Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README | |
3 | */ | |
4 | /* Reiserfs block (de)allocator, bitmap-based. */ | |
5 | ||
1da177e4 LT |
6 | #include <linux/time.h> |
7 | #include <linux/reiserfs_fs.h> | |
8 | #include <linux/errno.h> | |
9 | #include <linux/buffer_head.h> | |
10 | #include <linux/kernel.h> | |
11 | #include <linux/pagemap.h> | |
6f01046b | 12 | #include <linux/vmalloc.h> |
1da177e4 LT |
13 | #include <linux/reiserfs_fs_sb.h> |
14 | #include <linux/reiserfs_fs_i.h> | |
15 | #include <linux/quotaops.h> | |
c3aa0776 | 16 | #include <linux/seq_file.h> |
1da177e4 LT |
17 | |
18 | #define PREALLOCATION_SIZE 9 | |
19 | ||
20 | /* different reiserfs block allocator options */ | |
21 | ||
22 | #define SB_ALLOC_OPTS(s) (REISERFS_SB(s)->s_alloc_options.bits) | |
23 | ||
24 | #define _ALLOC_concentrating_formatted_nodes 0 | |
25 | #define _ALLOC_displacing_large_files 1 | |
26 | #define _ALLOC_displacing_new_packing_localities 2 | |
27 | #define _ALLOC_old_hashed_relocation 3 | |
28 | #define _ALLOC_new_hashed_relocation 4 | |
29 | #define _ALLOC_skip_busy 5 | |
30 | #define _ALLOC_displace_based_on_dirid 6 | |
31 | #define _ALLOC_hashed_formatted_nodes 7 | |
32 | #define _ALLOC_old_way 8 | |
33 | #define _ALLOC_hundredth_slices 9 | |
34 | #define _ALLOC_dirid_groups 10 | |
35 | #define _ALLOC_oid_groups 11 | |
36 | #define _ALLOC_packing_groups 12 | |
37 | ||
38 | #define concentrating_formatted_nodes(s) test_bit(_ALLOC_concentrating_formatted_nodes, &SB_ALLOC_OPTS(s)) | |
39 | #define displacing_large_files(s) test_bit(_ALLOC_displacing_large_files, &SB_ALLOC_OPTS(s)) | |
40 | #define displacing_new_packing_localities(s) test_bit(_ALLOC_displacing_new_packing_localities, &SB_ALLOC_OPTS(s)) | |
41 | ||
42 | #define SET_OPTION(optname) \ | |
43 | do { \ | |
1d889d99 JM |
44 | reiserfs_info(s, "block allocator option \"%s\" is set", #optname); \ |
45 | set_bit(_ALLOC_ ## optname , &SB_ALLOC_OPTS(s)); \ | |
1da177e4 LT |
46 | } while(0) |
47 | #define TEST_OPTION(optname, s) \ | |
48 | test_bit(_ALLOC_ ## optname , &SB_ALLOC_OPTS(s)) | |
49 | ||
bd4c625c | 50 | static inline void get_bit_address(struct super_block *s, |
3ee16670 JM |
51 | b_blocknr_t block, |
52 | unsigned int *bmap_nr, | |
53 | unsigned int *offset) | |
1da177e4 | 54 | { |
bd4c625c LT |
55 | /* It is in the bitmap block number equal to the block |
56 | * number divided by the number of bits in a block. */ | |
e1fabd3c | 57 | *bmap_nr = block >> (s->s_blocksize_bits + 3); |
bd4c625c LT |
58 | /* Within that bitmap block it is located at bit offset *offset. */ |
59 | *offset = block & ((s->s_blocksize << 3) - 1); | |
1da177e4 LT |
60 | } |
61 | ||
bd4c625c | 62 | int is_reusable(struct super_block *s, b_blocknr_t block, int bit_value) |
1da177e4 | 63 | { |
3ee16670 | 64 | unsigned int bmap, offset; |
cb680c1b | 65 | unsigned int bmap_count = reiserfs_bmap_count(s); |
1da177e4 | 66 | |
bd4c625c | 67 | if (block == 0 || block >= SB_BLOCK_COUNT(s)) { |
0030b645 JM |
68 | reiserfs_error(s, "vs-4010", |
69 | "block number is out of range %lu (%u)", | |
70 | block, SB_BLOCK_COUNT(s)); | |
bd4c625c | 71 | return 0; |
1da177e4 | 72 | } |
1da177e4 | 73 | |
e1fabd3c JM |
74 | get_bit_address(s, block, &bmap, &offset); |
75 | ||
76 | /* Old format filesystem? Unlikely, but the bitmaps are all up front so | |
77 | * we need to account for it. */ | |
78 | if (unlikely(test_bit(REISERFS_OLD_FORMAT, | |
79 | &(REISERFS_SB(s)->s_properties)))) { | |
80 | b_blocknr_t bmap1 = REISERFS_SB(s)->s_sbh->b_blocknr + 1; | |
cb680c1b JM |
81 | if (block >= bmap1 && |
82 | block <= bmap1 + bmap_count) { | |
0030b645 JM |
83 | reiserfs_error(s, "vs-4019", "bitmap block %lu(%u) " |
84 | "can't be freed or reused", | |
85 | block, bmap_count); | |
e1fabd3c JM |
86 | return 0; |
87 | } | |
88 | } else { | |
89 | if (offset == 0) { | |
0030b645 JM |
90 | reiserfs_error(s, "vs-4020", "bitmap block %lu(%u) " |
91 | "can't be freed or reused", | |
92 | block, bmap_count); | |
bd4c625c LT |
93 | return 0; |
94 | } | |
e1fabd3c | 95 | } |
1da177e4 | 96 | |
cb680c1b | 97 | if (bmap >= bmap_count) { |
0030b645 JM |
98 | reiserfs_error(s, "vs-4030", "bitmap for requested block " |
99 | "is out of range: block=%lu, bitmap_nr=%u", | |
100 | block, bmap); | |
bd4c625c LT |
101 | return 0; |
102 | } | |
1da177e4 | 103 | |
bd4c625c | 104 | if (bit_value == 0 && block == SB_ROOT_BLOCK(s)) { |
0030b645 JM |
105 | reiserfs_error(s, "vs-4050", "this is root block (%u), " |
106 | "it must be busy", SB_ROOT_BLOCK(s)); | |
bd4c625c LT |
107 | return 0; |
108 | } | |
109 | ||
110 | return 1; | |
1da177e4 | 111 | } |
1da177e4 LT |
112 | |
113 | /* searches in journal structures for a given block number (bmap, off). If block | |
114 | is found in reiserfs journal it suggests next free block candidate to test. */ | |
3ee16670 JM |
115 | static inline int is_block_in_journal(struct super_block *s, unsigned int bmap, |
116 | int off, int *next) | |
1da177e4 | 117 | { |
bd4c625c LT |
118 | b_blocknr_t tmp; |
119 | ||
120 | if (reiserfs_in_journal(s, bmap, off, 1, &tmp)) { | |
121 | if (tmp) { /* hint supplied */ | |
122 | *next = tmp; | |
123 | PROC_INFO_INC(s, scan_bitmap.in_journal_hint); | |
124 | } else { | |
125 | (*next) = off + 1; /* inc offset to avoid looping. */ | |
126 | PROC_INFO_INC(s, scan_bitmap.in_journal_nohint); | |
127 | } | |
128 | PROC_INFO_INC(s, scan_bitmap.retry); | |
129 | return 1; | |
1da177e4 | 130 | } |
bd4c625c | 131 | return 0; |
1da177e4 LT |
132 | } |
133 | ||
134 | /* it searches for a window of zero bits with given minimum and maximum lengths in one bitmap | |
135 | * block; */ | |
bd4c625c | 136 | static int scan_bitmap_block(struct reiserfs_transaction_handle *th, |
3ee16670 JM |
137 | unsigned int bmap_n, int *beg, int boundary, |
138 | int min, int max, int unfm) | |
1da177e4 | 139 | { |
bd4c625c LT |
140 | struct super_block *s = th->t_super; |
141 | struct reiserfs_bitmap_info *bi = &SB_AP_BITMAP(s)[bmap_n]; | |
0b3dc17b | 142 | struct buffer_head *bh; |
bd4c625c LT |
143 | int end, next; |
144 | int org = *beg; | |
1da177e4 | 145 | |
bd4c625c | 146 | BUG_ON(!th->t_trans_id); |
1da177e4 | 147 | |
cb680c1b JM |
148 | RFALSE(bmap_n >= reiserfs_bmap_count(s), "Bitmap %u is out of " |
149 | "range (0..%u)", bmap_n, reiserfs_bmap_count(s) - 1); | |
bd4c625c | 150 | PROC_INFO_INC(s, scan_bitmap.bmap); |
1da177e4 LT |
151 | /* this is unclear and lacks comments, explain how journal bitmaps |
152 | work here for the reader. Convey a sense of the design here. What | |
153 | is a window? */ | |
154 | /* - I mean `a window of zero bits' as in description of this function - Zam. */ | |
1da177e4 | 155 | |
bd4c625c | 156 | if (!bi) { |
0030b645 JM |
157 | reiserfs_error(s, "jdm-4055", "NULL bitmap info pointer " |
158 | "for bitmap %d", bmap_n); | |
bd4c625c LT |
159 | return 0; |
160 | } | |
0b3dc17b | 161 | |
5065227b JM |
162 | bh = reiserfs_read_bitmap_block(s, bmap_n); |
163 | if (bh == NULL) | |
164 | return 0; | |
1da177e4 | 165 | |
bd4c625c LT |
166 | while (1) { |
167 | cont: | |
0b3dc17b JM |
168 | if (bi->free_count < min) { |
169 | brelse(bh); | |
bd4c625c | 170 | return 0; // No free blocks in this bitmap |
0b3dc17b | 171 | } |
bd4c625c | 172 | |
3ad2f3fb | 173 | /* search for a first zero bit -- beginning of a window */ |
bd4c625c | 174 | *beg = reiserfs_find_next_zero_le_bit |
0b3dc17b | 175 | ((unsigned long *)(bh->b_data), boundary, *beg); |
bd4c625c LT |
176 | |
177 | if (*beg + min > boundary) { /* search for a zero bit fails or the rest of bitmap block | |
178 | * cannot contain a zero window of minimum size */ | |
0b3dc17b | 179 | brelse(bh); |
bd4c625c | 180 | return 0; |
1da177e4 | 181 | } |
1da177e4 | 182 | |
bd4c625c LT |
183 | if (unfm && is_block_in_journal(s, bmap_n, *beg, beg)) |
184 | continue; | |
185 | /* first zero bit found; we check next bits */ | |
186 | for (end = *beg + 1;; end++) { | |
187 | if (end >= *beg + max || end >= boundary | |
0b3dc17b | 188 | || reiserfs_test_le_bit(end, bh->b_data)) { |
bd4c625c LT |
189 | next = end; |
190 | break; | |
191 | } | |
192 | /* finding the other end of zero bit window requires looking into journal structures (in | |
193 | * case of searching for free blocks for unformatted nodes) */ | |
194 | if (unfm && is_block_in_journal(s, bmap_n, end, &next)) | |
195 | break; | |
196 | } | |
1da177e4 | 197 | |
bd4c625c LT |
198 | /* now (*beg) points to beginning of zero bits window, |
199 | * (end) points to one bit after the window end */ | |
200 | if (end - *beg >= min) { /* it seems we have found window of proper size */ | |
201 | int i; | |
0b3dc17b | 202 | reiserfs_prepare_for_journal(s, bh, 1); |
bd4c625c LT |
203 | /* try to set all blocks used checking are they still free */ |
204 | for (i = *beg; i < end; i++) { | |
205 | /* It seems that we should not check in journal again. */ | |
206 | if (reiserfs_test_and_set_le_bit | |
0b3dc17b | 207 | (i, bh->b_data)) { |
bd4c625c LT |
208 | /* bit was set by another process |
209 | * while we slept in prepare_for_journal() */ | |
210 | PROC_INFO_INC(s, scan_bitmap.stolen); | |
211 | if (i >= *beg + min) { /* we can continue with smaller set of allocated blocks, | |
212 | * if length of this set is more or equal to `min' */ | |
213 | end = i; | |
214 | break; | |
215 | } | |
216 | /* otherwise we clear all bit were set ... */ | |
217 | while (--i >= *beg) | |
0c2fd1bf | 218 | reiserfs_clear_le_bit |
0b3dc17b JM |
219 | (i, bh->b_data); |
220 | reiserfs_restore_prepared_buffer(s, bh); | |
bd4c625c LT |
221 | *beg = org; |
222 | /* ... and search again in current block from beginning */ | |
223 | goto cont; | |
224 | } | |
225 | } | |
226 | bi->free_count -= (end - *beg); | |
0b3dc17b JM |
227 | journal_mark_dirty(th, s, bh); |
228 | brelse(bh); | |
bd4c625c LT |
229 | |
230 | /* free block count calculation */ | |
231 | reiserfs_prepare_for_journal(s, SB_BUFFER_WITH_SB(s), | |
232 | 1); | |
233 | PUT_SB_FREE_BLOCKS(s, SB_FREE_BLOCKS(s) - (end - *beg)); | |
234 | journal_mark_dirty(th, s, SB_BUFFER_WITH_SB(s)); | |
235 | ||
236 | return end - (*beg); | |
237 | } else { | |
238 | *beg = next; | |
239 | } | |
1da177e4 | 240 | } |
1da177e4 LT |
241 | } |
242 | ||
bd4c625c LT |
243 | static int bmap_hash_id(struct super_block *s, u32 id) |
244 | { | |
245 | char *hash_in = NULL; | |
246 | unsigned long hash; | |
247 | unsigned bm; | |
248 | ||
249 | if (id <= 2) { | |
250 | bm = 1; | |
251 | } else { | |
252 | hash_in = (char *)(&id); | |
253 | hash = keyed_hash(hash_in, 4); | |
cb680c1b | 254 | bm = hash % reiserfs_bmap_count(s); |
bd4c625c LT |
255 | if (!bm) |
256 | bm = 1; | |
257 | } | |
258 | /* this can only be true when SB_BMAP_NR = 1 */ | |
cb680c1b | 259 | if (bm >= reiserfs_bmap_count(s)) |
bd4c625c LT |
260 | bm = 0; |
261 | return bm; | |
1da177e4 LT |
262 | } |
263 | ||
264 | /* | |
265 | * hashes the id and then returns > 0 if the block group for the | |
266 | * corresponding hash is full | |
267 | */ | |
bd4c625c LT |
268 | static inline int block_group_used(struct super_block *s, u32 id) |
269 | { | |
5065227b JM |
270 | int bm = bmap_hash_id(s, id); |
271 | struct reiserfs_bitmap_info *info = &SB_AP_BITMAP(s)[bm]; | |
272 | ||
273 | /* If we don't have cached information on this bitmap block, we're | |
274 | * going to have to load it later anyway. Loading it here allows us | |
c78bad11 | 275 | * to make a better decision. This favors long-term performance gain |
5065227b JM |
276 | * with a better on-disk layout vs. a short term gain of skipping the |
277 | * read and potentially having a bad placement. */ | |
4d20851d | 278 | if (info->free_count == UINT_MAX) { |
5065227b JM |
279 | struct buffer_head *bh = reiserfs_read_bitmap_block(s, bm); |
280 | brelse(bh); | |
281 | } | |
282 | ||
283 | if (info->free_count > ((s->s_blocksize << 3) * 60 / 100)) { | |
bd4c625c LT |
284 | return 0; |
285 | } | |
286 | return 1; | |
1da177e4 LT |
287 | } |
288 | ||
289 | /* | |
290 | * the packing is returned in disk byte order | |
291 | */ | |
bd4c625c | 292 | __le32 reiserfs_choose_packing(struct inode * dir) |
3e8962be | 293 | { |
bd4c625c LT |
294 | __le32 packing; |
295 | if (TEST_OPTION(packing_groups, dir->i_sb)) { | |
296 | u32 parent_dir = le32_to_cpu(INODE_PKEY(dir)->k_dir_id); | |
297 | /* | |
298 | * some versions of reiserfsck expect packing locality 1 to be | |
299 | * special | |
300 | */ | |
301 | if (parent_dir == 1 || block_group_used(dir->i_sb, parent_dir)) | |
302 | packing = INODE_PKEY(dir)->k_objectid; | |
303 | else | |
304 | packing = INODE_PKEY(dir)->k_dir_id; | |
305 | } else | |
306 | packing = INODE_PKEY(dir)->k_objectid; | |
307 | return packing; | |
1da177e4 | 308 | } |
bd4c625c | 309 | |
1da177e4 LT |
310 | /* Tries to find contiguous zero bit window (given size) in given region of |
311 | * bitmap and place new blocks there. Returns number of allocated blocks. */ | |
bd4c625c LT |
312 | static int scan_bitmap(struct reiserfs_transaction_handle *th, |
313 | b_blocknr_t * start, b_blocknr_t finish, | |
3ee16670 | 314 | int min, int max, int unfm, sector_t file_block) |
1da177e4 | 315 | { |
bd4c625c LT |
316 | int nr_allocated = 0; |
317 | struct super_block *s = th->t_super; | |
318 | /* find every bm and bmap and bmap_nr in this file, and change them all to bitmap_blocknr | |
319 | * - Hans, it is not a block number - Zam. */ | |
320 | ||
3ee16670 JM |
321 | unsigned int bm, off; |
322 | unsigned int end_bm, end_off; | |
323 | unsigned int off_max = s->s_blocksize << 3; | |
bd4c625c LT |
324 | |
325 | BUG_ON(!th->t_trans_id); | |
326 | ||
327 | PROC_INFO_INC(s, scan_bitmap.call); | |
328 | if (SB_FREE_BLOCKS(s) <= 0) | |
329 | return 0; // No point in looking for more free blocks | |
330 | ||
331 | get_bit_address(s, *start, &bm, &off); | |
332 | get_bit_address(s, finish, &end_bm, &end_off); | |
cb680c1b | 333 | if (bm > reiserfs_bmap_count(s)) |
bd4c625c | 334 | return 0; |
cb680c1b JM |
335 | if (end_bm > reiserfs_bmap_count(s)) |
336 | end_bm = reiserfs_bmap_count(s); | |
bd4c625c LT |
337 | |
338 | /* When the bitmap is more than 10% free, anyone can allocate. | |
339 | * When it's less than 10% free, only files that already use the | |
340 | * bitmap are allowed. Once we pass 80% full, this restriction | |
341 | * is lifted. | |
342 | * | |
343 | * We do this so that files that grow later still have space close to | |
344 | * their original allocation. This improves locality, and presumably | |
345 | * performance as a result. | |
346 | * | |
347 | * This is only an allocation policy and does not make up for getting a | |
348 | * bad hint. Decent hinting must be implemented for this to work well. | |
349 | */ | |
350 | if (TEST_OPTION(skip_busy, s) | |
351 | && SB_FREE_BLOCKS(s) > SB_BLOCK_COUNT(s) / 20) { | |
352 | for (; bm < end_bm; bm++, off = 0) { | |
353 | if ((off && (!unfm || (file_block != 0))) | |
354 | || SB_AP_BITMAP(s)[bm].free_count > | |
355 | (s->s_blocksize << 3) / 10) | |
356 | nr_allocated = | |
357 | scan_bitmap_block(th, bm, &off, off_max, | |
358 | min, max, unfm); | |
359 | if (nr_allocated) | |
360 | goto ret; | |
361 | } | |
362 | /* we know from above that start is a reasonable number */ | |
363 | get_bit_address(s, *start, &bm, &off); | |
364 | } | |
365 | ||
366 | for (; bm < end_bm; bm++, off = 0) { | |
367 | nr_allocated = | |
368 | scan_bitmap_block(th, bm, &off, off_max, min, max, unfm); | |
369 | if (nr_allocated) | |
370 | goto ret; | |
371 | } | |
372 | ||
373 | nr_allocated = | |
374 | scan_bitmap_block(th, bm, &off, end_off + 1, min, max, unfm); | |
375 | ||
376 | ret: | |
377 | *start = bm * off_max + off; | |
378 | return nr_allocated; | |
1da177e4 LT |
379 | |
380 | } | |
381 | ||
bd4c625c LT |
382 | static void _reiserfs_free_block(struct reiserfs_transaction_handle *th, |
383 | struct inode *inode, b_blocknr_t block, | |
384 | int for_unformatted) | |
1da177e4 | 385 | { |
bd4c625c LT |
386 | struct super_block *s = th->t_super; |
387 | struct reiserfs_super_block *rs; | |
0b3dc17b | 388 | struct buffer_head *sbh, *bmbh; |
bd4c625c | 389 | struct reiserfs_bitmap_info *apbi; |
3ee16670 | 390 | unsigned int nr, offset; |
1da177e4 | 391 | |
bd4c625c | 392 | BUG_ON(!th->t_trans_id); |
1da177e4 | 393 | |
bd4c625c | 394 | PROC_INFO_INC(s, free_block); |
1da177e4 | 395 | |
bd4c625c LT |
396 | rs = SB_DISK_SUPER_BLOCK(s); |
397 | sbh = SB_BUFFER_WITH_SB(s); | |
398 | apbi = SB_AP_BITMAP(s); | |
1da177e4 | 399 | |
bd4c625c | 400 | get_bit_address(s, block, &nr, &offset); |
1da177e4 | 401 | |
cb680c1b | 402 | if (nr >= reiserfs_bmap_count(s)) { |
0030b645 JM |
403 | reiserfs_error(s, "vs-4075", "block %lu is out of range", |
404 | block); | |
bd4c625c LT |
405 | return; |
406 | } | |
407 | ||
5065227b JM |
408 | bmbh = reiserfs_read_bitmap_block(s, nr); |
409 | if (!bmbh) | |
410 | return; | |
0b3dc17b JM |
411 | |
412 | reiserfs_prepare_for_journal(s, bmbh, 1); | |
bd4c625c LT |
413 | |
414 | /* clear bit for the given block in bit map */ | |
0b3dc17b | 415 | if (!reiserfs_test_and_clear_le_bit(offset, bmbh->b_data)) { |
0030b645 JM |
416 | reiserfs_error(s, "vs-4080", |
417 | "block %lu: bit already cleared", block); | |
bd4c625c LT |
418 | } |
419 | apbi[nr].free_count++; | |
0b3dc17b JM |
420 | journal_mark_dirty(th, s, bmbh); |
421 | brelse(bmbh); | |
bd4c625c LT |
422 | |
423 | reiserfs_prepare_for_journal(s, sbh, 1); | |
424 | /* update super block */ | |
425 | set_sb_free_blocks(rs, sb_free_blocks(rs) + 1); | |
426 | ||
427 | journal_mark_dirty(th, s, sbh); | |
428 | if (for_unformatted) | |
5dd4056d | 429 | dquot_free_block_nodirty(inode, 1); |
1da177e4 LT |
430 | } |
431 | ||
bd4c625c LT |
432 | void reiserfs_free_block(struct reiserfs_transaction_handle *th, |
433 | struct inode *inode, b_blocknr_t block, | |
434 | int for_unformatted) | |
1da177e4 | 435 | { |
bd4c625c | 436 | struct super_block *s = th->t_super; |
bd4c625c | 437 | BUG_ON(!th->t_trans_id); |
1da177e4 | 438 | |
bd4c625c | 439 | RFALSE(!s, "vs-4061: trying to free block on nonexistent device"); |
d4c3d19d JM |
440 | if (!is_reusable(s, block, 1)) |
441 | return; | |
442 | ||
443 | if (block > sb_block_count(REISERFS_SB(s)->s_rs)) { | |
0030b645 | 444 | reiserfs_error(th->t_super, "bitmap-4072", |
d4c3d19d JM |
445 | "Trying to free block outside file system " |
446 | "boundaries (%lu > %lu)", | |
447 | block, sb_block_count(REISERFS_SB(s)->s_rs)); | |
448 | return; | |
449 | } | |
bd4c625c LT |
450 | /* mark it before we clear it, just in case */ |
451 | journal_mark_freed(th, s, block); | |
452 | _reiserfs_free_block(th, inode, block, for_unformatted); | |
1da177e4 LT |
453 | } |
454 | ||
455 | /* preallocated blocks don't need to be run through journal_mark_freed */ | |
bd4c625c LT |
456 | static void reiserfs_free_prealloc_block(struct reiserfs_transaction_handle *th, |
457 | struct inode *inode, b_blocknr_t block) | |
458 | { | |
d4c3d19d | 459 | BUG_ON(!th->t_trans_id); |
bd4c625c LT |
460 | RFALSE(!th->t_super, |
461 | "vs-4060: trying to free block on nonexistent device"); | |
d4c3d19d JM |
462 | if (!is_reusable(th->t_super, block, 1)) |
463 | return; | |
bd4c625c | 464 | _reiserfs_free_block(th, inode, block, 1); |
1da177e4 LT |
465 | } |
466 | ||
bd4c625c LT |
467 | static void __discard_prealloc(struct reiserfs_transaction_handle *th, |
468 | struct reiserfs_inode_info *ei) | |
1da177e4 | 469 | { |
bd4c625c LT |
470 | unsigned long save = ei->i_prealloc_block; |
471 | int dirty = 0; | |
472 | struct inode *inode = &ei->vfs_inode; | |
473 | BUG_ON(!th->t_trans_id); | |
1da177e4 | 474 | #ifdef CONFIG_REISERFS_CHECK |
bd4c625c | 475 | if (ei->i_prealloc_count < 0) |
0030b645 JM |
476 | reiserfs_error(th->t_super, "zam-4001", |
477 | "inode has negative prealloc blocks count."); | |
1da177e4 | 478 | #endif |
bd4c625c LT |
479 | while (ei->i_prealloc_count > 0) { |
480 | reiserfs_free_prealloc_block(th, inode, ei->i_prealloc_block); | |
481 | ei->i_prealloc_block++; | |
482 | ei->i_prealloc_count--; | |
483 | dirty = 1; | |
484 | } | |
485 | if (dirty) | |
486 | reiserfs_update_sd(th, inode); | |
487 | ei->i_prealloc_block = save; | |
488 | list_del_init(&(ei->i_prealloc_list)); | |
1da177e4 LT |
489 | } |
490 | ||
491 | /* FIXME: It should be inline function */ | |
bd4c625c LT |
492 | void reiserfs_discard_prealloc(struct reiserfs_transaction_handle *th, |
493 | struct inode *inode) | |
1da177e4 | 494 | { |
bd4c625c LT |
495 | struct reiserfs_inode_info *ei = REISERFS_I(inode); |
496 | BUG_ON(!th->t_trans_id); | |
497 | if (ei->i_prealloc_count) | |
498 | __discard_prealloc(th, ei); | |
1da177e4 LT |
499 | } |
500 | ||
bd4c625c | 501 | void reiserfs_discard_all_prealloc(struct reiserfs_transaction_handle *th) |
1da177e4 | 502 | { |
bd4c625c | 503 | struct list_head *plist = &SB_JOURNAL(th->t_super)->j_prealloc_list; |
1da177e4 | 504 | |
bd4c625c | 505 | BUG_ON(!th->t_trans_id); |
1da177e4 | 506 | |
bd4c625c LT |
507 | while (!list_empty(plist)) { |
508 | struct reiserfs_inode_info *ei; | |
509 | ei = list_entry(plist->next, struct reiserfs_inode_info, | |
510 | i_prealloc_list); | |
1da177e4 | 511 | #ifdef CONFIG_REISERFS_CHECK |
bd4c625c | 512 | if (!ei->i_prealloc_count) { |
0030b645 JM |
513 | reiserfs_error(th->t_super, "zam-4001", |
514 | "inode is in prealloc list but has " | |
515 | "no preallocated blocks."); | |
bd4c625c | 516 | } |
1da177e4 | 517 | #endif |
bd4c625c LT |
518 | __discard_prealloc(th, ei); |
519 | } | |
1da177e4 LT |
520 | } |
521 | ||
bd4c625c | 522 | void reiserfs_init_alloc_options(struct super_block *s) |
1da177e4 | 523 | { |
bd4c625c LT |
524 | set_bit(_ALLOC_skip_busy, &SB_ALLOC_OPTS(s)); |
525 | set_bit(_ALLOC_dirid_groups, &SB_ALLOC_OPTS(s)); | |
526 | set_bit(_ALLOC_packing_groups, &SB_ALLOC_OPTS(s)); | |
1da177e4 LT |
527 | } |
528 | ||
529 | /* block allocator related options are parsed here */ | |
bd4c625c | 530 | int reiserfs_parse_alloc_options(struct super_block *s, char *options) |
1da177e4 | 531 | { |
bd4c625c LT |
532 | char *this_char, *value; |
533 | ||
534 | REISERFS_SB(s)->s_alloc_options.bits = 0; /* clear default settings */ | |
535 | ||
536 | while ((this_char = strsep(&options, ":")) != NULL) { | |
537 | if ((value = strchr(this_char, '=')) != NULL) | |
538 | *value++ = 0; | |
539 | ||
540 | if (!strcmp(this_char, "concentrating_formatted_nodes")) { | |
541 | int temp; | |
542 | SET_OPTION(concentrating_formatted_nodes); | |
543 | temp = (value | |
544 | && *value) ? simple_strtoul(value, &value, | |
545 | 0) : 10; | |
546 | if (temp <= 0 || temp > 100) { | |
547 | REISERFS_SB(s)->s_alloc_options.border = 10; | |
548 | } else { | |
549 | REISERFS_SB(s)->s_alloc_options.border = | |
550 | 100 / temp; | |
551 | } | |
552 | continue; | |
553 | } | |
554 | if (!strcmp(this_char, "displacing_large_files")) { | |
555 | SET_OPTION(displacing_large_files); | |
556 | REISERFS_SB(s)->s_alloc_options.large_file_size = | |
557 | (value | |
558 | && *value) ? simple_strtoul(value, &value, 0) : 16; | |
559 | continue; | |
560 | } | |
561 | if (!strcmp(this_char, "displacing_new_packing_localities")) { | |
562 | SET_OPTION(displacing_new_packing_localities); | |
563 | continue; | |
564 | }; | |
565 | ||
566 | if (!strcmp(this_char, "old_hashed_relocation")) { | |
567 | SET_OPTION(old_hashed_relocation); | |
568 | continue; | |
569 | } | |
1da177e4 | 570 | |
bd4c625c LT |
571 | if (!strcmp(this_char, "new_hashed_relocation")) { |
572 | SET_OPTION(new_hashed_relocation); | |
573 | continue; | |
574 | } | |
1da177e4 | 575 | |
bd4c625c LT |
576 | if (!strcmp(this_char, "dirid_groups")) { |
577 | SET_OPTION(dirid_groups); | |
578 | continue; | |
579 | } | |
580 | if (!strcmp(this_char, "oid_groups")) { | |
581 | SET_OPTION(oid_groups); | |
582 | continue; | |
583 | } | |
584 | if (!strcmp(this_char, "packing_groups")) { | |
585 | SET_OPTION(packing_groups); | |
586 | continue; | |
587 | } | |
588 | if (!strcmp(this_char, "hashed_formatted_nodes")) { | |
589 | SET_OPTION(hashed_formatted_nodes); | |
590 | continue; | |
591 | } | |
1da177e4 | 592 | |
bd4c625c LT |
593 | if (!strcmp(this_char, "skip_busy")) { |
594 | SET_OPTION(skip_busy); | |
595 | continue; | |
596 | } | |
1da177e4 | 597 | |
bd4c625c LT |
598 | if (!strcmp(this_char, "hundredth_slices")) { |
599 | SET_OPTION(hundredth_slices); | |
600 | continue; | |
601 | } | |
1da177e4 | 602 | |
bd4c625c LT |
603 | if (!strcmp(this_char, "old_way")) { |
604 | SET_OPTION(old_way); | |
605 | continue; | |
606 | } | |
1da177e4 | 607 | |
bd4c625c LT |
608 | if (!strcmp(this_char, "displace_based_on_dirid")) { |
609 | SET_OPTION(displace_based_on_dirid); | |
610 | continue; | |
611 | } | |
1da177e4 | 612 | |
bd4c625c LT |
613 | if (!strcmp(this_char, "preallocmin")) { |
614 | REISERFS_SB(s)->s_alloc_options.preallocmin = | |
615 | (value | |
616 | && *value) ? simple_strtoul(value, &value, 0) : 4; | |
617 | continue; | |
618 | } | |
619 | ||
620 | if (!strcmp(this_char, "preallocsize")) { | |
621 | REISERFS_SB(s)->s_alloc_options.preallocsize = | |
622 | (value | |
623 | && *value) ? simple_strtoul(value, &value, | |
624 | 0) : | |
625 | PREALLOCATION_SIZE; | |
626 | continue; | |
627 | } | |
1da177e4 | 628 | |
45b03d5e JM |
629 | reiserfs_warning(s, "zam-4001", "unknown option - %s", |
630 | this_char); | |
bd4c625c | 631 | return 1; |
1da177e4 LT |
632 | } |
633 | ||
1d889d99 | 634 | reiserfs_info(s, "allocator options = [%08x]\n", SB_ALLOC_OPTS(s)); |
bd4c625c | 635 | return 0; |
1da177e4 | 636 | } |
bd4c625c | 637 | |
c3aa0776 JK |
638 | static void print_sep(struct seq_file *seq, int *first) |
639 | { | |
640 | if (!*first) | |
641 | seq_puts(seq, ":"); | |
642 | else | |
643 | *first = 0; | |
644 | } | |
645 | ||
646 | void show_alloc_options(struct seq_file *seq, struct super_block *s) | |
647 | { | |
648 | int first = 1; | |
649 | ||
650 | if (SB_ALLOC_OPTS(s) == ((1 << _ALLOC_skip_busy) | | |
651 | (1 << _ALLOC_dirid_groups) | (1 << _ALLOC_packing_groups))) | |
652 | return; | |
653 | ||
654 | seq_puts(seq, ",alloc="); | |
655 | ||
656 | if (TEST_OPTION(concentrating_formatted_nodes, s)) { | |
657 | print_sep(seq, &first); | |
658 | if (REISERFS_SB(s)->s_alloc_options.border != 10) { | |
659 | seq_printf(seq, "concentrating_formatted_nodes=%d", | |
660 | 100 / REISERFS_SB(s)->s_alloc_options.border); | |
661 | } else | |
662 | seq_puts(seq, "concentrating_formatted_nodes"); | |
663 | } | |
664 | if (TEST_OPTION(displacing_large_files, s)) { | |
665 | print_sep(seq, &first); | |
666 | if (REISERFS_SB(s)->s_alloc_options.large_file_size != 16) { | |
667 | seq_printf(seq, "displacing_large_files=%lu", | |
668 | REISERFS_SB(s)->s_alloc_options.large_file_size); | |
669 | } else | |
670 | seq_puts(seq, "displacing_large_files"); | |
671 | } | |
672 | if (TEST_OPTION(displacing_new_packing_localities, s)) { | |
673 | print_sep(seq, &first); | |
674 | seq_puts(seq, "displacing_new_packing_localities"); | |
675 | } | |
676 | if (TEST_OPTION(old_hashed_relocation, s)) { | |
677 | print_sep(seq, &first); | |
678 | seq_puts(seq, "old_hashed_relocation"); | |
679 | } | |
680 | if (TEST_OPTION(new_hashed_relocation, s)) { | |
681 | print_sep(seq, &first); | |
682 | seq_puts(seq, "new_hashed_relocation"); | |
683 | } | |
684 | if (TEST_OPTION(dirid_groups, s)) { | |
685 | print_sep(seq, &first); | |
686 | seq_puts(seq, "dirid_groups"); | |
687 | } | |
688 | if (TEST_OPTION(oid_groups, s)) { | |
689 | print_sep(seq, &first); | |
690 | seq_puts(seq, "oid_groups"); | |
691 | } | |
692 | if (TEST_OPTION(packing_groups, s)) { | |
693 | print_sep(seq, &first); | |
694 | seq_puts(seq, "packing_groups"); | |
695 | } | |
696 | if (TEST_OPTION(hashed_formatted_nodes, s)) { | |
697 | print_sep(seq, &first); | |
698 | seq_puts(seq, "hashed_formatted_nodes"); | |
699 | } | |
700 | if (TEST_OPTION(skip_busy, s)) { | |
701 | print_sep(seq, &first); | |
702 | seq_puts(seq, "skip_busy"); | |
703 | } | |
704 | if (TEST_OPTION(hundredth_slices, s)) { | |
705 | print_sep(seq, &first); | |
706 | seq_puts(seq, "hundredth_slices"); | |
707 | } | |
708 | if (TEST_OPTION(old_way, s)) { | |
709 | print_sep(seq, &first); | |
710 | seq_puts(seq, "old_way"); | |
711 | } | |
712 | if (TEST_OPTION(displace_based_on_dirid, s)) { | |
713 | print_sep(seq, &first); | |
714 | seq_puts(seq, "displace_based_on_dirid"); | |
715 | } | |
716 | if (REISERFS_SB(s)->s_alloc_options.preallocmin != 0) { | |
717 | print_sep(seq, &first); | |
718 | seq_printf(seq, "preallocmin=%d", | |
719 | REISERFS_SB(s)->s_alloc_options.preallocmin); | |
720 | } | |
721 | if (REISERFS_SB(s)->s_alloc_options.preallocsize != 17) { | |
722 | print_sep(seq, &first); | |
723 | seq_printf(seq, "preallocsize=%d", | |
724 | REISERFS_SB(s)->s_alloc_options.preallocsize); | |
725 | } | |
726 | } | |
727 | ||
bd4c625c | 728 | static inline void new_hashed_relocation(reiserfs_blocknr_hint_t * hint) |
1da177e4 | 729 | { |
bd4c625c LT |
730 | char *hash_in; |
731 | if (hint->formatted_node) { | |
732 | hash_in = (char *)&hint->key.k_dir_id; | |
733 | } else { | |
734 | if (!hint->inode) { | |
735 | //hint->search_start = hint->beg; | |
736 | hash_in = (char *)&hint->key.k_dir_id; | |
737 | } else | |
738 | if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super)) | |
739 | hash_in = (char *)(&INODE_PKEY(hint->inode)->k_dir_id); | |
740 | else | |
741 | hash_in = | |
742 | (char *)(&INODE_PKEY(hint->inode)->k_objectid); | |
743 | } | |
1da177e4 | 744 | |
bd4c625c LT |
745 | hint->search_start = |
746 | hint->beg + keyed_hash(hash_in, 4) % (hint->end - hint->beg); | |
1da177e4 LT |
747 | } |
748 | ||
749 | /* | |
750 | * Relocation based on dirid, hashing them into a given bitmap block | |
c78bad11 | 751 | * files. Formatted nodes are unaffected, a separate policy covers them |
1da177e4 | 752 | */ |
bd4c625c | 753 | static void dirid_groups(reiserfs_blocknr_hint_t * hint) |
1da177e4 | 754 | { |
bd4c625c LT |
755 | unsigned long hash; |
756 | __u32 dirid = 0; | |
757 | int bm = 0; | |
758 | struct super_block *sb = hint->th->t_super; | |
1da177e4 | 759 | if (hint->inode) |
bd4c625c LT |
760 | dirid = le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id); |
761 | else if (hint->formatted_node) | |
762 | dirid = hint->key.k_dir_id; | |
763 | ||
764 | if (dirid) { | |
765 | bm = bmap_hash_id(sb, dirid); | |
766 | hash = bm * (sb->s_blocksize << 3); | |
767 | /* give a portion of the block group to metadata */ | |
768 | if (hint->inode) | |
769 | hash += sb->s_blocksize / 2; | |
770 | hint->search_start = hash; | |
771 | } | |
1da177e4 LT |
772 | } |
773 | ||
774 | /* | |
775 | * Relocation based on oid, hashing them into a given bitmap block | |
c78bad11 | 776 | * files. Formatted nodes are unaffected, a separate policy covers them |
1da177e4 | 777 | */ |
bd4c625c | 778 | static void oid_groups(reiserfs_blocknr_hint_t * hint) |
1da177e4 | 779 | { |
bd4c625c LT |
780 | if (hint->inode) { |
781 | unsigned long hash; | |
782 | __u32 oid; | |
783 | __u32 dirid; | |
784 | int bm; | |
785 | ||
786 | dirid = le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id); | |
787 | ||
788 | /* keep the root dir and it's first set of subdirs close to | |
789 | * the start of the disk | |
790 | */ | |
791 | if (dirid <= 2) | |
792 | hash = (hint->inode->i_sb->s_blocksize << 3); | |
793 | else { | |
794 | oid = le32_to_cpu(INODE_PKEY(hint->inode)->k_objectid); | |
795 | bm = bmap_hash_id(hint->inode->i_sb, oid); | |
796 | hash = bm * (hint->inode->i_sb->s_blocksize << 3); | |
797 | } | |
798 | hint->search_start = hash; | |
1da177e4 | 799 | } |
1da177e4 LT |
800 | } |
801 | ||
802 | /* returns 1 if it finds an indirect item and gets valid hint info | |
803 | * from it, otherwise 0 | |
804 | */ | |
bd4c625c | 805 | static int get_left_neighbor(reiserfs_blocknr_hint_t * hint) |
1da177e4 | 806 | { |
fec6d055 | 807 | struct treepath *path; |
bd4c625c LT |
808 | struct buffer_head *bh; |
809 | struct item_head *ih; | |
810 | int pos_in_item; | |
811 | __le32 *item; | |
812 | int ret = 0; | |
813 | ||
814 | if (!hint->path) /* reiserfs code can call this function w/o pointer to path | |
1da177e4 | 815 | * structure supplied; then we rely on supplied search_start */ |
bd4c625c LT |
816 | return 0; |
817 | ||
818 | path = hint->path; | |
819 | bh = get_last_bh(path); | |
820 | RFALSE(!bh, "green-4002: Illegal path specified to get_left_neighbor"); | |
821 | ih = get_ih(path); | |
822 | pos_in_item = path->pos_in_item; | |
823 | item = get_item(path); | |
824 | ||
825 | hint->search_start = bh->b_blocknr; | |
826 | ||
827 | if (!hint->formatted_node && is_indirect_le_ih(ih)) { | |
828 | /* for indirect item: go to left and look for the first non-hole entry | |
829 | in the indirect item */ | |
830 | if (pos_in_item == I_UNFM_NUM(ih)) | |
831 | pos_in_item--; | |
832 | // pos_in_item = I_UNFM_NUM (ih) - 1; | |
833 | while (pos_in_item >= 0) { | |
834 | int t = get_block_num(item, pos_in_item); | |
835 | if (t) { | |
836 | hint->search_start = t; | |
837 | ret = 1; | |
838 | break; | |
839 | } | |
840 | pos_in_item--; | |
841 | } | |
1da177e4 | 842 | } |
1da177e4 | 843 | |
bd4c625c LT |
844 | /* does result value fit into specified region? */ |
845 | return ret; | |
1da177e4 LT |
846 | } |
847 | ||
848 | /* should be, if formatted node, then try to put on first part of the device | |
849 | specified as number of percent with mount option device, else try to put | |
850 | on last of device. This is not to say it is good code to do so, | |
851 | but the effect should be measured. */ | |
bd4c625c LT |
852 | static inline void set_border_in_hint(struct super_block *s, |
853 | reiserfs_blocknr_hint_t * hint) | |
1da177e4 | 854 | { |
bd4c625c LT |
855 | b_blocknr_t border = |
856 | SB_BLOCK_COUNT(s) / REISERFS_SB(s)->s_alloc_options.border; | |
1da177e4 | 857 | |
bd4c625c LT |
858 | if (hint->formatted_node) |
859 | hint->end = border - 1; | |
860 | else | |
861 | hint->beg = border; | |
1da177e4 LT |
862 | } |
863 | ||
bd4c625c | 864 | static inline void displace_large_file(reiserfs_blocknr_hint_t * hint) |
1da177e4 | 865 | { |
bd4c625c LT |
866 | if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super)) |
867 | hint->search_start = | |
868 | hint->beg + | |
869 | keyed_hash((char *)(&INODE_PKEY(hint->inode)->k_dir_id), | |
870 | 4) % (hint->end - hint->beg); | |
871 | else | |
872 | hint->search_start = | |
873 | hint->beg + | |
874 | keyed_hash((char *)(&INODE_PKEY(hint->inode)->k_objectid), | |
875 | 4) % (hint->end - hint->beg); | |
1da177e4 LT |
876 | } |
877 | ||
bd4c625c | 878 | static inline void hash_formatted_node(reiserfs_blocknr_hint_t * hint) |
1da177e4 | 879 | { |
bd4c625c | 880 | char *hash_in; |
1da177e4 | 881 | |
bd4c625c LT |
882 | if (!hint->inode) |
883 | hash_in = (char *)&hint->key.k_dir_id; | |
884 | else if (TEST_OPTION(displace_based_on_dirid, hint->th->t_super)) | |
885 | hash_in = (char *)(&INODE_PKEY(hint->inode)->k_dir_id); | |
886 | else | |
887 | hash_in = (char *)(&INODE_PKEY(hint->inode)->k_objectid); | |
1da177e4 | 888 | |
bd4c625c LT |
889 | hint->search_start = |
890 | hint->beg + keyed_hash(hash_in, 4) % (hint->end - hint->beg); | |
1da177e4 LT |
891 | } |
892 | ||
bd4c625c LT |
893 | static inline int |
894 | this_blocknr_allocation_would_make_it_a_large_file(reiserfs_blocknr_hint_t * | |
895 | hint) | |
1da177e4 | 896 | { |
bd4c625c LT |
897 | return hint->block == |
898 | REISERFS_SB(hint->th->t_super)->s_alloc_options.large_file_size; | |
1da177e4 LT |
899 | } |
900 | ||
901 | #ifdef DISPLACE_NEW_PACKING_LOCALITIES | |
bd4c625c | 902 | static inline void displace_new_packing_locality(reiserfs_blocknr_hint_t * hint) |
1da177e4 | 903 | { |
bd4c625c | 904 | struct in_core_key *key = &hint->key; |
1da177e4 | 905 | |
bd4c625c LT |
906 | hint->th->displace_new_blocks = 0; |
907 | hint->search_start = | |
908 | hint->beg + keyed_hash((char *)(&key->k_objectid), | |
909 | 4) % (hint->end - hint->beg); | |
1da177e4 | 910 | } |
bd4c625c | 911 | #endif |
1da177e4 | 912 | |
bd4c625c | 913 | static inline int old_hashed_relocation(reiserfs_blocknr_hint_t * hint) |
1da177e4 | 914 | { |
bd4c625c LT |
915 | b_blocknr_t border; |
916 | u32 hash_in; | |
1da177e4 | 917 | |
bd4c625c LT |
918 | if (hint->formatted_node || hint->inode == NULL) { |
919 | return 0; | |
920 | } | |
1da177e4 | 921 | |
bd4c625c LT |
922 | hash_in = le32_to_cpu((INODE_PKEY(hint->inode))->k_dir_id); |
923 | border = | |
924 | hint->beg + (u32) keyed_hash(((char *)(&hash_in)), | |
925 | 4) % (hint->end - hint->beg - 1); | |
926 | if (border > hint->search_start) | |
927 | hint->search_start = border; | |
1da177e4 | 928 | |
bd4c625c | 929 | return 1; |
1da177e4 LT |
930 | } |
931 | ||
bd4c625c | 932 | static inline int old_way(reiserfs_blocknr_hint_t * hint) |
1da177e4 | 933 | { |
bd4c625c LT |
934 | b_blocknr_t border; |
935 | ||
936 | if (hint->formatted_node || hint->inode == NULL) { | |
937 | return 0; | |
938 | } | |
939 | ||
940 | border = | |
941 | hint->beg + | |
942 | le32_to_cpu(INODE_PKEY(hint->inode)->k_dir_id) % (hint->end - | |
943 | hint->beg); | |
944 | if (border > hint->search_start) | |
945 | hint->search_start = border; | |
1da177e4 | 946 | |
bd4c625c LT |
947 | return 1; |
948 | } | |
949 | ||
950 | static inline void hundredth_slices(reiserfs_blocknr_hint_t * hint) | |
951 | { | |
952 | struct in_core_key *key = &hint->key; | |
953 | b_blocknr_t slice_start; | |
954 | ||
955 | slice_start = | |
956 | (keyed_hash((char *)(&key->k_dir_id), 4) % 100) * (hint->end / 100); | |
957 | if (slice_start > hint->search_start | |
958 | || slice_start + (hint->end / 100) <= hint->search_start) { | |
959 | hint->search_start = slice_start; | |
960 | } | |
1da177e4 | 961 | } |
bd4c625c LT |
962 | |
963 | static void determine_search_start(reiserfs_blocknr_hint_t * hint, | |
964 | int amount_needed) | |
1da177e4 | 965 | { |
bd4c625c LT |
966 | struct super_block *s = hint->th->t_super; |
967 | int unfm_hint; | |
1da177e4 | 968 | |
bd4c625c LT |
969 | hint->beg = 0; |
970 | hint->end = SB_BLOCK_COUNT(s) - 1; | |
1da177e4 | 971 | |
bd4c625c LT |
972 | /* This is former border algorithm. Now with tunable border offset */ |
973 | if (concentrating_formatted_nodes(s)) | |
974 | set_border_in_hint(s, hint); | |
1da177e4 LT |
975 | |
976 | #ifdef DISPLACE_NEW_PACKING_LOCALITIES | |
bd4c625c LT |
977 | /* whenever we create a new directory, we displace it. At first we will |
978 | hash for location, later we might look for a moderately empty place for | |
979 | it */ | |
980 | if (displacing_new_packing_localities(s) | |
981 | && hint->th->displace_new_blocks) { | |
982 | displace_new_packing_locality(hint); | |
983 | ||
984 | /* we do not continue determine_search_start, | |
985 | * if new packing locality is being displaced */ | |
986 | return; | |
987 | } | |
1da177e4 | 988 | #endif |
1da177e4 | 989 | |
bd4c625c LT |
990 | /* all persons should feel encouraged to add more special cases here and |
991 | * test them */ | |
1da177e4 | 992 | |
bd4c625c LT |
993 | if (displacing_large_files(s) && !hint->formatted_node |
994 | && this_blocknr_allocation_would_make_it_a_large_file(hint)) { | |
995 | displace_large_file(hint); | |
996 | return; | |
997 | } | |
1da177e4 | 998 | |
bd4c625c LT |
999 | /* if none of our special cases is relevant, use the left neighbor in the |
1000 | tree order of the new node we are allocating for */ | |
1001 | if (hint->formatted_node && TEST_OPTION(hashed_formatted_nodes, s)) { | |
1002 | hash_formatted_node(hint); | |
1003 | return; | |
1004 | } | |
1da177e4 | 1005 | |
bd4c625c LT |
1006 | unfm_hint = get_left_neighbor(hint); |
1007 | ||
1008 | /* Mimic old block allocator behaviour, that is if VFS allowed for preallocation, | |
1009 | new blocks are displaced based on directory ID. Also, if suggested search_start | |
1010 | is less than last preallocated block, we start searching from it, assuming that | |
1011 | HDD dataflow is faster in forward direction */ | |
1012 | if (TEST_OPTION(old_way, s)) { | |
1013 | if (!hint->formatted_node) { | |
1014 | if (!reiserfs_hashed_relocation(s)) | |
1015 | old_way(hint); | |
1016 | else if (!reiserfs_no_unhashed_relocation(s)) | |
1017 | old_hashed_relocation(hint); | |
1018 | ||
1019 | if (hint->inode | |
1020 | && hint->search_start < | |
1021 | REISERFS_I(hint->inode)->i_prealloc_block) | |
1022 | hint->search_start = | |
1023 | REISERFS_I(hint->inode)->i_prealloc_block; | |
1024 | } | |
1025 | return; | |
1da177e4 | 1026 | } |
1da177e4 | 1027 | |
bd4c625c LT |
1028 | /* This is an approach proposed by Hans */ |
1029 | if (TEST_OPTION(hundredth_slices, s) | |
1030 | && !(displacing_large_files(s) && !hint->formatted_node)) { | |
1031 | hundredth_slices(hint); | |
1032 | return; | |
1033 | } | |
1da177e4 | 1034 | |
bd4c625c LT |
1035 | /* old_hashed_relocation only works on unformatted */ |
1036 | if (!unfm_hint && !hint->formatted_node && | |
1037 | TEST_OPTION(old_hashed_relocation, s)) { | |
1038 | old_hashed_relocation(hint); | |
1039 | } | |
1040 | /* new_hashed_relocation works with both formatted/unformatted nodes */ | |
1041 | if ((!unfm_hint || hint->formatted_node) && | |
1042 | TEST_OPTION(new_hashed_relocation, s)) { | |
1043 | new_hashed_relocation(hint); | |
1044 | } | |
1045 | /* dirid grouping works only on unformatted nodes */ | |
1046 | if (!unfm_hint && !hint->formatted_node && TEST_OPTION(dirid_groups, s)) { | |
1047 | dirid_groups(hint); | |
1048 | } | |
1da177e4 | 1049 | #ifdef DISPLACE_NEW_PACKING_LOCALITIES |
bd4c625c LT |
1050 | if (hint->formatted_node && TEST_OPTION(dirid_groups, s)) { |
1051 | dirid_groups(hint); | |
1052 | } | |
1da177e4 LT |
1053 | #endif |
1054 | ||
bd4c625c LT |
1055 | /* oid grouping works only on unformatted nodes */ |
1056 | if (!unfm_hint && !hint->formatted_node && TEST_OPTION(oid_groups, s)) { | |
1057 | oid_groups(hint); | |
1058 | } | |
1059 | return; | |
1da177e4 LT |
1060 | } |
1061 | ||
1062 | static int determine_prealloc_size(reiserfs_blocknr_hint_t * hint) | |
1063 | { | |
bd4c625c LT |
1064 | /* make minimum size a mount option and benchmark both ways */ |
1065 | /* we preallocate blocks only for regular files, specific size */ | |
1066 | /* benchmark preallocating always and see what happens */ | |
1067 | ||
1068 | hint->prealloc_size = 0; | |
1069 | ||
1070 | if (!hint->formatted_node && hint->preallocate) { | |
1071 | if (S_ISREG(hint->inode->i_mode) | |
1072 | && hint->inode->i_size >= | |
1073 | REISERFS_SB(hint->th->t_super)->s_alloc_options. | |
1074 | preallocmin * hint->inode->i_sb->s_blocksize) | |
1075 | hint->prealloc_size = | |
1076 | REISERFS_SB(hint->th->t_super)->s_alloc_options. | |
1077 | preallocsize - 1; | |
1078 | } | |
1079 | return CARRY_ON; | |
1da177e4 LT |
1080 | } |
1081 | ||
1082 | /* XXX I know it could be merged with upper-level function; | |
1083 | but may be result function would be too complex. */ | |
bd4c625c LT |
1084 | static inline int allocate_without_wrapping_disk(reiserfs_blocknr_hint_t * hint, |
1085 | b_blocknr_t * new_blocknrs, | |
1086 | b_blocknr_t start, | |
1087 | b_blocknr_t finish, int min, | |
1088 | int amount_needed, | |
1089 | int prealloc_size) | |
1da177e4 | 1090 | { |
bd4c625c LT |
1091 | int rest = amount_needed; |
1092 | int nr_allocated; | |
1093 | ||
1094 | while (rest > 0 && start <= finish) { | |
1095 | nr_allocated = scan_bitmap(hint->th, &start, finish, min, | |
1096 | rest + prealloc_size, | |
1097 | !hint->formatted_node, hint->block); | |
1098 | ||
1099 | if (nr_allocated == 0) /* no new blocks allocated, return */ | |
1100 | break; | |
1101 | ||
1102 | /* fill free_blocknrs array first */ | |
1103 | while (rest > 0 && nr_allocated > 0) { | |
1104 | *new_blocknrs++ = start++; | |
1105 | rest--; | |
1106 | nr_allocated--; | |
1107 | } | |
1da177e4 | 1108 | |
bd4c625c LT |
1109 | /* do we have something to fill prealloc. array also ? */ |
1110 | if (nr_allocated > 0) { | |
1111 | /* it means prealloc_size was greater that 0 and we do preallocation */ | |
1112 | list_add(&REISERFS_I(hint->inode)->i_prealloc_list, | |
1113 | &SB_JOURNAL(hint->th->t_super)-> | |
1114 | j_prealloc_list); | |
1115 | REISERFS_I(hint->inode)->i_prealloc_block = start; | |
1116 | REISERFS_I(hint->inode)->i_prealloc_count = | |
1117 | nr_allocated; | |
1118 | break; | |
1119 | } | |
1da177e4 | 1120 | } |
1da177e4 | 1121 | |
bd4c625c | 1122 | return (amount_needed - rest); |
1da177e4 LT |
1123 | } |
1124 | ||
1125 | static inline int blocknrs_and_prealloc_arrays_from_search_start | |
bd4c625c LT |
1126 | (reiserfs_blocknr_hint_t * hint, b_blocknr_t * new_blocknrs, |
1127 | int amount_needed) { | |
1128 | struct super_block *s = hint->th->t_super; | |
1129 | b_blocknr_t start = hint->search_start; | |
1130 | b_blocknr_t finish = SB_BLOCK_COUNT(s) - 1; | |
1131 | int passno = 0; | |
1132 | int nr_allocated = 0; | |
bd4c625c LT |
1133 | |
1134 | determine_prealloc_size(hint); | |
1135 | if (!hint->formatted_node) { | |
1136 | int quota_ret; | |
1da177e4 | 1137 | #ifdef REISERQUOTA_DEBUG |
bd4c625c LT |
1138 | reiserfs_debug(s, REISERFS_DEBUG_CODE, |
1139 | "reiserquota: allocating %d blocks id=%u", | |
1140 | amount_needed, hint->inode->i_uid); | |
1da177e4 | 1141 | #endif |
bd4c625c | 1142 | quota_ret = |
5dd4056d | 1143 | dquot_alloc_block_nodirty(hint->inode, amount_needed); |
bd4c625c LT |
1144 | if (quota_ret) /* Quota exceeded? */ |
1145 | return QUOTA_EXCEEDED; | |
1146 | if (hint->preallocate && hint->prealloc_size) { | |
1da177e4 | 1147 | #ifdef REISERQUOTA_DEBUG |
bd4c625c LT |
1148 | reiserfs_debug(s, REISERFS_DEBUG_CODE, |
1149 | "reiserquota: allocating (prealloc) %d blocks id=%u", | |
1150 | hint->prealloc_size, hint->inode->i_uid); | |
1da177e4 | 1151 | #endif |
5dd4056d | 1152 | quota_ret = dquot_prealloc_block_nodirty(hint->inode, |
bd4c625c LT |
1153 | hint->prealloc_size); |
1154 | if (quota_ret) | |
1155 | hint->preallocate = hint->prealloc_size = 0; | |
1156 | } | |
1157 | /* for unformatted nodes, force large allocations */ | |
1da177e4 | 1158 | } |
1da177e4 | 1159 | |
bd4c625c | 1160 | do { |
bd4c625c LT |
1161 | switch (passno++) { |
1162 | case 0: /* Search from hint->search_start to end of disk */ | |
1163 | start = hint->search_start; | |
1164 | finish = SB_BLOCK_COUNT(s) - 1; | |
1165 | break; | |
1166 | case 1: /* Search from hint->beg to hint->search_start */ | |
1167 | start = hint->beg; | |
1168 | finish = hint->search_start; | |
1169 | break; | |
1170 | case 2: /* Last chance: Search from 0 to hint->beg */ | |
1171 | start = 0; | |
1172 | finish = hint->beg; | |
1173 | break; | |
1174 | default: /* We've tried searching everywhere, not enough space */ | |
1175 | /* Free the blocks */ | |
1176 | if (!hint->formatted_node) { | |
1da177e4 | 1177 | #ifdef REISERQUOTA_DEBUG |
bd4c625c LT |
1178 | reiserfs_debug(s, REISERFS_DEBUG_CODE, |
1179 | "reiserquota: freeing (nospace) %d blocks id=%u", | |
1180 | amount_needed + | |
1181 | hint->prealloc_size - | |
1182 | nr_allocated, | |
1183 | hint->inode->i_uid); | |
1da177e4 | 1184 | #endif |
77db4f25 | 1185 | /* Free not allocated blocks */ |
5dd4056d | 1186 | dquot_free_block_nodirty(hint->inode, |
77db4f25 JK |
1187 | amount_needed + hint->prealloc_size - |
1188 | nr_allocated); | |
bd4c625c LT |
1189 | } |
1190 | while (nr_allocated--) | |
1191 | reiserfs_free_block(hint->th, hint->inode, | |
1192 | new_blocknrs[nr_allocated], | |
1193 | !hint->formatted_node); | |
1194 | ||
1195 | return NO_DISK_SPACE; | |
1196 | } | |
1197 | } while ((nr_allocated += allocate_without_wrapping_disk(hint, | |
1198 | new_blocknrs + | |
1199 | nr_allocated, | |
1200 | start, finish, | |
9ea0f949 | 1201 | 1, |
bd4c625c LT |
1202 | amount_needed - |
1203 | nr_allocated, | |
1204 | hint-> | |
1205 | prealloc_size)) | |
1206 | < amount_needed); | |
1207 | if (!hint->formatted_node && | |
1208 | amount_needed + hint->prealloc_size > | |
1209 | nr_allocated + REISERFS_I(hint->inode)->i_prealloc_count) { | |
1210 | /* Some of preallocation blocks were not allocated */ | |
1da177e4 | 1211 | #ifdef REISERQUOTA_DEBUG |
bd4c625c LT |
1212 | reiserfs_debug(s, REISERFS_DEBUG_CODE, |
1213 | "reiserquota: freeing (failed prealloc) %d blocks id=%u", | |
1214 | amount_needed + hint->prealloc_size - | |
1215 | nr_allocated - | |
1216 | REISERFS_I(hint->inode)->i_prealloc_count, | |
1217 | hint->inode->i_uid); | |
1da177e4 | 1218 | #endif |
5dd4056d | 1219 | dquot_free_block_nodirty(hint->inode, amount_needed + |
bd4c625c LT |
1220 | hint->prealloc_size - nr_allocated - |
1221 | REISERFS_I(hint->inode)-> | |
1222 | i_prealloc_count); | |
1223 | } | |
1da177e4 | 1224 | |
bd4c625c | 1225 | return CARRY_ON; |
1da177e4 LT |
1226 | } |
1227 | ||
1228 | /* grab new blocknrs from preallocated list */ | |
1229 | /* return amount still needed after using them */ | |
bd4c625c LT |
1230 | static int use_preallocated_list_if_available(reiserfs_blocknr_hint_t * hint, |
1231 | b_blocknr_t * new_blocknrs, | |
1232 | int amount_needed) | |
1da177e4 | 1233 | { |
bd4c625c | 1234 | struct inode *inode = hint->inode; |
1da177e4 | 1235 | |
bd4c625c LT |
1236 | if (REISERFS_I(inode)->i_prealloc_count > 0) { |
1237 | while (amount_needed) { | |
1da177e4 | 1238 | |
bd4c625c LT |
1239 | *new_blocknrs++ = REISERFS_I(inode)->i_prealloc_block++; |
1240 | REISERFS_I(inode)->i_prealloc_count--; | |
1da177e4 | 1241 | |
bd4c625c | 1242 | amount_needed--; |
1da177e4 | 1243 | |
bd4c625c LT |
1244 | if (REISERFS_I(inode)->i_prealloc_count <= 0) { |
1245 | list_del(&REISERFS_I(inode)->i_prealloc_list); | |
1246 | break; | |
1247 | } | |
1248 | } | |
1da177e4 | 1249 | } |
bd4c625c LT |
1250 | /* return amount still needed after using preallocated blocks */ |
1251 | return amount_needed; | |
1da177e4 LT |
1252 | } |
1253 | ||
bd4c625c LT |
1254 | int reiserfs_allocate_blocknrs(reiserfs_blocknr_hint_t * hint, b_blocknr_t * new_blocknrs, int amount_needed, int reserved_by_us /* Amount of blocks we have |
1255 | already reserved */ ) | |
1da177e4 | 1256 | { |
bd4c625c LT |
1257 | int initial_amount_needed = amount_needed; |
1258 | int ret; | |
1259 | struct super_block *s = hint->th->t_super; | |
1260 | ||
1261 | /* Check if there is enough space, taking into account reserved space */ | |
1262 | if (SB_FREE_BLOCKS(s) - REISERFS_SB(s)->reserved_blocks < | |
1263 | amount_needed - reserved_by_us) | |
1264 | return NO_DISK_SPACE; | |
1265 | /* should this be if !hint->inode && hint->preallocate? */ | |
1266 | /* do you mean hint->formatted_node can be removed ? - Zam */ | |
1267 | /* hint->formatted_node cannot be removed because we try to access | |
1268 | inode information here, and there is often no inode assotiated with | |
1269 | metadata allocations - green */ | |
1270 | ||
1271 | if (!hint->formatted_node && hint->preallocate) { | |
1272 | amount_needed = use_preallocated_list_if_available | |
1273 | (hint, new_blocknrs, amount_needed); | |
1274 | if (amount_needed == 0) /* all blocknrs we need we got from | |
1275 | prealloc. list */ | |
1276 | return CARRY_ON; | |
1277 | new_blocknrs += (initial_amount_needed - amount_needed); | |
1278 | } | |
1279 | ||
1280 | /* find search start and save it in hint structure */ | |
1281 | determine_search_start(hint, amount_needed); | |
1282 | if (hint->search_start >= SB_BLOCK_COUNT(s)) | |
1283 | hint->search_start = SB_BLOCK_COUNT(s) - 1; | |
1284 | ||
1285 | /* allocation itself; fill new_blocknrs and preallocation arrays */ | |
1286 | ret = blocknrs_and_prealloc_arrays_from_search_start | |
1da177e4 | 1287 | (hint, new_blocknrs, amount_needed); |
bd4c625c LT |
1288 | |
1289 | /* we used prealloc. list to fill (partially) new_blocknrs array. If final allocation fails we | |
1290 | * need to return blocks back to prealloc. list or just free them. -- Zam (I chose second | |
1291 | * variant) */ | |
1292 | ||
1293 | if (ret != CARRY_ON) { | |
1294 | while (amount_needed++ < initial_amount_needed) { | |
1295 | reiserfs_free_block(hint->th, hint->inode, | |
1296 | *(--new_blocknrs), 1); | |
1297 | } | |
1da177e4 | 1298 | } |
bd4c625c | 1299 | return ret; |
1da177e4 LT |
1300 | } |
1301 | ||
6f01046b JM |
1302 | void reiserfs_cache_bitmap_metadata(struct super_block *sb, |
1303 | struct buffer_head *bh, | |
1304 | struct reiserfs_bitmap_info *info) | |
1305 | { | |
1306 | unsigned long *cur = (unsigned long *)(bh->b_data + bh->b_size); | |
1307 | ||
4d20851d | 1308 | /* The first bit must ALWAYS be 1 */ |
0030b645 JM |
1309 | if (!reiserfs_test_le_bit(0, (unsigned long *)bh->b_data)) |
1310 | reiserfs_error(sb, "reiserfs-2025", "bitmap block %lu is " | |
1311 | "corrupted: first bit must be 1", bh->b_blocknr); | |
4d20851d JM |
1312 | |
1313 | info->free_count = 0; | |
6f01046b JM |
1314 | |
1315 | while (--cur >= (unsigned long *)bh->b_data) { | |
6f01046b | 1316 | /* 0 and ~0 are special, we can optimize for them */ |
4d20851d | 1317 | if (*cur == 0) |
6f01046b | 1318 | info->free_count += BITS_PER_LONG; |
4d20851d | 1319 | else if (*cur != ~0L) /* A mix, investigate */ |
9d6bf5aa | 1320 | info->free_count += BITS_PER_LONG - hweight_long(*cur); |
6f01046b | 1321 | } |
6f01046b JM |
1322 | } |
1323 | ||
1324 | struct buffer_head *reiserfs_read_bitmap_block(struct super_block *sb, | |
1325 | unsigned int bitmap) | |
1326 | { | |
1327 | b_blocknr_t block = (sb->s_blocksize << 3) * bitmap; | |
5065227b | 1328 | struct reiserfs_bitmap_info *info = SB_AP_BITMAP(sb) + bitmap; |
6f01046b JM |
1329 | struct buffer_head *bh; |
1330 | ||
1331 | /* Way old format filesystems had the bitmaps packed up front. | |
1332 | * I doubt there are any of these left, but just in case... */ | |
1333 | if (unlikely(test_bit(REISERFS_OLD_FORMAT, | |
1334 | &(REISERFS_SB(sb)->s_properties)))) | |
1335 | block = REISERFS_SB(sb)->s_sbh->b_blocknr + 1 + bitmap; | |
1336 | else if (bitmap == 0) | |
1337 | block = (REISERFS_DISK_OFFSET_IN_BYTES >> sb->s_blocksize_bits) + 1; | |
1338 | ||
4c5eface | 1339 | reiserfs_write_unlock(sb); |
5065227b | 1340 | bh = sb_bread(sb, block); |
4c5eface | 1341 | reiserfs_write_lock(sb); |
5065227b | 1342 | if (bh == NULL) |
00079e04 | 1343 | reiserfs_warning(sb, "sh-2029: %s: bitmap block (#%u) " |
fbe5498b | 1344 | "reading failed", __func__, block); |
5065227b JM |
1345 | else { |
1346 | if (buffer_locked(bh)) { | |
1347 | PROC_INFO_INC(sb, scan_bitmap.wait); | |
8ebc4232 | 1348 | reiserfs_write_unlock(sb); |
5065227b | 1349 | __wait_on_buffer(bh); |
8ebc4232 | 1350 | reiserfs_write_lock(sb); |
5065227b JM |
1351 | } |
1352 | BUG_ON(!buffer_uptodate(bh)); | |
1353 | BUG_ON(atomic_read(&bh->b_count) == 0); | |
1354 | ||
4d20851d | 1355 | if (info->free_count == UINT_MAX) |
5065227b JM |
1356 | reiserfs_cache_bitmap_metadata(sb, bh, info); |
1357 | } | |
6f01046b JM |
1358 | |
1359 | return bh; | |
1360 | } | |
1361 | ||
1362 | int reiserfs_init_bitmap_cache(struct super_block *sb) | |
1363 | { | |
1364 | struct reiserfs_bitmap_info *bitmap; | |
cb680c1b | 1365 | unsigned int bmap_nr = reiserfs_bmap_count(sb); |
6f01046b | 1366 | |
cb680c1b | 1367 | bitmap = vmalloc(sizeof(*bitmap) * bmap_nr); |
6f01046b JM |
1368 | if (bitmap == NULL) |
1369 | return -ENOMEM; | |
1370 | ||
cb680c1b | 1371 | memset(bitmap, 0xff, sizeof(*bitmap) * bmap_nr); |
6f01046b | 1372 | |
6f01046b JM |
1373 | SB_AP_BITMAP(sb) = bitmap; |
1374 | ||
1375 | return 0; | |
1376 | } | |
5065227b JM |
1377 | |
1378 | void reiserfs_free_bitmap_cache(struct super_block *sb) | |
1379 | { | |
1380 | if (SB_AP_BITMAP(sb)) { | |
1381 | vfree(SB_AP_BITMAP(sb)); | |
1382 | SB_AP_BITMAP(sb) = NULL; | |
1383 | } | |
1384 | } |