Commit | Line | Data |
---|---|---|
a86c6181 AT |
1 | /* |
2 | * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com | |
3 | * Written by Alex Tomas <alex@clusterfs.com> | |
4 | * | |
5 | * Architecture independence: | |
6 | * Copyright (c) 2005, Bull S.A. | |
7 | * Written by Pierre Peiffer <pierre.peiffer@bull.net> | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License version 2 as | |
11 | * published by the Free Software Foundation. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | * GNU General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public Licens | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- | |
21 | */ | |
22 | ||
23 | /* | |
24 | * Extents support for EXT4 | |
25 | * | |
26 | * TODO: | |
27 | * - ext4*_error() should be used in some situations | |
28 | * - analyze all BUG()/BUG_ON(), use -EIO where appropriate | |
29 | * - smart tree reduction | |
30 | */ | |
31 | ||
32 | #include <linux/module.h> | |
33 | #include <linux/fs.h> | |
34 | #include <linux/time.h> | |
35 | #include <linux/ext4_jbd2.h> | |
36 | #include <linux/jbd.h> | |
37 | #include <linux/smp_lock.h> | |
38 | #include <linux/highuid.h> | |
39 | #include <linux/pagemap.h> | |
40 | #include <linux/quotaops.h> | |
41 | #include <linux/string.h> | |
42 | #include <linux/slab.h> | |
43 | #include <linux/ext4_fs_extents.h> | |
44 | #include <asm/uaccess.h> | |
45 | ||
46 | ||
f65e6fba AT |
47 | /* this macro combines low and hi parts of phys. blocknr into ext4_fsblk_t */ |
48 | static inline ext4_fsblk_t ext_pblock(struct ext4_extent *ex) | |
49 | { | |
50 | ext4_fsblk_t block; | |
51 | ||
52 | block = le32_to_cpu(ex->ee_start); | |
53 | if (sizeof(ext4_fsblk_t) > 4) | |
54 | block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1; | |
55 | return block; | |
56 | } | |
57 | ||
58 | /* this macro combines low and hi parts of phys. blocknr into ext4_fsblk_t */ | |
59 | static inline ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix) | |
60 | { | |
61 | ext4_fsblk_t block; | |
62 | ||
63 | block = le32_to_cpu(ix->ei_leaf); | |
64 | if (sizeof(ext4_fsblk_t) > 4) | |
65 | block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1; | |
66 | return block; | |
67 | } | |
68 | ||
69 | /* the routine stores large phys. blocknr into extent breaking it into parts */ | |
70 | static inline void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb) | |
71 | { | |
72 | ex->ee_start = cpu_to_le32((unsigned long) (pb & 0xffffffff)); | |
73 | if (sizeof(ext4_fsblk_t) > 4) | |
74 | ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); | |
75 | } | |
76 | ||
77 | /* the routine stores large phys. blocknr into index breaking it into parts */ | |
78 | static inline void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb) | |
79 | { | |
80 | ix->ei_leaf = cpu_to_le32((unsigned long) (pb & 0xffffffff)); | |
81 | if (sizeof(ext4_fsblk_t) > 4) | |
82 | ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); | |
83 | } | |
84 | ||
a86c6181 AT |
85 | static int ext4_ext_check_header(const char *function, struct inode *inode, |
86 | struct ext4_extent_header *eh) | |
87 | { | |
88 | const char *error_msg = NULL; | |
89 | ||
90 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
91 | error_msg = "invalid magic"; | |
92 | goto corrupted; | |
93 | } | |
94 | if (unlikely(eh->eh_max == 0)) { | |
95 | error_msg = "invalid eh_max"; | |
96 | goto corrupted; | |
97 | } | |
98 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
99 | error_msg = "invalid eh_entries"; | |
100 | goto corrupted; | |
101 | } | |
102 | return 0; | |
103 | ||
104 | corrupted: | |
105 | ext4_error(inode->i_sb, function, | |
106 | "bad header in inode #%lu: %s - magic %x, " | |
107 | "entries %u, max %u, depth %u", | |
108 | inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic), | |
109 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | |
110 | le16_to_cpu(eh->eh_depth)); | |
111 | ||
112 | return -EIO; | |
113 | } | |
114 | ||
115 | static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed) | |
116 | { | |
117 | int err; | |
118 | ||
119 | if (handle->h_buffer_credits > needed) | |
120 | return handle; | |
121 | if (!ext4_journal_extend(handle, needed)) | |
122 | return handle; | |
123 | err = ext4_journal_restart(handle, needed); | |
124 | ||
125 | return handle; | |
126 | } | |
127 | ||
128 | /* | |
129 | * could return: | |
130 | * - EROFS | |
131 | * - ENOMEM | |
132 | */ | |
133 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
134 | struct ext4_ext_path *path) | |
135 | { | |
136 | if (path->p_bh) { | |
137 | /* path points to block */ | |
138 | return ext4_journal_get_write_access(handle, path->p_bh); | |
139 | } | |
140 | /* path points to leaf/index in inode body */ | |
141 | /* we use in-core data, no need to protect them */ | |
142 | return 0; | |
143 | } | |
144 | ||
145 | /* | |
146 | * could return: | |
147 | * - EROFS | |
148 | * - ENOMEM | |
149 | * - EIO | |
150 | */ | |
151 | static int ext4_ext_dirty(handle_t *handle, struct inode *inode, | |
152 | struct ext4_ext_path *path) | |
153 | { | |
154 | int err; | |
155 | if (path->p_bh) { | |
156 | /* path points to block */ | |
157 | err = ext4_journal_dirty_metadata(handle, path->p_bh); | |
158 | } else { | |
159 | /* path points to leaf/index in inode body */ | |
160 | err = ext4_mark_inode_dirty(handle, inode); | |
161 | } | |
162 | return err; | |
163 | } | |
164 | ||
f65e6fba | 165 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 166 | struct ext4_ext_path *path, |
f65e6fba | 167 | ext4_fsblk_t block) |
a86c6181 AT |
168 | { |
169 | struct ext4_inode_info *ei = EXT4_I(inode); | |
f65e6fba AT |
170 | ext4_fsblk_t bg_start; |
171 | ext4_grpblk_t colour; | |
a86c6181 AT |
172 | int depth; |
173 | ||
174 | if (path) { | |
175 | struct ext4_extent *ex; | |
176 | depth = path->p_depth; | |
177 | ||
178 | /* try to predict block placement */ | |
179 | if ((ex = path[depth].p_ext)) | |
f65e6fba | 180 | return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block)); |
a86c6181 AT |
181 | |
182 | /* it looks index is empty | |
183 | * try to find starting from index itself */ | |
184 | if (path[depth].p_bh) | |
185 | return path[depth].p_bh->b_blocknr; | |
186 | } | |
187 | ||
188 | /* OK. use inode's group */ | |
189 | bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) + | |
190 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block); | |
191 | colour = (current->pid % 16) * | |
192 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); | |
193 | return bg_start + colour + block; | |
194 | } | |
195 | ||
f65e6fba | 196 | static ext4_fsblk_t |
a86c6181 AT |
197 | ext4_ext_new_block(handle_t *handle, struct inode *inode, |
198 | struct ext4_ext_path *path, | |
199 | struct ext4_extent *ex, int *err) | |
200 | { | |
f65e6fba | 201 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
202 | |
203 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
204 | newblock = ext4_new_block(handle, inode, goal, err); | |
205 | return newblock; | |
206 | } | |
207 | ||
208 | static inline int ext4_ext_space_block(struct inode *inode) | |
209 | { | |
210 | int size; | |
211 | ||
212 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
213 | / sizeof(struct ext4_extent); | |
214 | #ifdef AGRESSIVE_TEST | |
215 | if (size > 6) | |
216 | size = 6; | |
217 | #endif | |
218 | return size; | |
219 | } | |
220 | ||
221 | static inline int ext4_ext_space_block_idx(struct inode *inode) | |
222 | { | |
223 | int size; | |
224 | ||
225 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
226 | / sizeof(struct ext4_extent_idx); | |
227 | #ifdef AGRESSIVE_TEST | |
228 | if (size > 5) | |
229 | size = 5; | |
230 | #endif | |
231 | return size; | |
232 | } | |
233 | ||
234 | static inline int ext4_ext_space_root(struct inode *inode) | |
235 | { | |
236 | int size; | |
237 | ||
238 | size = sizeof(EXT4_I(inode)->i_data); | |
239 | size -= sizeof(struct ext4_extent_header); | |
240 | size /= sizeof(struct ext4_extent); | |
241 | #ifdef AGRESSIVE_TEST | |
242 | if (size > 3) | |
243 | size = 3; | |
244 | #endif | |
245 | return size; | |
246 | } | |
247 | ||
248 | static inline int ext4_ext_space_root_idx(struct inode *inode) | |
249 | { | |
250 | int size; | |
251 | ||
252 | size = sizeof(EXT4_I(inode)->i_data); | |
253 | size -= sizeof(struct ext4_extent_header); | |
254 | size /= sizeof(struct ext4_extent_idx); | |
255 | #ifdef AGRESSIVE_TEST | |
256 | if (size > 4) | |
257 | size = 4; | |
258 | #endif | |
259 | return size; | |
260 | } | |
261 | ||
262 | #ifdef EXT_DEBUG | |
263 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
264 | { | |
265 | int k, l = path->p_depth; | |
266 | ||
267 | ext_debug("path:"); | |
268 | for (k = 0; k <= l; k++, path++) { | |
269 | if (path->p_idx) { | |
f65e6fba AT |
270 | ext_debug(" %d->"E3FSBLK, le32_to_cpu(path->p_idx->ei_block), |
271 | idx_pblock(path->p_idx)); | |
a86c6181 | 272 | } else if (path->p_ext) { |
f65e6fba | 273 | ext_debug(" %d:%d:"E3FSBLK" ", |
a86c6181 AT |
274 | le32_to_cpu(path->p_ext->ee_block), |
275 | le16_to_cpu(path->p_ext->ee_len), | |
f65e6fba | 276 | ext_pblock(path->p_ext)); |
a86c6181 AT |
277 | } else |
278 | ext_debug(" []"); | |
279 | } | |
280 | ext_debug("\n"); | |
281 | } | |
282 | ||
283 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
284 | { | |
285 | int depth = ext_depth(inode); | |
286 | struct ext4_extent_header *eh; | |
287 | struct ext4_extent *ex; | |
288 | int i; | |
289 | ||
290 | if (!path) | |
291 | return; | |
292 | ||
293 | eh = path[depth].p_hdr; | |
294 | ex = EXT_FIRST_EXTENT(eh); | |
295 | ||
296 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { | |
f65e6fba AT |
297 | ext_debug("%d:%d:"E3FSBLK" ", le32_to_cpu(ex->ee_block), |
298 | le16_to_cpu(ex->ee_len), ext_pblock(ex)); | |
a86c6181 AT |
299 | } |
300 | ext_debug("\n"); | |
301 | } | |
302 | #else | |
303 | #define ext4_ext_show_path(inode,path) | |
304 | #define ext4_ext_show_leaf(inode,path) | |
305 | #endif | |
306 | ||
307 | static void ext4_ext_drop_refs(struct ext4_ext_path *path) | |
308 | { | |
309 | int depth = path->p_depth; | |
310 | int i; | |
311 | ||
312 | for (i = 0; i <= depth; i++, path++) | |
313 | if (path->p_bh) { | |
314 | brelse(path->p_bh); | |
315 | path->p_bh = NULL; | |
316 | } | |
317 | } | |
318 | ||
319 | /* | |
320 | * binary search for closest index by given block | |
321 | */ | |
322 | static void | |
323 | ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int block) | |
324 | { | |
325 | struct ext4_extent_header *eh = path->p_hdr; | |
326 | struct ext4_extent_idx *r, *l, *m; | |
327 | ||
328 | BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC); | |
329 | BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max)); | |
330 | BUG_ON(le16_to_cpu(eh->eh_entries) <= 0); | |
331 | ||
332 | ext_debug("binsearch for %d(idx): ", block); | |
333 | ||
334 | l = EXT_FIRST_INDEX(eh) + 1; | |
335 | r = EXT_FIRST_INDEX(eh) + le16_to_cpu(eh->eh_entries) - 1; | |
336 | while (l <= r) { | |
337 | m = l + (r - l) / 2; | |
338 | if (block < le32_to_cpu(m->ei_block)) | |
339 | r = m - 1; | |
340 | else | |
341 | l = m + 1; | |
342 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ei_block, | |
343 | m, m->ei_block, r, r->ei_block); | |
344 | } | |
345 | ||
346 | path->p_idx = l - 1; | |
f65e6fba AT |
347 | ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), |
348 | idx_block(path->p_idx)); | |
a86c6181 AT |
349 | |
350 | #ifdef CHECK_BINSEARCH | |
351 | { | |
352 | struct ext4_extent_idx *chix, *ix; | |
353 | int k; | |
354 | ||
355 | chix = ix = EXT_FIRST_INDEX(eh); | |
356 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
357 | if (k != 0 && | |
358 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
359 | printk("k=%d, ix=0x%p, first=0x%p\n", k, | |
360 | ix, EXT_FIRST_INDEX(eh)); | |
361 | printk("%u <= %u\n", | |
362 | le32_to_cpu(ix->ei_block), | |
363 | le32_to_cpu(ix[-1].ei_block)); | |
364 | } | |
365 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
366 | <= le32_to_cpu(ix[-1].ei_block)); | |
367 | if (block < le32_to_cpu(ix->ei_block)) | |
368 | break; | |
369 | chix = ix; | |
370 | } | |
371 | BUG_ON(chix != path->p_idx); | |
372 | } | |
373 | #endif | |
374 | ||
375 | } | |
376 | ||
377 | /* | |
378 | * binary search for closest extent by given block | |
379 | */ | |
380 | static void | |
381 | ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block) | |
382 | { | |
383 | struct ext4_extent_header *eh = path->p_hdr; | |
384 | struct ext4_extent *r, *l, *m; | |
385 | ||
386 | BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC); | |
387 | BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max)); | |
388 | ||
389 | if (eh->eh_entries == 0) { | |
390 | /* | |
391 | * this leaf is empty yet: | |
392 | * we get such a leaf in split/add case | |
393 | */ | |
394 | return; | |
395 | } | |
396 | ||
397 | ext_debug("binsearch for %d: ", block); | |
398 | ||
399 | l = EXT_FIRST_EXTENT(eh) + 1; | |
400 | r = EXT_FIRST_EXTENT(eh) + le16_to_cpu(eh->eh_entries) - 1; | |
401 | ||
402 | while (l <= r) { | |
403 | m = l + (r - l) / 2; | |
404 | if (block < le32_to_cpu(m->ee_block)) | |
405 | r = m - 1; | |
406 | else | |
407 | l = m + 1; | |
408 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ee_block, | |
409 | m, m->ee_block, r, r->ee_block); | |
410 | } | |
411 | ||
412 | path->p_ext = l - 1; | |
f65e6fba | 413 | ext_debug(" -> %d:"E3FSBLK":%d ", |
a86c6181 | 414 | le32_to_cpu(path->p_ext->ee_block), |
f65e6fba AT |
415 | ext_pblock(path->p_ext), |
416 | le16_to_cpu(path->p_ext->ee_len)); | |
a86c6181 AT |
417 | |
418 | #ifdef CHECK_BINSEARCH | |
419 | { | |
420 | struct ext4_extent *chex, *ex; | |
421 | int k; | |
422 | ||
423 | chex = ex = EXT_FIRST_EXTENT(eh); | |
424 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
425 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
426 | <= le32_to_cpu(ex[-1].ee_block)); | |
427 | if (block < le32_to_cpu(ex->ee_block)) | |
428 | break; | |
429 | chex = ex; | |
430 | } | |
431 | BUG_ON(chex != path->p_ext); | |
432 | } | |
433 | #endif | |
434 | ||
435 | } | |
436 | ||
437 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
438 | { | |
439 | struct ext4_extent_header *eh; | |
440 | ||
441 | eh = ext_inode_hdr(inode); | |
442 | eh->eh_depth = 0; | |
443 | eh->eh_entries = 0; | |
444 | eh->eh_magic = EXT4_EXT_MAGIC; | |
445 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode)); | |
446 | ext4_mark_inode_dirty(handle, inode); | |
447 | ext4_ext_invalidate_cache(inode); | |
448 | return 0; | |
449 | } | |
450 | ||
451 | struct ext4_ext_path * | |
452 | ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path) | |
453 | { | |
454 | struct ext4_extent_header *eh; | |
455 | struct buffer_head *bh; | |
456 | short int depth, i, ppos = 0, alloc = 0; | |
457 | ||
458 | eh = ext_inode_hdr(inode); | |
459 | BUG_ON(eh == NULL); | |
460 | if (ext4_ext_check_header(__FUNCTION__, inode, eh)) | |
461 | return ERR_PTR(-EIO); | |
462 | ||
463 | i = depth = ext_depth(inode); | |
464 | ||
465 | /* account possible depth increase */ | |
466 | if (!path) { | |
467 | path = kmalloc(sizeof(struct ext4_ext_path) * (depth + 2), | |
468 | GFP_NOFS); | |
469 | if (!path) | |
470 | return ERR_PTR(-ENOMEM); | |
471 | alloc = 1; | |
472 | } | |
473 | memset(path, 0, sizeof(struct ext4_ext_path) * (depth + 1)); | |
474 | path[0].p_hdr = eh; | |
475 | ||
476 | /* walk through the tree */ | |
477 | while (i) { | |
478 | ext_debug("depth %d: num %d, max %d\n", | |
479 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
480 | ext4_ext_binsearch_idx(inode, path + ppos, block); | |
f65e6fba | 481 | path[ppos].p_block = idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
482 | path[ppos].p_depth = i; |
483 | path[ppos].p_ext = NULL; | |
484 | ||
485 | bh = sb_bread(inode->i_sb, path[ppos].p_block); | |
486 | if (!bh) | |
487 | goto err; | |
488 | ||
489 | eh = ext_block_hdr(bh); | |
490 | ppos++; | |
491 | BUG_ON(ppos > depth); | |
492 | path[ppos].p_bh = bh; | |
493 | path[ppos].p_hdr = eh; | |
494 | i--; | |
495 | ||
496 | if (ext4_ext_check_header(__FUNCTION__, inode, eh)) | |
497 | goto err; | |
498 | } | |
499 | ||
500 | path[ppos].p_depth = i; | |
501 | path[ppos].p_hdr = eh; | |
502 | path[ppos].p_ext = NULL; | |
503 | path[ppos].p_idx = NULL; | |
504 | ||
505 | if (ext4_ext_check_header(__FUNCTION__, inode, eh)) | |
506 | goto err; | |
507 | ||
508 | /* find extent */ | |
509 | ext4_ext_binsearch(inode, path + ppos, block); | |
510 | ||
511 | ext4_ext_show_path(inode, path); | |
512 | ||
513 | return path; | |
514 | ||
515 | err: | |
516 | ext4_ext_drop_refs(path); | |
517 | if (alloc) | |
518 | kfree(path); | |
519 | return ERR_PTR(-EIO); | |
520 | } | |
521 | ||
522 | /* | |
523 | * insert new index [logical;ptr] into the block at cupr | |
524 | * it check where to insert: before curp or after curp | |
525 | */ | |
526 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, | |
527 | struct ext4_ext_path *curp, | |
f65e6fba | 528 | int logical, ext4_fsblk_t ptr) |
a86c6181 AT |
529 | { |
530 | struct ext4_extent_idx *ix; | |
531 | int len, err; | |
532 | ||
533 | if ((err = ext4_ext_get_access(handle, inode, curp))) | |
534 | return err; | |
535 | ||
536 | BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block)); | |
537 | len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; | |
538 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { | |
539 | /* insert after */ | |
540 | if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { | |
541 | len = (len - 1) * sizeof(struct ext4_extent_idx); | |
542 | len = len < 0 ? 0 : len; | |
543 | ext_debug("insert new index %d after: %d. " | |
544 | "move %d from 0x%p to 0x%p\n", | |
545 | logical, ptr, len, | |
546 | (curp->p_idx + 1), (curp->p_idx + 2)); | |
547 | memmove(curp->p_idx + 2, curp->p_idx + 1, len); | |
548 | } | |
549 | ix = curp->p_idx + 1; | |
550 | } else { | |
551 | /* insert before */ | |
552 | len = len * sizeof(struct ext4_extent_idx); | |
553 | len = len < 0 ? 0 : len; | |
554 | ext_debug("insert new index %d before: %d. " | |
555 | "move %d from 0x%p to 0x%p\n", | |
556 | logical, ptr, len, | |
557 | curp->p_idx, (curp->p_idx + 1)); | |
558 | memmove(curp->p_idx + 1, curp->p_idx, len); | |
559 | ix = curp->p_idx; | |
560 | } | |
561 | ||
562 | ix->ei_block = cpu_to_le32(logical); | |
f65e6fba | 563 | ext4_idx_store_pblock(ix, ptr); |
a86c6181 AT |
564 | curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1); |
565 | ||
566 | BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries) | |
567 | > le16_to_cpu(curp->p_hdr->eh_max)); | |
568 | BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr)); | |
569 | ||
570 | err = ext4_ext_dirty(handle, inode, curp); | |
571 | ext4_std_error(inode->i_sb, err); | |
572 | ||
573 | return err; | |
574 | } | |
575 | ||
576 | /* | |
577 | * routine inserts new subtree into the path, using free index entry | |
578 | * at depth 'at: | |
579 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
580 | * - makes decision where to split | |
581 | * - moves remaining extens and index entries (right to the split point) | |
582 | * into the newly allocated blocks | |
583 | * - initialize subtree | |
584 | */ | |
585 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
586 | struct ext4_ext_path *path, | |
587 | struct ext4_extent *newext, int at) | |
588 | { | |
589 | struct buffer_head *bh = NULL; | |
590 | int depth = ext_depth(inode); | |
591 | struct ext4_extent_header *neh; | |
592 | struct ext4_extent_idx *fidx; | |
593 | struct ext4_extent *ex; | |
594 | int i = at, k, m, a; | |
f65e6fba | 595 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 596 | __le32 border; |
f65e6fba | 597 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
598 | int err = 0; |
599 | ||
600 | /* make decision: where to split? */ | |
601 | /* FIXME: now desicion is simplest: at current extent */ | |
602 | ||
603 | /* if current leaf will be splitted, then we should use | |
604 | * border from split point */ | |
605 | BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr)); | |
606 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
607 | border = path[depth].p_ext[1].ee_block; | |
608 | ext_debug("leaf will be splitted." | |
609 | " next leaf starts at %d\n", | |
610 | le32_to_cpu(border)); | |
611 | } else { | |
612 | border = newext->ee_block; | |
613 | ext_debug("leaf will be added." | |
614 | " next leaf starts at %d\n", | |
615 | le32_to_cpu(border)); | |
616 | } | |
617 | ||
618 | /* | |
619 | * if error occurs, then we break processing | |
620 | * and turn filesystem read-only. so, index won't | |
621 | * be inserted and tree will be in consistent | |
622 | * state. next mount will repair buffers too | |
623 | */ | |
624 | ||
625 | /* | |
626 | * get array to track all allocated blocks | |
627 | * we need this to handle errors and free blocks | |
628 | * upon them | |
629 | */ | |
f65e6fba | 630 | ablocks = kmalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
631 | if (!ablocks) |
632 | return -ENOMEM; | |
f65e6fba | 633 | memset(ablocks, 0, sizeof(ext4_fsblk_t) * depth); |
a86c6181 AT |
634 | |
635 | /* allocate all needed blocks */ | |
636 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
637 | for (a = 0; a < depth - at; a++) { | |
638 | newblock = ext4_ext_new_block(handle, inode, path, newext, &err); | |
639 | if (newblock == 0) | |
640 | goto cleanup; | |
641 | ablocks[a] = newblock; | |
642 | } | |
643 | ||
644 | /* initialize new leaf */ | |
645 | newblock = ablocks[--a]; | |
646 | BUG_ON(newblock == 0); | |
647 | bh = sb_getblk(inode->i_sb, newblock); | |
648 | if (!bh) { | |
649 | err = -EIO; | |
650 | goto cleanup; | |
651 | } | |
652 | lock_buffer(bh); | |
653 | ||
654 | if ((err = ext4_journal_get_create_access(handle, bh))) | |
655 | goto cleanup; | |
656 | ||
657 | neh = ext_block_hdr(bh); | |
658 | neh->eh_entries = 0; | |
659 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); | |
660 | neh->eh_magic = EXT4_EXT_MAGIC; | |
661 | neh->eh_depth = 0; | |
662 | ex = EXT_FIRST_EXTENT(neh); | |
663 | ||
664 | /* move remain of path[depth] to the new leaf */ | |
665 | BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max); | |
666 | /* start copy from next extent */ | |
667 | /* TODO: we could do it by single memmove */ | |
668 | m = 0; | |
669 | path[depth].p_ext++; | |
670 | while (path[depth].p_ext <= | |
671 | EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
f65e6fba | 672 | ext_debug("move %d:"E3FSBLK":%d in new leaf "E3FSBLK"\n", |
a86c6181 | 673 | le32_to_cpu(path[depth].p_ext->ee_block), |
f65e6fba | 674 | ext_pblock(path[depth].p_ext), |
a86c6181 AT |
675 | le16_to_cpu(path[depth].p_ext->ee_len), |
676 | newblock); | |
677 | /*memmove(ex++, path[depth].p_ext++, | |
678 | sizeof(struct ext4_extent)); | |
679 | neh->eh_entries++;*/ | |
680 | path[depth].p_ext++; | |
681 | m++; | |
682 | } | |
683 | if (m) { | |
684 | memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m); | |
685 | neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m); | |
686 | } | |
687 | ||
688 | set_buffer_uptodate(bh); | |
689 | unlock_buffer(bh); | |
690 | ||
691 | if ((err = ext4_journal_dirty_metadata(handle, bh))) | |
692 | goto cleanup; | |
693 | brelse(bh); | |
694 | bh = NULL; | |
695 | ||
696 | /* correct old leaf */ | |
697 | if (m) { | |
698 | if ((err = ext4_ext_get_access(handle, inode, path + depth))) | |
699 | goto cleanup; | |
700 | path[depth].p_hdr->eh_entries = | |
701 | cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m); | |
702 | if ((err = ext4_ext_dirty(handle, inode, path + depth))) | |
703 | goto cleanup; | |
704 | ||
705 | } | |
706 | ||
707 | /* create intermediate indexes */ | |
708 | k = depth - at - 1; | |
709 | BUG_ON(k < 0); | |
710 | if (k) | |
711 | ext_debug("create %d intermediate indices\n", k); | |
712 | /* insert new index into current index block */ | |
713 | /* current depth stored in i var */ | |
714 | i = depth - 1; | |
715 | while (k--) { | |
716 | oldblock = newblock; | |
717 | newblock = ablocks[--a]; | |
f65e6fba | 718 | bh = sb_getblk(inode->i_sb, (ext4_fsblk_t)newblock); |
a86c6181 AT |
719 | if (!bh) { |
720 | err = -EIO; | |
721 | goto cleanup; | |
722 | } | |
723 | lock_buffer(bh); | |
724 | ||
725 | if ((err = ext4_journal_get_create_access(handle, bh))) | |
726 | goto cleanup; | |
727 | ||
728 | neh = ext_block_hdr(bh); | |
729 | neh->eh_entries = cpu_to_le16(1); | |
730 | neh->eh_magic = EXT4_EXT_MAGIC; | |
731 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); | |
732 | neh->eh_depth = cpu_to_le16(depth - i); | |
733 | fidx = EXT_FIRST_INDEX(neh); | |
734 | fidx->ei_block = border; | |
f65e6fba | 735 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 736 | |
f65e6fba | 737 | ext_debug("int.index at %d (block "E3FSBLK"): %lu -> "E3FSBLK"\n", i, |
a86c6181 AT |
738 | newblock, (unsigned long) le32_to_cpu(border), |
739 | oldblock); | |
740 | /* copy indexes */ | |
741 | m = 0; | |
742 | path[i].p_idx++; | |
743 | ||
744 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
745 | EXT_MAX_INDEX(path[i].p_hdr)); | |
746 | BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) != | |
747 | EXT_LAST_INDEX(path[i].p_hdr)); | |
748 | while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) { | |
f65e6fba | 749 | ext_debug("%d: move %d:%d in new index "E3FSBLK"\n", i, |
a86c6181 | 750 | le32_to_cpu(path[i].p_idx->ei_block), |
f65e6fba | 751 | idx_pblock(path[i].p_idx), |
a86c6181 AT |
752 | newblock); |
753 | /*memmove(++fidx, path[i].p_idx++, | |
754 | sizeof(struct ext4_extent_idx)); | |
755 | neh->eh_entries++; | |
756 | BUG_ON(neh->eh_entries > neh->eh_max);*/ | |
757 | path[i].p_idx++; | |
758 | m++; | |
759 | } | |
760 | if (m) { | |
761 | memmove(++fidx, path[i].p_idx - m, | |
762 | sizeof(struct ext4_extent_idx) * m); | |
763 | neh->eh_entries = | |
764 | cpu_to_le16(le16_to_cpu(neh->eh_entries) + m); | |
765 | } | |
766 | set_buffer_uptodate(bh); | |
767 | unlock_buffer(bh); | |
768 | ||
769 | if ((err = ext4_journal_dirty_metadata(handle, bh))) | |
770 | goto cleanup; | |
771 | brelse(bh); | |
772 | bh = NULL; | |
773 | ||
774 | /* correct old index */ | |
775 | if (m) { | |
776 | err = ext4_ext_get_access(handle, inode, path + i); | |
777 | if (err) | |
778 | goto cleanup; | |
779 | path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m); | |
780 | err = ext4_ext_dirty(handle, inode, path + i); | |
781 | if (err) | |
782 | goto cleanup; | |
783 | } | |
784 | ||
785 | i--; | |
786 | } | |
787 | ||
788 | /* insert new index */ | |
789 | if (err) | |
790 | goto cleanup; | |
791 | ||
792 | err = ext4_ext_insert_index(handle, inode, path + at, | |
793 | le32_to_cpu(border), newblock); | |
794 | ||
795 | cleanup: | |
796 | if (bh) { | |
797 | if (buffer_locked(bh)) | |
798 | unlock_buffer(bh); | |
799 | brelse(bh); | |
800 | } | |
801 | ||
802 | if (err) { | |
803 | /* free all allocated blocks in error case */ | |
804 | for (i = 0; i < depth; i++) { | |
805 | if (!ablocks[i]) | |
806 | continue; | |
807 | ext4_free_blocks(handle, inode, ablocks[i], 1); | |
808 | } | |
809 | } | |
810 | kfree(ablocks); | |
811 | ||
812 | return err; | |
813 | } | |
814 | ||
815 | /* | |
816 | * routine implements tree growing procedure: | |
817 | * - allocates new block | |
818 | * - moves top-level data (index block or leaf) into the new block | |
819 | * - initialize new top-level, creating index that points to the | |
820 | * just created block | |
821 | */ | |
822 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
823 | struct ext4_ext_path *path, | |
824 | struct ext4_extent *newext) | |
825 | { | |
826 | struct ext4_ext_path *curp = path; | |
827 | struct ext4_extent_header *neh; | |
828 | struct ext4_extent_idx *fidx; | |
829 | struct buffer_head *bh; | |
f65e6fba | 830 | ext4_fsblk_t newblock; |
a86c6181 AT |
831 | int err = 0; |
832 | ||
833 | newblock = ext4_ext_new_block(handle, inode, path, newext, &err); | |
834 | if (newblock == 0) | |
835 | return err; | |
836 | ||
837 | bh = sb_getblk(inode->i_sb, newblock); | |
838 | if (!bh) { | |
839 | err = -EIO; | |
840 | ext4_std_error(inode->i_sb, err); | |
841 | return err; | |
842 | } | |
843 | lock_buffer(bh); | |
844 | ||
845 | if ((err = ext4_journal_get_create_access(handle, bh))) { | |
846 | unlock_buffer(bh); | |
847 | goto out; | |
848 | } | |
849 | ||
850 | /* move top-level index/leaf into new block */ | |
851 | memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data)); | |
852 | ||
853 | /* set size of new block */ | |
854 | neh = ext_block_hdr(bh); | |
855 | /* old root could have indexes or leaves | |
856 | * so calculate e_max right way */ | |
857 | if (ext_depth(inode)) | |
858 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); | |
859 | else | |
860 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); | |
861 | neh->eh_magic = EXT4_EXT_MAGIC; | |
862 | set_buffer_uptodate(bh); | |
863 | unlock_buffer(bh); | |
864 | ||
865 | if ((err = ext4_journal_dirty_metadata(handle, bh))) | |
866 | goto out; | |
867 | ||
868 | /* create index in new top-level index: num,max,pointer */ | |
869 | if ((err = ext4_ext_get_access(handle, inode, curp))) | |
870 | goto out; | |
871 | ||
872 | curp->p_hdr->eh_magic = EXT4_EXT_MAGIC; | |
873 | curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode)); | |
874 | curp->p_hdr->eh_entries = cpu_to_le16(1); | |
875 | curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); | |
876 | /* FIXME: it works, but actually path[0] can be index */ | |
877 | curp->p_idx->ei_block = EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; | |
f65e6fba | 878 | ext4_idx_store_pblock(curp->p_idx, newblock); |
a86c6181 AT |
879 | |
880 | neh = ext_inode_hdr(inode); | |
881 | fidx = EXT_FIRST_INDEX(neh); | |
f65e6fba | 882 | ext_debug("new root: num %d(%d), lblock %d, ptr "E3FSBLK"\n", |
a86c6181 | 883 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
f65e6fba | 884 | le32_to_cpu(fidx->ei_block), idx_pblock(fidx)); |
a86c6181 AT |
885 | |
886 | neh->eh_depth = cpu_to_le16(path->p_depth + 1); | |
887 | err = ext4_ext_dirty(handle, inode, curp); | |
888 | out: | |
889 | brelse(bh); | |
890 | ||
891 | return err; | |
892 | } | |
893 | ||
894 | /* | |
895 | * routine finds empty index and adds new leaf. if no free index found | |
896 | * then it requests in-depth growing | |
897 | */ | |
898 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
899 | struct ext4_ext_path *path, | |
900 | struct ext4_extent *newext) | |
901 | { | |
902 | struct ext4_ext_path *curp; | |
903 | int depth, i, err = 0; | |
904 | ||
905 | repeat: | |
906 | i = depth = ext_depth(inode); | |
907 | ||
908 | /* walk up to the tree and look for free index entry */ | |
909 | curp = path + depth; | |
910 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
911 | i--; | |
912 | curp--; | |
913 | } | |
914 | ||
915 | /* we use already allocated block for index block | |
916 | * so, subsequent data blocks should be contigoues */ | |
917 | if (EXT_HAS_FREE_INDEX(curp)) { | |
918 | /* if we found index with free entry, then use that | |
919 | * entry: create all needed subtree and add new leaf */ | |
920 | err = ext4_ext_split(handle, inode, path, newext, i); | |
921 | ||
922 | /* refill path */ | |
923 | ext4_ext_drop_refs(path); | |
924 | path = ext4_ext_find_extent(inode, | |
925 | le32_to_cpu(newext->ee_block), | |
926 | path); | |
927 | if (IS_ERR(path)) | |
928 | err = PTR_ERR(path); | |
929 | } else { | |
930 | /* tree is full, time to grow in depth */ | |
931 | err = ext4_ext_grow_indepth(handle, inode, path, newext); | |
932 | if (err) | |
933 | goto out; | |
934 | ||
935 | /* refill path */ | |
936 | ext4_ext_drop_refs(path); | |
937 | path = ext4_ext_find_extent(inode, | |
938 | le32_to_cpu(newext->ee_block), | |
939 | path); | |
940 | if (IS_ERR(path)) { | |
941 | err = PTR_ERR(path); | |
942 | goto out; | |
943 | } | |
944 | ||
945 | /* | |
946 | * only first (depth 0 -> 1) produces free space | |
947 | * in all other cases we have to split growed tree | |
948 | */ | |
949 | depth = ext_depth(inode); | |
950 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
951 | /* now we need split */ | |
952 | goto repeat; | |
953 | } | |
954 | } | |
955 | ||
956 | out: | |
957 | return err; | |
958 | } | |
959 | ||
960 | /* | |
961 | * returns allocated block in subsequent extent or EXT_MAX_BLOCK | |
962 | * NOTE: it consider block number from index entry as | |
963 | * allocated block. thus, index entries have to be consistent | |
964 | * with leafs | |
965 | */ | |
966 | static unsigned long | |
967 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) | |
968 | { | |
969 | int depth; | |
970 | ||
971 | BUG_ON(path == NULL); | |
972 | depth = path->p_depth; | |
973 | ||
974 | if (depth == 0 && path->p_ext == NULL) | |
975 | return EXT_MAX_BLOCK; | |
976 | ||
977 | while (depth >= 0) { | |
978 | if (depth == path->p_depth) { | |
979 | /* leaf */ | |
980 | if (path[depth].p_ext != | |
981 | EXT_LAST_EXTENT(path[depth].p_hdr)) | |
982 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
983 | } else { | |
984 | /* index */ | |
985 | if (path[depth].p_idx != | |
986 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
987 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
988 | } | |
989 | depth--; | |
990 | } | |
991 | ||
992 | return EXT_MAX_BLOCK; | |
993 | } | |
994 | ||
995 | /* | |
996 | * returns first allocated block from next leaf or EXT_MAX_BLOCK | |
997 | */ | |
998 | static unsigned ext4_ext_next_leaf_block(struct inode *inode, | |
999 | struct ext4_ext_path *path) | |
1000 | { | |
1001 | int depth; | |
1002 | ||
1003 | BUG_ON(path == NULL); | |
1004 | depth = path->p_depth; | |
1005 | ||
1006 | /* zero-tree has no leaf blocks at all */ | |
1007 | if (depth == 0) | |
1008 | return EXT_MAX_BLOCK; | |
1009 | ||
1010 | /* go to index block */ | |
1011 | depth--; | |
1012 | ||
1013 | while (depth >= 0) { | |
1014 | if (path[depth].p_idx != | |
1015 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1016 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1017 | depth--; | |
1018 | } | |
1019 | ||
1020 | return EXT_MAX_BLOCK; | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * if leaf gets modified and modified extent is first in the leaf | |
1025 | * then we have to correct all indexes above | |
1026 | * TODO: do we need to correct tree in all cases? | |
1027 | */ | |
1028 | int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, | |
1029 | struct ext4_ext_path *path) | |
1030 | { | |
1031 | struct ext4_extent_header *eh; | |
1032 | int depth = ext_depth(inode); | |
1033 | struct ext4_extent *ex; | |
1034 | __le32 border; | |
1035 | int k, err = 0; | |
1036 | ||
1037 | eh = path[depth].p_hdr; | |
1038 | ex = path[depth].p_ext; | |
1039 | BUG_ON(ex == NULL); | |
1040 | BUG_ON(eh == NULL); | |
1041 | ||
1042 | if (depth == 0) { | |
1043 | /* there is no tree at all */ | |
1044 | return 0; | |
1045 | } | |
1046 | ||
1047 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1048 | /* we correct tree if first leaf got modified only */ | |
1049 | return 0; | |
1050 | } | |
1051 | ||
1052 | /* | |
1053 | * TODO: we need correction if border is smaller then current one | |
1054 | */ | |
1055 | k = depth - 1; | |
1056 | border = path[depth].p_ext->ee_block; | |
1057 | if ((err = ext4_ext_get_access(handle, inode, path + k))) | |
1058 | return err; | |
1059 | path[k].p_idx->ei_block = border; | |
1060 | if ((err = ext4_ext_dirty(handle, inode, path + k))) | |
1061 | return err; | |
1062 | ||
1063 | while (k--) { | |
1064 | /* change all left-side indexes */ | |
1065 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1066 | break; | |
1067 | if ((err = ext4_ext_get_access(handle, inode, path + k))) | |
1068 | break; | |
1069 | path[k].p_idx->ei_block = border; | |
1070 | if ((err = ext4_ext_dirty(handle, inode, path + k))) | |
1071 | break; | |
1072 | } | |
1073 | ||
1074 | return err; | |
1075 | } | |
1076 | ||
1077 | static int inline | |
1078 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, | |
1079 | struct ext4_extent *ex2) | |
1080 | { | |
a86c6181 AT |
1081 | if (le32_to_cpu(ex1->ee_block) + le16_to_cpu(ex1->ee_len) |
1082 | != le32_to_cpu(ex2->ee_block)) | |
1083 | return 0; | |
1084 | ||
1085 | #ifdef AGRESSIVE_TEST | |
1086 | if (le16_to_cpu(ex1->ee_len) >= 4) | |
1087 | return 0; | |
1088 | #endif | |
1089 | ||
f65e6fba | 1090 | if (ext_pblock(ex1) + le16_to_cpu(ex1->ee_len) == ext_pblock(ex2)) |
a86c6181 AT |
1091 | return 1; |
1092 | return 0; | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * this routine tries to merge requsted extent into the existing | |
1097 | * extent or inserts requested extent as new one into the tree, | |
1098 | * creating new leaf in no-space case | |
1099 | */ | |
1100 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1101 | struct ext4_ext_path *path, | |
1102 | struct ext4_extent *newext) | |
1103 | { | |
1104 | struct ext4_extent_header * eh; | |
1105 | struct ext4_extent *ex, *fex; | |
1106 | struct ext4_extent *nearex; /* nearest extent */ | |
1107 | struct ext4_ext_path *npath = NULL; | |
1108 | int depth, len, err, next; | |
1109 | ||
1110 | BUG_ON(newext->ee_len == 0); | |
1111 | depth = ext_depth(inode); | |
1112 | ex = path[depth].p_ext; | |
1113 | BUG_ON(path[depth].p_hdr == NULL); | |
1114 | ||
1115 | /* try to insert block into found extent and return */ | |
1116 | if (ex && ext4_can_extents_be_merged(inode, ex, newext)) { | |
f65e6fba | 1117 | ext_debug("append %d block to %d:%d (from "E3FSBLK")\n", |
a86c6181 AT |
1118 | le16_to_cpu(newext->ee_len), |
1119 | le32_to_cpu(ex->ee_block), | |
f65e6fba | 1120 | le16_to_cpu(ex->ee_len), ext_pblock(ex)); |
a86c6181 AT |
1121 | if ((err = ext4_ext_get_access(handle, inode, path + depth))) |
1122 | return err; | |
1123 | ex->ee_len = cpu_to_le16(le16_to_cpu(ex->ee_len) | |
1124 | + le16_to_cpu(newext->ee_len)); | |
1125 | eh = path[depth].p_hdr; | |
1126 | nearex = ex; | |
1127 | goto merge; | |
1128 | } | |
1129 | ||
1130 | repeat: | |
1131 | depth = ext_depth(inode); | |
1132 | eh = path[depth].p_hdr; | |
1133 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1134 | goto has_space; | |
1135 | ||
1136 | /* probably next leaf has space for us? */ | |
1137 | fex = EXT_LAST_EXTENT(eh); | |
1138 | next = ext4_ext_next_leaf_block(inode, path); | |
1139 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block) | |
1140 | && next != EXT_MAX_BLOCK) { | |
1141 | ext_debug("next leaf block - %d\n", next); | |
1142 | BUG_ON(npath != NULL); | |
1143 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1144 | if (IS_ERR(npath)) | |
1145 | return PTR_ERR(npath); | |
1146 | BUG_ON(npath->p_depth != path->p_depth); | |
1147 | eh = npath[depth].p_hdr; | |
1148 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
1149 | ext_debug("next leaf isnt full(%d)\n", | |
1150 | le16_to_cpu(eh->eh_entries)); | |
1151 | path = npath; | |
1152 | goto repeat; | |
1153 | } | |
1154 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1155 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1156 | } | |
1157 | ||
1158 | /* | |
1159 | * there is no free space in found leaf | |
1160 | * we're gonna add new leaf in the tree | |
1161 | */ | |
1162 | err = ext4_ext_create_new_leaf(handle, inode, path, newext); | |
1163 | if (err) | |
1164 | goto cleanup; | |
1165 | depth = ext_depth(inode); | |
1166 | eh = path[depth].p_hdr; | |
1167 | ||
1168 | has_space: | |
1169 | nearex = path[depth].p_ext; | |
1170 | ||
1171 | if ((err = ext4_ext_get_access(handle, inode, path + depth))) | |
1172 | goto cleanup; | |
1173 | ||
1174 | if (!nearex) { | |
1175 | /* there is no extent in this leaf, create first one */ | |
f65e6fba | 1176 | ext_debug("first extent in the leaf: %d:"E3FSBLK":%d\n", |
a86c6181 | 1177 | le32_to_cpu(newext->ee_block), |
f65e6fba | 1178 | ext_pblock(newext), |
a86c6181 AT |
1179 | le16_to_cpu(newext->ee_len)); |
1180 | path[depth].p_ext = EXT_FIRST_EXTENT(eh); | |
1181 | } else if (le32_to_cpu(newext->ee_block) | |
1182 | > le32_to_cpu(nearex->ee_block)) { | |
1183 | /* BUG_ON(newext->ee_block == nearex->ee_block); */ | |
1184 | if (nearex != EXT_LAST_EXTENT(eh)) { | |
1185 | len = EXT_MAX_EXTENT(eh) - nearex; | |
1186 | len = (len - 1) * sizeof(struct ext4_extent); | |
1187 | len = len < 0 ? 0 : len; | |
f65e6fba | 1188 | ext_debug("insert %d:"E3FSBLK":%d after: nearest 0x%p, " |
a86c6181 AT |
1189 | "move %d from 0x%p to 0x%p\n", |
1190 | le32_to_cpu(newext->ee_block), | |
f65e6fba | 1191 | ext_pblock(newext), |
a86c6181 AT |
1192 | le16_to_cpu(newext->ee_len), |
1193 | nearex, len, nearex + 1, nearex + 2); | |
1194 | memmove(nearex + 2, nearex + 1, len); | |
1195 | } | |
1196 | path[depth].p_ext = nearex + 1; | |
1197 | } else { | |
1198 | BUG_ON(newext->ee_block == nearex->ee_block); | |
1199 | len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); | |
1200 | len = len < 0 ? 0 : len; | |
f65e6fba | 1201 | ext_debug("insert %d:"E3FSBLK":%d before: nearest 0x%p, " |
a86c6181 AT |
1202 | "move %d from 0x%p to 0x%p\n", |
1203 | le32_to_cpu(newext->ee_block), | |
f65e6fba | 1204 | ext_pblock(newext), |
a86c6181 AT |
1205 | le16_to_cpu(newext->ee_len), |
1206 | nearex, len, nearex + 1, nearex + 2); | |
1207 | memmove(nearex + 1, nearex, len); | |
1208 | path[depth].p_ext = nearex; | |
1209 | } | |
1210 | ||
1211 | eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1); | |
1212 | nearex = path[depth].p_ext; | |
1213 | nearex->ee_block = newext->ee_block; | |
1214 | nearex->ee_start = newext->ee_start; | |
f65e6fba | 1215 | nearex->ee_start_hi = newext->ee_start_hi; |
a86c6181 | 1216 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1217 | |
1218 | merge: | |
1219 | /* try to merge extents to the right */ | |
1220 | while (nearex < EXT_LAST_EXTENT(eh)) { | |
1221 | if (!ext4_can_extents_be_merged(inode, nearex, nearex + 1)) | |
1222 | break; | |
1223 | /* merge with next extent! */ | |
1224 | nearex->ee_len = cpu_to_le16(le16_to_cpu(nearex->ee_len) | |
1225 | + le16_to_cpu(nearex[1].ee_len)); | |
1226 | if (nearex + 1 < EXT_LAST_EXTENT(eh)) { | |
1227 | len = (EXT_LAST_EXTENT(eh) - nearex - 1) | |
1228 | * sizeof(struct ext4_extent); | |
1229 | memmove(nearex + 1, nearex + 2, len); | |
1230 | } | |
1231 | eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1); | |
1232 | BUG_ON(eh->eh_entries == 0); | |
1233 | } | |
1234 | ||
1235 | /* try to merge extents to the left */ | |
1236 | ||
1237 | /* time to correct all indexes above */ | |
1238 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1239 | if (err) | |
1240 | goto cleanup; | |
1241 | ||
1242 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1243 | ||
1244 | cleanup: | |
1245 | if (npath) { | |
1246 | ext4_ext_drop_refs(npath); | |
1247 | kfree(npath); | |
1248 | } | |
1249 | ext4_ext_tree_changed(inode); | |
1250 | ext4_ext_invalidate_cache(inode); | |
1251 | return err; | |
1252 | } | |
1253 | ||
1254 | int ext4_ext_walk_space(struct inode *inode, unsigned long block, | |
1255 | unsigned long num, ext_prepare_callback func, | |
1256 | void *cbdata) | |
1257 | { | |
1258 | struct ext4_ext_path *path = NULL; | |
1259 | struct ext4_ext_cache cbex; | |
1260 | struct ext4_extent *ex; | |
1261 | unsigned long next, start = 0, end = 0; | |
1262 | unsigned long last = block + num; | |
1263 | int depth, exists, err = 0; | |
1264 | ||
1265 | BUG_ON(func == NULL); | |
1266 | BUG_ON(inode == NULL); | |
1267 | ||
1268 | while (block < last && block != EXT_MAX_BLOCK) { | |
1269 | num = last - block; | |
1270 | /* find extent for this block */ | |
1271 | path = ext4_ext_find_extent(inode, block, path); | |
1272 | if (IS_ERR(path)) { | |
1273 | err = PTR_ERR(path); | |
1274 | path = NULL; | |
1275 | break; | |
1276 | } | |
1277 | ||
1278 | depth = ext_depth(inode); | |
1279 | BUG_ON(path[depth].p_hdr == NULL); | |
1280 | ex = path[depth].p_ext; | |
1281 | next = ext4_ext_next_allocated_block(path); | |
1282 | ||
1283 | exists = 0; | |
1284 | if (!ex) { | |
1285 | /* there is no extent yet, so try to allocate | |
1286 | * all requested space */ | |
1287 | start = block; | |
1288 | end = block + num; | |
1289 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
1290 | /* need to allocate space before found extent */ | |
1291 | start = block; | |
1292 | end = le32_to_cpu(ex->ee_block); | |
1293 | if (block + num < end) | |
1294 | end = block + num; | |
1295 | } else if (block >= | |
1296 | le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len)) { | |
1297 | /* need to allocate space after found extent */ | |
1298 | start = block; | |
1299 | end = block + num; | |
1300 | if (end >= next) | |
1301 | end = next; | |
1302 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
1303 | /* | |
1304 | * some part of requested space is covered | |
1305 | * by found extent | |
1306 | */ | |
1307 | start = block; | |
1308 | end = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len); | |
1309 | if (block + num < end) | |
1310 | end = block + num; | |
1311 | exists = 1; | |
1312 | } else { | |
1313 | BUG(); | |
1314 | } | |
1315 | BUG_ON(end <= start); | |
1316 | ||
1317 | if (!exists) { | |
1318 | cbex.ec_block = start; | |
1319 | cbex.ec_len = end - start; | |
1320 | cbex.ec_start = 0; | |
1321 | cbex.ec_type = EXT4_EXT_CACHE_GAP; | |
1322 | } else { | |
1323 | cbex.ec_block = le32_to_cpu(ex->ee_block); | |
1324 | cbex.ec_len = le16_to_cpu(ex->ee_len); | |
f65e6fba | 1325 | cbex.ec_start = ext_pblock(ex); |
a86c6181 AT |
1326 | cbex.ec_type = EXT4_EXT_CACHE_EXTENT; |
1327 | } | |
1328 | ||
1329 | BUG_ON(cbex.ec_len == 0); | |
1330 | err = func(inode, path, &cbex, cbdata); | |
1331 | ext4_ext_drop_refs(path); | |
1332 | ||
1333 | if (err < 0) | |
1334 | break; | |
1335 | if (err == EXT_REPEAT) | |
1336 | continue; | |
1337 | else if (err == EXT_BREAK) { | |
1338 | err = 0; | |
1339 | break; | |
1340 | } | |
1341 | ||
1342 | if (ext_depth(inode) != depth) { | |
1343 | /* depth was changed. we have to realloc path */ | |
1344 | kfree(path); | |
1345 | path = NULL; | |
1346 | } | |
1347 | ||
1348 | block = cbex.ec_block + cbex.ec_len; | |
1349 | } | |
1350 | ||
1351 | if (path) { | |
1352 | ext4_ext_drop_refs(path); | |
1353 | kfree(path); | |
1354 | } | |
1355 | ||
1356 | return err; | |
1357 | } | |
1358 | ||
1359 | static inline void | |
1360 | ext4_ext_put_in_cache(struct inode *inode, __u32 block, | |
1361 | __u32 len, __u32 start, int type) | |
1362 | { | |
1363 | struct ext4_ext_cache *cex; | |
1364 | BUG_ON(len == 0); | |
1365 | cex = &EXT4_I(inode)->i_cached_extent; | |
1366 | cex->ec_type = type; | |
1367 | cex->ec_block = block; | |
1368 | cex->ec_len = len; | |
1369 | cex->ec_start = start; | |
1370 | } | |
1371 | ||
1372 | /* | |
1373 | * this routine calculate boundaries of the gap requested block fits into | |
1374 | * and cache this gap | |
1375 | */ | |
1376 | static inline void | |
1377 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, | |
1378 | unsigned long block) | |
1379 | { | |
1380 | int depth = ext_depth(inode); | |
1381 | unsigned long lblock, len; | |
1382 | struct ext4_extent *ex; | |
1383 | ||
1384 | ex = path[depth].p_ext; | |
1385 | if (ex == NULL) { | |
1386 | /* there is no extent yet, so gap is [0;-] */ | |
1387 | lblock = 0; | |
1388 | len = EXT_MAX_BLOCK; | |
1389 | ext_debug("cache gap(whole file):"); | |
1390 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
1391 | lblock = block; | |
1392 | len = le32_to_cpu(ex->ee_block) - block; | |
1393 | ext_debug("cache gap(before): %lu [%lu:%lu]", | |
1394 | (unsigned long) block, | |
1395 | (unsigned long) le32_to_cpu(ex->ee_block), | |
1396 | (unsigned long) le16_to_cpu(ex->ee_len)); | |
1397 | } else if (block >= le32_to_cpu(ex->ee_block) | |
1398 | + le16_to_cpu(ex->ee_len)) { | |
1399 | lblock = le32_to_cpu(ex->ee_block) | |
1400 | + le16_to_cpu(ex->ee_len); | |
1401 | len = ext4_ext_next_allocated_block(path); | |
1402 | ext_debug("cache gap(after): [%lu:%lu] %lu", | |
1403 | (unsigned long) le32_to_cpu(ex->ee_block), | |
1404 | (unsigned long) le16_to_cpu(ex->ee_len), | |
1405 | (unsigned long) block); | |
1406 | BUG_ON(len == lblock); | |
1407 | len = len - lblock; | |
1408 | } else { | |
1409 | lblock = len = 0; | |
1410 | BUG(); | |
1411 | } | |
1412 | ||
1413 | ext_debug(" -> %lu:%lu\n", (unsigned long) lblock, len); | |
1414 | ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP); | |
1415 | } | |
1416 | ||
1417 | static inline int | |
1418 | ext4_ext_in_cache(struct inode *inode, unsigned long block, | |
1419 | struct ext4_extent *ex) | |
1420 | { | |
1421 | struct ext4_ext_cache *cex; | |
1422 | ||
1423 | cex = &EXT4_I(inode)->i_cached_extent; | |
1424 | ||
1425 | /* has cache valid data? */ | |
1426 | if (cex->ec_type == EXT4_EXT_CACHE_NO) | |
1427 | return EXT4_EXT_CACHE_NO; | |
1428 | ||
1429 | BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP && | |
1430 | cex->ec_type != EXT4_EXT_CACHE_EXTENT); | |
1431 | if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) { | |
1432 | ex->ee_block = cpu_to_le32(cex->ec_block); | |
f65e6fba | 1433 | ext4_ext_store_pblock(ex, cex->ec_start); |
a86c6181 | 1434 | ex->ee_len = cpu_to_le16(cex->ec_len); |
f65e6fba | 1435 | ext_debug("%lu cached by %lu:%lu:"E3FSBLK"\n", |
a86c6181 AT |
1436 | (unsigned long) block, |
1437 | (unsigned long) cex->ec_block, | |
1438 | (unsigned long) cex->ec_len, | |
f65e6fba | 1439 | cex->ec_start); |
a86c6181 AT |
1440 | return cex->ec_type; |
1441 | } | |
1442 | ||
1443 | /* not in cache */ | |
1444 | return EXT4_EXT_CACHE_NO; | |
1445 | } | |
1446 | ||
1447 | /* | |
1448 | * routine removes index from the index block | |
1449 | * it's used in truncate case only. thus all requests are for | |
1450 | * last index in the block only | |
1451 | */ | |
1452 | int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, | |
1453 | struct ext4_ext_path *path) | |
1454 | { | |
1455 | struct buffer_head *bh; | |
1456 | int err; | |
f65e6fba | 1457 | ext4_fsblk_t leaf; |
a86c6181 AT |
1458 | |
1459 | /* free index block */ | |
1460 | path--; | |
f65e6fba | 1461 | leaf = idx_pblock(path->p_idx); |
a86c6181 AT |
1462 | BUG_ON(path->p_hdr->eh_entries == 0); |
1463 | if ((err = ext4_ext_get_access(handle, inode, path))) | |
1464 | return err; | |
1465 | path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1); | |
1466 | if ((err = ext4_ext_dirty(handle, inode, path))) | |
1467 | return err; | |
f65e6fba | 1468 | ext_debug("index is empty, remove it, free block "E3FSBLK"\n", leaf); |
a86c6181 AT |
1469 | bh = sb_find_get_block(inode->i_sb, leaf); |
1470 | ext4_forget(handle, 1, inode, bh, leaf); | |
1471 | ext4_free_blocks(handle, inode, leaf, 1); | |
1472 | return err; | |
1473 | } | |
1474 | ||
1475 | /* | |
1476 | * This routine returns max. credits extent tree can consume. | |
1477 | * It should be OK for low-performance paths like ->writepage() | |
1478 | * To allow many writing process to fit a single transaction, | |
1479 | * caller should calculate credits under truncate_mutex and | |
1480 | * pass actual path. | |
1481 | */ | |
1482 | int inline ext4_ext_calc_credits_for_insert(struct inode *inode, | |
1483 | struct ext4_ext_path *path) | |
1484 | { | |
1485 | int depth, needed; | |
1486 | ||
1487 | if (path) { | |
1488 | /* probably there is space in leaf? */ | |
1489 | depth = ext_depth(inode); | |
1490 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) | |
1491 | < le16_to_cpu(path[depth].p_hdr->eh_max)) | |
1492 | return 1; | |
1493 | } | |
1494 | ||
1495 | /* | |
1496 | * given 32bit logical block (4294967296 blocks), max. tree | |
1497 | * can be 4 levels in depth -- 4 * 340^4 == 53453440000. | |
1498 | * let's also add one more level for imbalance. | |
1499 | */ | |
1500 | depth = 5; | |
1501 | ||
1502 | /* allocation of new data block(s) */ | |
1503 | needed = 2; | |
1504 | ||
1505 | /* | |
1506 | * tree can be full, so it'd need to grow in depth: | |
1507 | * allocation + old root + new root | |
1508 | */ | |
1509 | needed += 2 + 1 + 1; | |
1510 | ||
1511 | /* | |
1512 | * Index split can happen, we'd need: | |
1513 | * allocate intermediate indexes (bitmap + group) | |
1514 | * + change two blocks at each level, but root (already included) | |
1515 | */ | |
1516 | needed = (depth * 2) + (depth * 2); | |
1517 | ||
1518 | /* any allocation modifies superblock */ | |
1519 | needed += 1; | |
1520 | ||
1521 | return needed; | |
1522 | } | |
1523 | ||
1524 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
1525 | struct ext4_extent *ex, | |
1526 | unsigned long from, unsigned long to) | |
1527 | { | |
1528 | struct buffer_head *bh; | |
1529 | int i; | |
1530 | ||
1531 | #ifdef EXTENTS_STATS | |
1532 | { | |
1533 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1534 | unsigned short ee_len = le16_to_cpu(ex->ee_len); | |
1535 | spin_lock(&sbi->s_ext_stats_lock); | |
1536 | sbi->s_ext_blocks += ee_len; | |
1537 | sbi->s_ext_extents++; | |
1538 | if (ee_len < sbi->s_ext_min) | |
1539 | sbi->s_ext_min = ee_len; | |
1540 | if (ee_len > sbi->s_ext_max) | |
1541 | sbi->s_ext_max = ee_len; | |
1542 | if (ext_depth(inode) > sbi->s_depth_max) | |
1543 | sbi->s_depth_max = ext_depth(inode); | |
1544 | spin_unlock(&sbi->s_ext_stats_lock); | |
1545 | } | |
1546 | #endif | |
1547 | if (from >= le32_to_cpu(ex->ee_block) | |
1548 | && to == le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) { | |
1549 | /* tail removal */ | |
f65e6fba AT |
1550 | unsigned long num; |
1551 | ext4_fsblk_t start; | |
a86c6181 | 1552 | num = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - from; |
f65e6fba AT |
1553 | start = ext_pblock(ex) + le16_to_cpu(ex->ee_len) - num; |
1554 | ext_debug("free last %lu blocks starting "E3FSBLK"\n", num, start); | |
a86c6181 AT |
1555 | for (i = 0; i < num; i++) { |
1556 | bh = sb_find_get_block(inode->i_sb, start + i); | |
1557 | ext4_forget(handle, 0, inode, bh, start + i); | |
1558 | } | |
1559 | ext4_free_blocks(handle, inode, start, num); | |
1560 | } else if (from == le32_to_cpu(ex->ee_block) | |
1561 | && to <= le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) { | |
1562 | printk("strange request: removal %lu-%lu from %u:%u\n", | |
1563 | from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len)); | |
1564 | } else { | |
1565 | printk("strange request: removal(2) %lu-%lu from %u:%u\n", | |
1566 | from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len)); | |
1567 | } | |
1568 | return 0; | |
1569 | } | |
1570 | ||
1571 | static int | |
1572 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
1573 | struct ext4_ext_path *path, unsigned long start) | |
1574 | { | |
1575 | int err = 0, correct_index = 0; | |
1576 | int depth = ext_depth(inode), credits; | |
1577 | struct ext4_extent_header *eh; | |
1578 | unsigned a, b, block, num; | |
1579 | unsigned long ex_ee_block; | |
1580 | unsigned short ex_ee_len; | |
1581 | struct ext4_extent *ex; | |
1582 | ||
1583 | ext_debug("truncate since %lu in leaf\n", start); | |
1584 | if (!path[depth].p_hdr) | |
1585 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
1586 | eh = path[depth].p_hdr; | |
1587 | BUG_ON(eh == NULL); | |
1588 | BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max)); | |
1589 | BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC); | |
1590 | ||
1591 | /* find where to start removing */ | |
1592 | ex = EXT_LAST_EXTENT(eh); | |
1593 | ||
1594 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
1595 | ex_ee_len = le16_to_cpu(ex->ee_len); | |
1596 | ||
1597 | while (ex >= EXT_FIRST_EXTENT(eh) && | |
1598 | ex_ee_block + ex_ee_len > start) { | |
1599 | ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len); | |
1600 | path[depth].p_ext = ex; | |
1601 | ||
1602 | a = ex_ee_block > start ? ex_ee_block : start; | |
1603 | b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ? | |
1604 | ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK; | |
1605 | ||
1606 | ext_debug(" border %u:%u\n", a, b); | |
1607 | ||
1608 | if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) { | |
1609 | block = 0; | |
1610 | num = 0; | |
1611 | BUG(); | |
1612 | } else if (a != ex_ee_block) { | |
1613 | /* remove tail of the extent */ | |
1614 | block = ex_ee_block; | |
1615 | num = a - block; | |
1616 | } else if (b != ex_ee_block + ex_ee_len - 1) { | |
1617 | /* remove head of the extent */ | |
1618 | block = a; | |
1619 | num = b - a; | |
1620 | /* there is no "make a hole" API yet */ | |
1621 | BUG(); | |
1622 | } else { | |
1623 | /* remove whole extent: excellent! */ | |
1624 | block = ex_ee_block; | |
1625 | num = 0; | |
1626 | BUG_ON(a != ex_ee_block); | |
1627 | BUG_ON(b != ex_ee_block + ex_ee_len - 1); | |
1628 | } | |
1629 | ||
1630 | /* at present, extent can't cross block group */ | |
1631 | /* leaf + bitmap + group desc + sb + inode */ | |
1632 | credits = 5; | |
1633 | if (ex == EXT_FIRST_EXTENT(eh)) { | |
1634 | correct_index = 1; | |
1635 | credits += (ext_depth(inode)) + 1; | |
1636 | } | |
1637 | #ifdef CONFIG_QUOTA | |
1638 | credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); | |
1639 | #endif | |
1640 | ||
1641 | handle = ext4_ext_journal_restart(handle, credits); | |
1642 | if (IS_ERR(handle)) { | |
1643 | err = PTR_ERR(handle); | |
1644 | goto out; | |
1645 | } | |
1646 | ||
1647 | err = ext4_ext_get_access(handle, inode, path + depth); | |
1648 | if (err) | |
1649 | goto out; | |
1650 | ||
1651 | err = ext4_remove_blocks(handle, inode, ex, a, b); | |
1652 | if (err) | |
1653 | goto out; | |
1654 | ||
1655 | if (num == 0) { | |
1656 | /* this extent is removed entirely mark slot unused */ | |
f65e6fba | 1657 | ext4_ext_store_pblock(ex, 0); |
a86c6181 AT |
1658 | eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1); |
1659 | } | |
1660 | ||
1661 | ex->ee_block = cpu_to_le32(block); | |
1662 | ex->ee_len = cpu_to_le16(num); | |
1663 | ||
1664 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1665 | if (err) | |
1666 | goto out; | |
1667 | ||
f65e6fba AT |
1668 | ext_debug("new extent: %u:%u:"E3FSBLK"\n", block, num, |
1669 | ext_pblock(ex)); | |
a86c6181 AT |
1670 | ex--; |
1671 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
1672 | ex_ee_len = le16_to_cpu(ex->ee_len); | |
1673 | } | |
1674 | ||
1675 | if (correct_index && eh->eh_entries) | |
1676 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1677 | ||
1678 | /* if this leaf is free, then we should | |
1679 | * remove it from index block above */ | |
1680 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
1681 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
1682 | ||
1683 | out: | |
1684 | return err; | |
1685 | } | |
1686 | ||
1687 | /* | |
1688 | * returns 1 if current index have to be freed (even partial) | |
1689 | */ | |
1690 | static int inline | |
1691 | ext4_ext_more_to_rm(struct ext4_ext_path *path) | |
1692 | { | |
1693 | BUG_ON(path->p_idx == NULL); | |
1694 | ||
1695 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
1696 | return 0; | |
1697 | ||
1698 | /* | |
1699 | * if truncate on deeper level happened it it wasn't partial | |
1700 | * so we have to consider current index for truncation | |
1701 | */ | |
1702 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
1703 | return 0; | |
1704 | return 1; | |
1705 | } | |
1706 | ||
1707 | int ext4_ext_remove_space(struct inode *inode, unsigned long start) | |
1708 | { | |
1709 | struct super_block *sb = inode->i_sb; | |
1710 | int depth = ext_depth(inode); | |
1711 | struct ext4_ext_path *path; | |
1712 | handle_t *handle; | |
1713 | int i = 0, err = 0; | |
1714 | ||
1715 | ext_debug("truncate since %lu\n", start); | |
1716 | ||
1717 | /* probably first extent we're gonna free will be last in block */ | |
1718 | handle = ext4_journal_start(inode, depth + 1); | |
1719 | if (IS_ERR(handle)) | |
1720 | return PTR_ERR(handle); | |
1721 | ||
1722 | ext4_ext_invalidate_cache(inode); | |
1723 | ||
1724 | /* | |
1725 | * we start scanning from right side freeing all the blocks | |
1726 | * after i_size and walking into the deep | |
1727 | */ | |
1728 | path = kmalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL); | |
1729 | if (path == NULL) { | |
1730 | ext4_journal_stop(handle); | |
1731 | return -ENOMEM; | |
1732 | } | |
1733 | memset(path, 0, sizeof(struct ext4_ext_path) * (depth + 1)); | |
1734 | path[0].p_hdr = ext_inode_hdr(inode); | |
1735 | if (ext4_ext_check_header(__FUNCTION__, inode, path[0].p_hdr)) { | |
1736 | err = -EIO; | |
1737 | goto out; | |
1738 | } | |
1739 | path[0].p_depth = depth; | |
1740 | ||
1741 | while (i >= 0 && err == 0) { | |
1742 | if (i == depth) { | |
1743 | /* this is leaf block */ | |
1744 | err = ext4_ext_rm_leaf(handle, inode, path, start); | |
1745 | /* root level have p_bh == NULL, brelse() eats this */ | |
1746 | brelse(path[i].p_bh); | |
1747 | path[i].p_bh = NULL; | |
1748 | i--; | |
1749 | continue; | |
1750 | } | |
1751 | ||
1752 | /* this is index block */ | |
1753 | if (!path[i].p_hdr) { | |
1754 | ext_debug("initialize header\n"); | |
1755 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
1756 | if (ext4_ext_check_header(__FUNCTION__, inode, | |
1757 | path[i].p_hdr)) { | |
1758 | err = -EIO; | |
1759 | goto out; | |
1760 | } | |
1761 | } | |
1762 | ||
1763 | BUG_ON(le16_to_cpu(path[i].p_hdr->eh_entries) | |
1764 | > le16_to_cpu(path[i].p_hdr->eh_max)); | |
1765 | BUG_ON(path[i].p_hdr->eh_magic != EXT4_EXT_MAGIC); | |
1766 | ||
1767 | if (!path[i].p_idx) { | |
1768 | /* this level hasn't touched yet */ | |
1769 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); | |
1770 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
1771 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
1772 | path[i].p_hdr, | |
1773 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
1774 | } else { | |
1775 | /* we've already was here, see at next index */ | |
1776 | path[i].p_idx--; | |
1777 | } | |
1778 | ||
1779 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
1780 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
1781 | path[i].p_idx); | |
1782 | if (ext4_ext_more_to_rm(path + i)) { | |
1783 | /* go to the next level */ | |
f65e6fba AT |
1784 | ext_debug("move to level %d (block "E3FSBLK")\n", |
1785 | i + 1, idx_pblock(path[i].p_idx)); | |
a86c6181 AT |
1786 | memset(path + i + 1, 0, sizeof(*path)); |
1787 | path[i+1].p_bh = | |
f65e6fba | 1788 | sb_bread(sb, idx_pblock(path[i].p_idx)); |
a86c6181 AT |
1789 | if (!path[i+1].p_bh) { |
1790 | /* should we reset i_size? */ | |
1791 | err = -EIO; | |
1792 | break; | |
1793 | } | |
1794 | ||
1795 | /* put actual number of indexes to know is this | |
1796 | * number got changed at the next iteration */ | |
1797 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); | |
1798 | i++; | |
1799 | } else { | |
1800 | /* we finish processing this index, go up */ | |
1801 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { | |
1802 | /* index is empty, remove it | |
1803 | * handle must be already prepared by the | |
1804 | * truncatei_leaf() */ | |
1805 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
1806 | } | |
1807 | /* root level have p_bh == NULL, brelse() eats this */ | |
1808 | brelse(path[i].p_bh); | |
1809 | path[i].p_bh = NULL; | |
1810 | i--; | |
1811 | ext_debug("return to level %d\n", i); | |
1812 | } | |
1813 | } | |
1814 | ||
1815 | /* TODO: flexible tree reduction should be here */ | |
1816 | if (path->p_hdr->eh_entries == 0) { | |
1817 | /* | |
1818 | * truncate to zero freed all the tree | |
1819 | * so, we need to correct eh_depth | |
1820 | */ | |
1821 | err = ext4_ext_get_access(handle, inode, path); | |
1822 | if (err == 0) { | |
1823 | ext_inode_hdr(inode)->eh_depth = 0; | |
1824 | ext_inode_hdr(inode)->eh_max = | |
1825 | cpu_to_le16(ext4_ext_space_root(inode)); | |
1826 | err = ext4_ext_dirty(handle, inode, path); | |
1827 | } | |
1828 | } | |
1829 | out: | |
1830 | ext4_ext_tree_changed(inode); | |
1831 | ext4_ext_drop_refs(path); | |
1832 | kfree(path); | |
1833 | ext4_journal_stop(handle); | |
1834 | ||
1835 | return err; | |
1836 | } | |
1837 | ||
1838 | /* | |
1839 | * called at mount time | |
1840 | */ | |
1841 | void ext4_ext_init(struct super_block *sb) | |
1842 | { | |
1843 | /* | |
1844 | * possible initialization would be here | |
1845 | */ | |
1846 | ||
1847 | if (test_opt(sb, EXTENTS)) { | |
1848 | printk("EXT4-fs: file extents enabled"); | |
1849 | #ifdef AGRESSIVE_TEST | |
1850 | printk(", agressive tests"); | |
1851 | #endif | |
1852 | #ifdef CHECK_BINSEARCH | |
1853 | printk(", check binsearch"); | |
1854 | #endif | |
1855 | #ifdef EXTENTS_STATS | |
1856 | printk(", stats"); | |
1857 | #endif | |
1858 | printk("\n"); | |
1859 | #ifdef EXTENTS_STATS | |
1860 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
1861 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
1862 | EXT4_SB(sb)->s_ext_max = 0; | |
1863 | #endif | |
1864 | } | |
1865 | } | |
1866 | ||
1867 | /* | |
1868 | * called at umount time | |
1869 | */ | |
1870 | void ext4_ext_release(struct super_block *sb) | |
1871 | { | |
1872 | if (!test_opt(sb, EXTENTS)) | |
1873 | return; | |
1874 | ||
1875 | #ifdef EXTENTS_STATS | |
1876 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
1877 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1878 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
1879 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
1880 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
1881 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
1882 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
1883 | } | |
1884 | #endif | |
1885 | } | |
1886 | ||
f65e6fba AT |
1887 | int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, |
1888 | ext4_fsblk_t iblock, | |
a86c6181 AT |
1889 | unsigned long max_blocks, struct buffer_head *bh_result, |
1890 | int create, int extend_disksize) | |
1891 | { | |
1892 | struct ext4_ext_path *path = NULL; | |
1893 | struct ext4_extent newex, *ex; | |
f65e6fba AT |
1894 | ext4_fsblk_t goal, newblock; |
1895 | int err = 0, depth; | |
a86c6181 AT |
1896 | unsigned long allocated = 0; |
1897 | ||
1898 | __clear_bit(BH_New, &bh_result->b_state); | |
1899 | ext_debug("blocks %d/%lu requested for inode %u\n", (int) iblock, | |
1900 | max_blocks, (unsigned) inode->i_ino); | |
1901 | mutex_lock(&EXT4_I(inode)->truncate_mutex); | |
1902 | ||
1903 | /* check in cache */ | |
1904 | if ((goal = ext4_ext_in_cache(inode, iblock, &newex))) { | |
1905 | if (goal == EXT4_EXT_CACHE_GAP) { | |
1906 | if (!create) { | |
1907 | /* block isn't allocated yet and | |
1908 | * user don't want to allocate it */ | |
1909 | goto out2; | |
1910 | } | |
1911 | /* we should allocate requested block */ | |
1912 | } else if (goal == EXT4_EXT_CACHE_EXTENT) { | |
1913 | /* block is already allocated */ | |
1914 | newblock = iblock | |
1915 | - le32_to_cpu(newex.ee_block) | |
f65e6fba | 1916 | + ext_pblock(&newex); |
a86c6181 AT |
1917 | /* number of remain blocks in the extent */ |
1918 | allocated = le16_to_cpu(newex.ee_len) - | |
1919 | (iblock - le32_to_cpu(newex.ee_block)); | |
1920 | goto out; | |
1921 | } else { | |
1922 | BUG(); | |
1923 | } | |
1924 | } | |
1925 | ||
1926 | /* find extent for this block */ | |
1927 | path = ext4_ext_find_extent(inode, iblock, NULL); | |
1928 | if (IS_ERR(path)) { | |
1929 | err = PTR_ERR(path); | |
1930 | path = NULL; | |
1931 | goto out2; | |
1932 | } | |
1933 | ||
1934 | depth = ext_depth(inode); | |
1935 | ||
1936 | /* | |
1937 | * consistent leaf must not be empty | |
1938 | * this situations is possible, though, _during_ tree modification | |
1939 | * this is why assert can't be put in ext4_ext_find_extent() | |
1940 | */ | |
1941 | BUG_ON(path[depth].p_ext == NULL && depth != 0); | |
1942 | ||
1943 | if ((ex = path[depth].p_ext)) { | |
1944 | unsigned long ee_block = le32_to_cpu(ex->ee_block); | |
f65e6fba | 1945 | ext4_fsblk_t ee_start = ext_pblock(ex); |
a86c6181 AT |
1946 | unsigned short ee_len = le16_to_cpu(ex->ee_len); |
1947 | /* if found exent covers block, simple return it */ | |
1948 | if (iblock >= ee_block && iblock < ee_block + ee_len) { | |
1949 | newblock = iblock - ee_block + ee_start; | |
1950 | /* number of remain blocks in the extent */ | |
1951 | allocated = ee_len - (iblock - ee_block); | |
f65e6fba | 1952 | ext_debug("%d fit into %lu:%d -> "E3FSBLK"\n", (int) iblock, |
a86c6181 AT |
1953 | ee_block, ee_len, newblock); |
1954 | ext4_ext_put_in_cache(inode, ee_block, ee_len, | |
1955 | ee_start, EXT4_EXT_CACHE_EXTENT); | |
1956 | goto out; | |
1957 | } | |
1958 | } | |
1959 | ||
1960 | /* | |
1961 | * requested block isn't allocated yet | |
1962 | * we couldn't try to create block if create flag is zero | |
1963 | */ | |
1964 | if (!create) { | |
1965 | /* put just found gap into cache to speedup subsequest reqs */ | |
1966 | ext4_ext_put_gap_in_cache(inode, path, iblock); | |
1967 | goto out2; | |
1968 | } | |
1969 | /* | |
1970 | * Okay, we need to do block allocation. Lazily initialize the block | |
1971 | * allocation info here if necessary | |
1972 | */ | |
1973 | if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info)) | |
1974 | ext4_init_block_alloc_info(inode); | |
1975 | ||
1976 | /* allocate new block */ | |
1977 | goal = ext4_ext_find_goal(inode, path, iblock); | |
1978 | allocated = max_blocks; | |
1979 | newblock = ext4_new_blocks(handle, inode, goal, &allocated, &err); | |
1980 | if (!newblock) | |
1981 | goto out2; | |
f65e6fba | 1982 | ext_debug("allocate new block: goal "E3FSBLK", found "E3FSBLK"/%lu\n", |
a86c6181 AT |
1983 | goal, newblock, allocated); |
1984 | ||
1985 | /* try to insert new extent into found leaf and return */ | |
1986 | newex.ee_block = cpu_to_le32(iblock); | |
f65e6fba | 1987 | ext4_ext_store_pblock(&newex, newblock); |
a86c6181 AT |
1988 | newex.ee_len = cpu_to_le16(allocated); |
1989 | err = ext4_ext_insert_extent(handle, inode, path, &newex); | |
1990 | if (err) | |
1991 | goto out2; | |
1992 | ||
1993 | if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize) | |
1994 | EXT4_I(inode)->i_disksize = inode->i_size; | |
1995 | ||
1996 | /* previous routine could use block we allocated */ | |
f65e6fba | 1997 | newblock = ext_pblock(&newex); |
a86c6181 AT |
1998 | __set_bit(BH_New, &bh_result->b_state); |
1999 | ||
2000 | ext4_ext_put_in_cache(inode, iblock, allocated, newblock, | |
2001 | EXT4_EXT_CACHE_EXTENT); | |
2002 | out: | |
2003 | if (allocated > max_blocks) | |
2004 | allocated = max_blocks; | |
2005 | ext4_ext_show_leaf(inode, path); | |
2006 | __set_bit(BH_Mapped, &bh_result->b_state); | |
2007 | bh_result->b_bdev = inode->i_sb->s_bdev; | |
2008 | bh_result->b_blocknr = newblock; | |
2009 | out2: | |
2010 | if (path) { | |
2011 | ext4_ext_drop_refs(path); | |
2012 | kfree(path); | |
2013 | } | |
2014 | mutex_unlock(&EXT4_I(inode)->truncate_mutex); | |
2015 | ||
2016 | return err ? err : allocated; | |
2017 | } | |
2018 | ||
2019 | void ext4_ext_truncate(struct inode * inode, struct page *page) | |
2020 | { | |
2021 | struct address_space *mapping = inode->i_mapping; | |
2022 | struct super_block *sb = inode->i_sb; | |
2023 | unsigned long last_block; | |
2024 | handle_t *handle; | |
2025 | int err = 0; | |
2026 | ||
2027 | /* | |
2028 | * probably first extent we're gonna free will be last in block | |
2029 | */ | |
2030 | err = ext4_writepage_trans_blocks(inode) + 3; | |
2031 | handle = ext4_journal_start(inode, err); | |
2032 | if (IS_ERR(handle)) { | |
2033 | if (page) { | |
2034 | clear_highpage(page); | |
2035 | flush_dcache_page(page); | |
2036 | unlock_page(page); | |
2037 | page_cache_release(page); | |
2038 | } | |
2039 | return; | |
2040 | } | |
2041 | ||
2042 | if (page) | |
2043 | ext4_block_truncate_page(handle, page, mapping, inode->i_size); | |
2044 | ||
2045 | mutex_lock(&EXT4_I(inode)->truncate_mutex); | |
2046 | ext4_ext_invalidate_cache(inode); | |
2047 | ||
2048 | /* | |
2049 | * TODO: optimization is possible here | |
2050 | * probably we need not scaning at all, | |
2051 | * because page truncation is enough | |
2052 | */ | |
2053 | if (ext4_orphan_add(handle, inode)) | |
2054 | goto out_stop; | |
2055 | ||
2056 | /* we have to know where to truncate from in crash case */ | |
2057 | EXT4_I(inode)->i_disksize = inode->i_size; | |
2058 | ext4_mark_inode_dirty(handle, inode); | |
2059 | ||
2060 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
2061 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
2062 | err = ext4_ext_remove_space(inode, last_block); | |
2063 | ||
2064 | /* In a multi-transaction truncate, we only make the final | |
2065 | * transaction synchronous */ | |
2066 | if (IS_SYNC(inode)) | |
2067 | handle->h_sync = 1; | |
2068 | ||
2069 | out_stop: | |
2070 | /* | |
2071 | * If this was a simple ftruncate(), and the file will remain alive | |
2072 | * then we need to clear up the orphan record which we created above. | |
2073 | * However, if this was a real unlink then we were called by | |
2074 | * ext4_delete_inode(), and we allow that function to clean up the | |
2075 | * orphan info for us. | |
2076 | */ | |
2077 | if (inode->i_nlink) | |
2078 | ext4_orphan_del(handle, inode); | |
2079 | ||
2080 | mutex_unlock(&EXT4_I(inode)->truncate_mutex); | |
2081 | ext4_journal_stop(handle); | |
2082 | } | |
2083 | ||
2084 | /* | |
2085 | * this routine calculate max number of blocks we could modify | |
2086 | * in order to allocate new block for an inode | |
2087 | */ | |
2088 | int ext4_ext_writepage_trans_blocks(struct inode *inode, int num) | |
2089 | { | |
2090 | int needed; | |
2091 | ||
2092 | needed = ext4_ext_calc_credits_for_insert(inode, NULL); | |
2093 | ||
2094 | /* caller want to allocate num blocks, but note it includes sb */ | |
2095 | needed = needed * num - (num - 1); | |
2096 | ||
2097 | #ifdef CONFIG_QUOTA | |
2098 | needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); | |
2099 | #endif | |
2100 | ||
2101 | return needed; | |
2102 | } | |
2103 | ||
2104 | EXPORT_SYMBOL(ext4_mark_inode_dirty); | |
2105 | EXPORT_SYMBOL(ext4_ext_invalidate_cache); | |
2106 | EXPORT_SYMBOL(ext4_ext_insert_extent); | |
2107 | EXPORT_SYMBOL(ext4_ext_walk_space); | |
2108 | EXPORT_SYMBOL(ext4_ext_find_goal); | |
2109 | EXPORT_SYMBOL(ext4_ext_calc_credits_for_insert); | |
2110 |