target block(filesystem wide)
|
1517 |
|
|
* @count: target number of blocks to allocate
|
1518 |
|
|
* @errp: error code
|
1519 |
|
|
*
|
1520 |
|
|
* ext4_new_blocks uses a goal block to assist allocation. It tries to
|
1521 |
|
|
* allocate block(s) from the block group contains the goal block first. If that
|
1522 |
|
|
* fails, it will try to allocate block(s) from other block groups without
|
1523 |
|
|
* any specific goal block.
|
1524 |
|
|
*
|
1525 |
|
|
*/
|
1526 |
|
|
ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode,
|
1527 |
|
|
ext4_fsblk_t goal, unsigned long *count, int *errp)
|
1528 |
|
|
{
|
1529 |
|
|
struct buffer_head *bitmap_bh = NULL;
|
1530 |
|
|
struct buffer_head *gdp_bh;
|
1531 |
|
|
unsigned long group_no;
|
1532 |
|
|
int goal_group;
|
1533 |
|
|
ext4_grpblk_t grp_target_blk; /* blockgroup relative goal block */
|
1534 |
|
|
ext4_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
|
1535 |
|
|
ext4_fsblk_t ret_block; /* filesyetem-wide allocated block */
|
1536 |
|
|
int bgi; /* blockgroup iteration index */
|
1537 |
|
|
int fatal = 0, err;
|
1538 |
|
|
int performed_allocation = 0;
|
1539 |
|
|
ext4_grpblk_t free_blocks; /* number of free blocks in a group */
|
1540 |
|
|
struct super_block *sb;
|
1541 |
|
|
struct ext4_group_desc *gdp;
|
1542 |
|
|
struct ext4_super_block *es;
|
1543 |
|
|
struct ext4_sb_info *sbi;
|
1544 |
|
|
struct ext4_reserve_window_node *my_rsv = NULL;
|
1545 |
|
|
struct ext4_block_alloc_info *block_i;
|
1546 |
|
|
unsigned short windowsz = 0;
|
1547 |
|
|
#ifdef EXT4FS_DEBUG
|
1548 |
|
|
static int goal_hits, goal_attempts;
|
1549 |
|
|
#endif
|
1550 |
|
|
unsigned long ngroups;
|
1551 |
|
|
unsigned long num = *count;
|
1552 |
|
|
|
1553 |
|
|
*errp = -ENOSPC;
|
1554 |
|
|
sb = inode->i_sb;
|
1555 |
|
|
if (!sb) {
|
1556 |
|
|
printk("ext4_new_block: nonexistent device");
|
1557 |
|
|
return 0;
|
1558 |
|
|
}
|
1559 |
|
|
|
1560 |
|
|
/*
|
1561 |
|
|
* Check quota for allocation of this block.
|
1562 |
|
|
*/
|
1563 |
|
|
if (DQUOT_ALLOC_BLOCK(inode, num)) {
|
1564 |
|
|
*errp = -EDQUOT;
|
1565 |
|
|
return 0;
|
1566 |
|
|
}
|
1567 |
|
|
|
1568 |
|
|
sbi = EXT4_SB(sb);
|
1569 |
|
|
es = EXT4_SB(sb)->s_es;
|
1570 |
|
|
ext4_debug("goal=%lu.\n", goal);
|
1571 |
|
|
/*
|
1572 |
|
|
* Allocate a block from reservation only when
|
1573 |
|
|
* filesystem is mounted with reservation(default,-o reservation), and
|
1574 |
|
|
* it's a regular file, and
|
1575 |
|
|
* the desired window size is greater than 0 (One could use ioctl
|
1576 |
|
|
* command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off
|
1577 |
|
|
* reservation on that particular file)
|
1578 |
|
|
*/
|
1579 |
|
|
block_i = EXT4_I(inode)->i_block_alloc_info;
|
1580 |
|
|
if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
|
1581 |
|
|
my_rsv = &block_i->rsv_window_node;
|
1582 |
|
|
|
1583 |
|
|
if (!ext4_has_free_blocks(sbi)) {
|
1584 |
|
|
*errp = -ENOSPC;
|
1585 |
|
|
goto out;
|
1586 |
|
|
}
|
1587 |
|
|
|
1588 |
|
|
/*
|
1589 |
|
|
* First, test whether the goal block is free.
|
1590 |
|
|
*/
|
1591 |
|
|
if (goal < le32_to_cpu(es->s_first_data_block) ||
|
1592 |
|
|
goal >= ext4_blocks_count(es))
|
1593 |
|
|
goal = le32_to_cpu(es->s_first_data_block);
|
1594 |
|
|
ext4_get_group_no_and_offset(sb, goal, &group_no, &grp_target_blk);
|
1595 |
|
|
goal_group = group_no;
|
1596 |
|
|
retry_alloc:
|
1597 |
|
|
gdp = ext4_get_group_desc(sb, group_no, &gdp_bh);
|
1598 |
|
|
if (!gdp)
|
1599 |
|
|
goto io_error;
|
1600 |
|
|
|
1601 |
|
|
free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
|
1602 |
|
|
/*
|
1603 |
|
|
* if there is not enough free blocks to make a new resevation
|
1604 |
|
|
* turn off reservation for this allocation
|
1605 |
|
|
*/
|
1606 |
|
|
if (my_rsv && (free_blocks < windowsz)
|
1607 |
|
|
&& (rsv_is_empty(&my_rsv->rsv_window)))
|
1608 |
|
|
my_rsv = NULL;
|
1609 |
|
|
|
1610 |
|
|
if (free_blocks > 0) {
|
1611 |
|
|
bitmap_bh = read_block_bitmap(sb, group_no);
|
1612 |
|
|
if (!bitmap_bh)
|
1613 |
|
|
goto io_error;
|
1614 |
|
|
grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle,
|
1615 |
|
|
group_no, bitmap_bh, grp_target_blk,
|
1616 |
|
|
my_rsv, &num, &fatal);
|
1617 |
|
|
if (fatal)
|
1618 |
|
|
goto out;
|
1619 |
|
|
if (grp_alloc_blk >= 0)
|
1620 |
|
|
goto allocated;
|
1621 |
|
|
}
|
1622 |
|
|
|
1623 |
|
|
ngroups = EXT4_SB(sb)->s_groups_count;
|
1624 |
|
|
smp_rmb();
|
1625 |
|
|
|
1626 |
|
|
/*
|
1627 |
|
|
* Now search the rest of the groups. We assume that
|
1628 |
|
|
* i and gdp correctly point to the last group visited.
|
1629 |
|
|
*/
|
1630 |
|
|
for (bgi = 0; bgi < ngroups; bgi++) {
|
1631 |
|
|
group_no++;
|
1632 |
|
|
if (group_no >= ngroups)
|
1633 |
|
|
group_no = 0;
|
1634 |
|
|
gdp = ext4_get_group_desc(sb, group_no, &gdp_bh);
|
1635 |
|
|
if (!gdp)
|
1636 |
|
|
goto io_error;
|
1637 |
|
|
free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
|
1638 |
|
|
/*
|
1639 |
|
|
* skip this group if the number of
|
1640 |
|
|
* free blocks is less than half of the reservation
|
1641 |
|
|
* window size.
|
1642 |
|
|
*/
|
1643 |
|
|
if (free_blocks <= (windowsz/2))
|
1644 |
|
|
continue;
|
1645 |
|
|
|
1646 |
|
|
brelse(bitmap_bh);
|
1647 |
|
|
bitmap_bh = read_block_bitmap(sb, group_no);
|
1648 |
|
|
if (!bitmap_bh)
|
1649 |
|
|
goto io_error;
|
1650 |
|
|
/*
|
1651 |
|
|
* try to allocate block(s) from this group, without a goal(-1).
|
1652 |
|
|
*/
|
1653 |
|
|
grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle,
|
1654 |
|
|
group_no, bitmap_bh, -1, my_rsv,
|
1655 |
|
|
&num, &fatal);
|
1656 |
|
|
if (fatal)
|
1657 |
|
|
goto out;
|
1658 |
|
|
if (grp_alloc_blk >= 0)
|
1659 |
|
|
goto allocated;
|
1660 |
|
|
}
|
1661 |
|
|
/*
|
1662 |
|
|
* We may end up a bogus ealier ENOSPC error due to
|
1663 |
|
|
* filesystem is "full" of reservations, but
|
1664 |
|
|
* there maybe indeed free blocks avaliable on disk
|
1665 |
|
|
* In this case, we just forget about the reservations
|
1666 |
|
|
* just do block allocation as without reservations.
|
1667 |
|
|
*/
|
1668 |
|
|
if (my_rsv) {
|
1669 |
|
|
my_rsv = NULL;
|
1670 |
|
|
windowsz = 0;
|
1671 |
|
|
group_no = goal_group;
|
1672 |
|
|
goto retry_alloc;
|
1673 |
|
|
}
|
1674 |
|
|
/* No space left on the device */
|
1675 |
|
|
*errp = -ENOSPC;
|
1676 |
|
|
goto out;
|
1677 |
|
|
|
1678 |
|
|
allocated:
|
1679 |
|
|
|
1680 |
|
|
ext4_debug("using block group %d(%d)\n",
|
1681 |
|
|
group_no, gdp->bg_free_blocks_count);
|
1682 |
|
|
|
1683 |
|
|
BUFFER_TRACE(gdp_bh, "get_write_access");
|
1684 |
|
|
fatal = ext4_journal_get_write_access(handle, gdp_bh);
|
1685 |
|
|
if (fatal)
|
1686 |
|
|
goto out;
|
1687 |
|
|
|
1688 |
|
|
ret_block = grp_alloc_blk + ext4_group_first_block_no(sb, group_no);
|
1689 |
|
|
|
1690 |
|
|
if (in_range(ext4_block_bitmap(sb, gdp), ret_block, num) ||
|
1691 |
|
|
in_range(ext4_inode_bitmap(sb, gdp), ret_block, num) ||
|
1692 |
|
|
in_range(ret_block, ext4_inode_table(sb, gdp),
|
1693 |
|
|
EXT4_SB(sb)->s_itb_per_group) ||
|
1694 |
|
|
in_range(ret_block + num - 1, ext4_inode_table(sb, gdp),
|
1695 |
|
|
EXT4_SB(sb)->s_itb_per_group))
|
1696 |
|
|
ext4_error(sb, "ext4_new_block",
|
1697 |
|
|
"Allocating block in system zone - "
|
1698 |
|
|
"blocks from %llu, length %lu",
|
1699 |
|
|
ret_block, num);
|
1700 |
|
|
|
1701 |
|
|
performed_allocation = 1;
|
1702 |
|
|
|
1703 |
|
|
#ifdef CONFIG_JBD2_DEBUG
|
1704 |
|
|
{
|
1705 |
|
|
struct buffer_head *debug_bh;
|
1706 |
|
|
|
1707 |
|
|
/* Record bitmap buffer state in the newly allocated block */
|
1708 |
|
|
debug_bh = sb_find_get_block(sb, ret_block);
|
1709 |
|
|
if (debug_bh) {
|
1710 |
|
|
BUFFER_TRACE(debug_bh, "state when allocated");
|
1711 |
|
|
BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
|
1712 |
|
|
brelse(debug_bh);
|
1713 |
|
|
}
|
1714 |
|
|
}
|
1715 |
|
|
jbd_lock_bh_state(bitmap_bh);
|
1716 |
|
|
spin_lock(sb_bgl_lock(sbi, group_no));
|
1717 |
|
|
if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
|
1718 |
|
|
int i;
|
1719 |
|
|
|
1720 |
|
|
for (i = 0; i < num; i++) {
|
1721 |
|
|
if (ext4_test_bit(grp_alloc_blk+i,
|
1722 |
|
|
bh2jh(bitmap_bh)->b_committed_data)) {
|
1723 |
|
|
printk("%s: block was unexpectedly set in "
|
1724 |
|
|
"b_committed_data\n", __FUNCTION__);
|
1725 |
|
|
}
|
1726 |
|
|
}
|
1727 |
|
|
}
|
1728 |
|
|
ext4_debug("found bit %d\n", grp_alloc_blk);
|
1729 |
|
|
spin_unlock(sb_bgl_lock(sbi, group_no));
|
1730 |
|
|
jbd_unlock_bh_state(bitmap_bh);
|
1731 |
|
|
#endif
|
1732 |
|
|
|
1733 |
|
|
if (ret_block + num - 1 >= ext4_blocks_count(es)) {
|
1734 |
|
|
ext4_error(sb, "ext4_new_block",
|
1735 |
|
|
"block(%llu) >= blocks count(%llu) - "
|
1736 |
|
|
"block_group = %lu, es == %p ", ret_block,
|
1737 |
|
|
ext4_blocks_count(es), group_no, es);
|
1738 |
|
|
goto out;
|
1739 |
|
|
}
|
1740 |
|
|
|
1741 |
|
|
/*
|
1742 |
|
|
* It is up to the caller to add the new buffer to a journal
|
1743 |
|
|
* list of some description. We don't know in advance whether
|
1744 |
|
|
* the caller wants to use it as metadata or data.
|
1745 |
|
|
*/
|
1746 |
|
|
ext4_debug("allocating block %lu. Goal hits %d of %d.\n",
|
1747 |
|
|
ret_block, goal_hits, goal_attempts);
|
1748 |
|
|
|
1749 |
|
|
spin_lock(sb_bgl_lock(sbi, group_no));
|
1750 |
|
|
if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))
|
1751 |
|
|
gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
|
1752 |
|
|
gdp->bg_free_blocks_count =
|
1753 |
|
|
cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count)-num);
|
1754 |
|
|
gdp->bg_checksum = ext4_group_desc_csum(sbi, group_no, gdp);
|
1755 |
|
|
spin_unlock(sb_bgl_lock(sbi, group_no));
|
1756 |
|
|
percpu_counter_sub(&sbi->s_freeblocks_counter, num);
|
1757 |
|
|
|
1758 |
|
|
BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
|
1759 |
|
|
err = ext4_journal_dirty_metadata(handle, gdp_bh);
|
1760 |
|
|
if (!fatal)
|
1761 |
|
|
fatal = err;
|
1762 |
|
|
|
1763 |
|
|
sb->s_dirt = 1;
|
1764 |
|
|
if (fatal)
|
1765 |
|
|
goto out;
|
1766 |
|
|
|
1767 |
|
|
*errp = 0;
|
1768 |
|
|
brelse(bitmap_bh);
|
1769 |
|
|
DQUOT_FREE_BLOCK(inode, *count-num);
|
1770 |
|
|
*count = num;
|
1771 |
|
|
return ret_block;
|
1772 |
|
|
|
1773 |
|
|
io_error:
|
1774 |
|
|
*errp = -EIO;
|
1775 |
|
|
out:
|
1776 |
|
|
if (fatal) {
|
1777 |
|
|
*errp = fatal;
|
1778 |
|
|
ext4_std_error(sb, fatal);
|
1779 |
|
|
}
|
1780 |
|
|
/*
|
1781 |
|
|
* Undo the block allocation
|
1782 |
|
|
*/
|
1783 |
|
|
if (!performed_allocation)
|
1784 |
|
|
DQUOT_FREE_BLOCK(inode, *count);
|
1785 |
|
|
brelse(bitmap_bh);
|
1786 |
|
|
return 0;
|
1787 |
|
|
}
|
1788 |
|
|
|
1789 |
|
|
ext4_fsblk_t ext4_new_block(handle_t *handle, struct inode *inode,
|
1790 |
|
|
ext4_fsblk_t goal, int *errp)
|
1791 |
|
|
{
|
1792 |
|
|
unsigned long count = 1;
|
1793 |
|
|
|
1794 |
|
|
return ext4_new_blocks(handle, inode, goal, &count, errp);
|
1795 |
|
|
}
|
1796 |
|
|
|
1797 |
|
|
/**
|
1798 |
|
|
* ext4_count_free_blocks() -- count filesystem free blocks
|
1799 |
|
|
* @sb: superblock
|
1800 |
|
|
*
|
1801 |
|
|
* Adds up the number of free blocks from each block group.
|
1802 |
|
|
*/
|
1803 |
|
|
ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
|
1804 |
|
|
{
|
1805 |
|
|
ext4_fsblk_t desc_count;
|
1806 |
|
|
struct ext4_group_desc *gdp;
|
1807 |
|
|
int i;
|
1808 |
|
|
unsigned long ngroups = EXT4_SB(sb)->s_groups_count;
|
1809 |
|
|
#ifdef EXT4FS_DEBUG
|
1810 |
|
|
struct ext4_super_block *es;
|
1811 |
|
|
ext4_fsblk_t bitmap_count;
|
1812 |
|
|
unsigned long x;
|
1813 |
|
|
struct buffer_head *bitmap_bh = NULL;
|
1814 |
|
|
|
1815 |
|
|
es = EXT4_SB(sb)->s_es;
|
1816 |
|
|
desc_count = 0;
|
1817 |
|
|
bitmap_count = 0;
|
1818 |
|
|
gdp = NULL;
|
1819 |
|
|
|
1820 |
|
|
smp_rmb();
|
1821 |
|
|
for (i = 0; i < ngroups; i++) {
|
1822 |
|
|
gdp = ext4_get_group_desc(sb, i, NULL);
|
1823 |
|
|
if (!gdp)
|
1824 |
|
|
continue;
|
1825 |
|
|
desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
|
1826 |
|
|
brelse(bitmap_bh);
|
1827 |
|
|
bitmap_bh = read_block_bitmap(sb, i);
|
1828 |
|
|
if (bitmap_bh == NULL)
|
1829 |
|
|
continue;
|
1830 |
|
|
|
1831 |
|
|
x = ext4_count_free(bitmap_bh, sb->s_blocksize);
|
1832 |
|
|
printk("group %d: stored = %d, counted = %lu\n",
|
1833 |
|
|
i, le16_to_cpu(gdp->bg_free_blocks_count), x);
|
1834 |
|
|
bitmap_count += x;
|
1835 |
|
|
}
|
1836 |
|
|
brelse(bitmap_bh);
|
1837 |
|
|
printk("ext4_count_free_blocks: stored = %llu"
|
1838 |
|
|
", computed = %llu, %llu\n",
|
1839 |
|
|
EXT4_FREE_BLOCKS_COUNT(es),
|
1840 |
|
|
desc_count, bitmap_count);
|
1841 |
|
|
return bitmap_count;
|
1842 |
|
|
#else
|
1843 |
|
|
desc_count = 0;
|
1844 |
|
|
smp_rmb();
|
1845 |
|
|
for (i = 0; i < ngroups; i++) {
|
1846 |
|
|
gdp = ext4_get_group_desc(sb, i, NULL);
|
1847 |
|
|
if (!gdp)
|
1848 |
|
|
continue;
|
1849 |
|
|
desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
|
1850 |
|
|
}
|
1851 |
|
|
|
1852 |
|
|
return desc_count;
|
1853 |
|
|
#endif
|
1854 |
|
|
}
|
1855 |
|
|
|
1856 |
|
|
static inline int test_root(int a, int b)
|
1857 |
|
|
{
|
1858 |
|
|
int num = b;
|
1859 |
|
|
|
1860 |
|
|
while (a > num)
|
1861 |
|
|
num *= b;
|
1862 |
|
|
return num == a;
|
1863 |
|
|
}
|
1864 |
|
|
|
1865 |
|
|
static int ext4_group_sparse(int group)
|
1866 |
|
|
{
|
1867 |
|
|
if (group <= 1)
|
1868 |
|
|
return 1;
|
1869 |
|
|
if (!(group & 1))
|
1870 |
|
|
return 0;
|
1871 |
|
|
return (test_root(group, 7) || test_root(group, 5) ||
|
1872 |
|
|
test_root(group, 3));
|
1873 |
|
|
}
|
1874 |
|
|
|
1875 |
|
|
/**
|
1876 |
|
|
* ext4_bg_has_super - number of blocks used by the superblock in group
|
1877 |
|
|
* @sb: superblock for filesystem
|
1878 |
|
|
* @group: group number to check
|
1879 |
|
|
*
|
1880 |
|
|
* Return the number of blocks used by the superblock (primary or backup)
|
1881 |
|
|
* in this group. Currently this will be only 0 or 1.
|
1882 |
|
|
*/
|
1883 |
|
|
int ext4_bg_has_super(struct super_block *sb, int group)
|
1884 |
|
|
{
|
1885 |
|
|
if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
|
1886 |
|
|
EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
|
1887 |
|
|
!ext4_group_sparse(group))
|
1888 |
|
|
return 0;
|
1889 |
|
|
return 1;
|
1890 |
|
|
}
|
1891 |
|
|
|
1892 |
|
|
static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, int group)
|
1893 |
|
|
{
|
1894 |
|
|
unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
|
1895 |
|
|
unsigned long first = metagroup * EXT4_DESC_PER_BLOCK(sb);
|
1896 |
|
|
unsigned long last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
|
1897 |
|
|
|
1898 |
|
|
if (group == first || group == first + 1 || group == last)
|
1899 |
|
|
return 1;
|
1900 |
|
|
return 0;
|
1901 |
|
|
}
|
1902 |
|
|
|
1903 |
|
|
static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, int group)
|
1904 |
|
|
{
|
1905 |
|
|
if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
|
1906 |
|
|
EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
|
1907 |
|
|
!ext4_group_sparse(group))
|
1908 |
|
|
return 0;
|
1909 |
|
|
return EXT4_SB(sb)->s_gdb_count;
|
1910 |
|
|
}
|
1911 |
|
|
|
1912 |
|
|
/**
|
1913 |
|
|
* ext4_bg_num_gdb - number of blocks used by the group table in group
|
1914 |
|
|
* @sb: superblock for filesystem
|
1915 |
|
|
* @group: group number to check
|
1916 |
|
|
*
|
1917 |
|
|
* Return the number of blocks used by the group descriptor table
|
1918 |
|
|
* (primary or backup) in this group. In the future there may be a
|
1919 |
|
|
* different number of descriptor blocks in each group.
|
1920 |
|
|
*/
|
1921 |
|
|
unsigned long ext4_bg_num_gdb(struct super_block *sb, int group)
|
1922 |
|
|
{
|
1923 |
|
|
unsigned long first_meta_bg =
|
1924 |
|
|
le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
|
1925 |
|
|
unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
|
1926 |
|
|
|
1927 |
|
|
if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
|
1928 |
|
|
metagroup < first_meta_bg)
|
1929 |
|
|
return ext4_bg_num_gdb_nometa(sb,group);
|
1930 |
|
|
|
1931 |
|
|
return ext4_bg_num_gdb_meta(sb,group);
|
1932 |
|
|
|
1933 |
|
|
}
|