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